cb06820 Merge pull request #115 from bnema/integrate-upstream-prs · zut

M README.md +19 -4


@@ -145,12 +145,24 @@ $ZUT_HOME/
145	145		└── logs/ # app log files
146	146		```
147	147
148		-	Drop a `SYSTEM.md` in `$ZUT_HOME` to replace the built-in identity for every run. Existing append addenda, including `AGENTS.md`, skills, and enabled Ponytail coding mode, remain layered on top. `--system-prompt` wins per invocation unless a selected replace-mode subagent profile supplies the child's identity. Delete the file to revert to the default identity.
	148	+	Drop a `SYSTEM.md` in `$ZUT_HOME` to replace the built-in identity and docs guidance for every run. Existing append addenda, including `AGENTS.md`, skills, and enabled Ponytail coding mode, remain layered on top. `--system-prompt` wins per invocation unless a selected replace-mode subagent profile supplies the child's identity. Delete the file to revert to the default identity.
149	149
150	150		Ponytail coding mode is enabled by default, including when an existing `config.json` does not contain a `ponytail_enabled` field. When disabled, its compact guidance is omitted from resolved system prompts; when enabled, the guidance tells the model to apply itself only to engineering work. The setting persists its explicit on/off choice in `$ZUT_HOME/config.json` and can be changed from `/settings`. It is advisory guidance only: it does not change tool permissions, confirmations, jail behavior, or sandboxing.
151	151
152	152		A selected subagent profile with `systemPromptMode: replace` supplies the child's base identity even when `--system-prompt` was provided for the parent run. Global append addenda, including enabled Ponytail guidance, still follow their normal inclusion rules.
153	153
	154	+	### HTTP proxy
	155	+
	156	+	To route zut-managed HTTP and HTTPS requests through one proxy, add `http_proxy` to `$ZUT_HOME/config.json`:
	157	+
	158	+	```json
	159	+	{
	160	+	"http_proxy": "http://127.0.0.1:7890"
	161	+	}
	162	+	```
	163	+
	164	+	The setting is applied at startup to both HTTP and HTTPS traffic. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. `NO_PROXY` and `no_proxy` continue to control bypasses. Restart zut after changing the config file. If the URL contains proxy credentials, prefer protected environment variables because `config.json` is not a credential store.
	165	+
154	166		## Persistent instructions (AGENTS.md)
155	167
156	168		Use `AGENTS.md` to give zut standing instructions that layer on top of the default system prompt, without replacing it. This is the friendliest way to shape behavior (for example, taming local models that jump straight to code edits) because it adds guidance rather than taking over the base identity the way `SYSTEM.md` does.

@@ -174,9 +186,9 @@ Treat questions and discussions as requests for explanation. Do not edit files o
174	186		\|---\|---\|---\|
175	187		\| `$ZUT_HOME/AGENTS.md` \| global, every run \| appended to the default prompt \|
176	188		\| `./AGENTS.md` (and parent dirs) \| project \| appended to the default prompt \|
177		-	\| `$ZUT_HOME/SYSTEM.md` \| global, every run \| replaces the built-in identity; append addenda remain \|
	189	+	\| `$ZUT_HOME/SYSTEM.md` \| global, every run \| replaces the built-in identity and docs guidance; append addenda remain \|
178	190		\| `--append-system-prompt <text>` \| single run \| appends for one invocation (repeatable) \|
179		-	\| `--system-prompt <text>` \| single run \| replaces the built-in identity for one invocation; append addenda remain \|
	191	+	\| `--system-prompt <text>` \| single run \| replaces the built-in identity and docs guidance for one invocation; append addenda remain \|
180	192
181	193		> Note: zut does not read a `CLAUDE.md` instruction file. The only Claude-compatible thing it picks up is skills under `.claude/skills/`. If you are migrating from Claude Code, move that content into `AGENTS.md` (global or per-project) and zut will use it.
182	194

@@ -190,6 +202,7 @@ The first time you launch a newer zut binary, the TUI shows the GitHub release n
190	202		zut # interactive tui
191	203		zut "fix the failing test" # tui, pre-filled prompt
192	204		zut -p "list all go files" # print final text, exit
	205	+	cat README.md \| zut "summarize this text" # pipe input and infer print mode
193	206		zut -p --stats stats.json "task" # print final text and write generation stats
194	207		zut --json "refactor main.go" # newline-delimited json events, exit
195	208		zut -p --orchestrate --provider openai --model gpt-5 "implement and synthesize"

@@ -228,6 +241,7 @@ Print-mode stats contain `provider`, `model`, `prompt_tokens`, `reasoning_tokens
228	241		\| `-e`, `--ext <path>` \| Load an extension from `<path>` for this run (repeatable; wins against installed extensions of the same name). \|
229	242		\| `--no-ext` \| Skip extension discovery for this run. `--ext` still works on top, so `--no-ext --ext ./x` runs only `x`. \|
230	243		\| `--no-skill` \| Disable all skills, including built-ins. No `skill` tool is registered and the system prompt has no skill manifest. \|
	244	+	\| `--no-context-files`, `-nc` \| Disable discovery and loading of global and project `AGENTS.md` files. \|
231	245		\| `--no-yolo` \| Confirm every tool call before it runs (interactive TUI only). A dialog shows the tool name and a one-line preview of its args, while edit calls show the proposed diff in the tool panel, with four choices: yes, yes-always-this-tool-this-session, yes-always-this-session, no. Press `/` while the dialog is focused to run a slash command first; closing its input or child dialog returns to the pending confirmation. Ignored with a stderr warning in print / json / rpc modes, where tools still run freely so scripts and automation keep working. \|
232	246
233	247		## Tools

@@ -248,6 +262,7 @@ When the sandbox is on (see `/jail`), filesystem tools and LSP workspace edits r
248	262		- Interactive (default): chat TUI with streaming output, spinner, cost meter, slash commands.
249	263		- Print: `zut -p "prompt"` runs the agent to completion and writes only the final assistant text to stdout.
250	264		- Stream: `zut --stream "prompt"` runs without the TUI and writes assistant text to stdout as it arrives. Tool activity goes to stderr.
	265	+	- Piped input: stdin implies print mode when no explicit mode is selected. Print, stream, and JSON modes prepend piped input to the positional prompt, separated by a newline. For example, `cat README.md \| zut "summarize this text"`.
251	266		- JSON: `zut --json "prompt"` emits one JSON object per agent event to stdout, newline-delimited. The schema is documented in [docs/rpc.md](docs/rpc.md).
252	267		- RPC: `zut rpc` runs as a long-lived child process; commands in on stdin, events and responses out on stdout, both as NDJSON. Designed for embedding zut in third-party apps written in any language. See [docs/rpc.md](docs/rpc.md) for the wire schema and `examples/rpc/{python,node,shell,go}` for working clients.
253	268

@@ -518,7 +533,7 @@ Place a `models.json` in `$ZUT_HOME` (`$XDG_STATE_HOME/zut/` when set, otherwise
518	533
519	534		Supported fields per model: `id` (required), `name`, `reasoning`, `reasoningLevelMap`, `contextWindow`, `maxTokens`, `baseUrl`, `priceInput`, `priceOutput`, `priceCacheRead`, `priceCacheWrite`.
520	535
521		-	`reasoningLevelMap` is optional. Protocol defaults apply when it is omitted. Add only model-specific exceptions using `minimum`, `low`, `medium`, `high`, `xhigh`, or `max` as keys. Map a key to another level when both inputs are equivalent, or to an empty string or `off` to remove it. The same effective mapping drives `/reasoning` and provider requests.
	536	+	`reasoningLevelMap` is optional. Protocol defaults apply when it is omitted. Add only model-specific exceptions using `minimum`, `low`, `medium`, `high`, `xhigh`, or `max` as keys. Map a key to another level when both inputs are equivalent, use an identity mapping such as `"max": "max"` to enable a level beyond the protocol default, or map it to an empty string or `off` to remove it. The same effective mapping drives `/reasoning` and provider requests.
522	537
523	538		Provider keys are normalized: `openai-codex` and `openai-responses` map to `openai`, `anthropic-messages` maps to `anthropic`, `moonshot`, `moonshot-ai`, and `kimi-code` map to `kimi`, and `deepseek-chat` and `deepseek-ai` map to `deepseek`. Built-in provider ids such as `groq`, `openrouter`, `github-copilot`, `amazon-bedrock`, `google-vertex`, `azure-openai-responses`, `fireworks`, `vercel-ai-gateway`, `mistral`, and `xai` can also be used directly.
524	539

 └── logs/               # app log files
 ```
-Drop a `SYSTEM.md` in `$ZUT_HOME` to replace the built-in identity for every run. Existing append addenda, including `AGENTS.md`, skills, and enabled Ponytail coding mode, remain layered on top. `--system-prompt` wins per invocation unless a selected replace-mode subagent profile supplies the child's identity. Delete the file to revert to the default identity.
+Drop a `SYSTEM.md` in `$ZUT_HOME` to replace the built-in identity and docs guidance for every run. Existing append addenda, including `AGENTS.md`, skills, and enabled Ponytail coding mode, remain layered on top. `--system-prompt` wins per invocation unless a selected replace-mode subagent profile supplies the child's identity. Delete the file to revert to the default identity.
 Ponytail coding mode is enabled by default, including when an existing `config.json` does not contain a `ponytail_enabled` field. When disabled, its compact guidance is omitted from resolved system prompts; when enabled, the guidance tells the model to apply itself only to engineering work. The setting persists its explicit on/off choice in `$ZUT_HOME/config.json` and can be changed from `/settings`. It is advisory guidance only: it does not change tool permissions, confirmations, jail behavior, or sandboxing.
 A selected subagent profile with `systemPromptMode: replace` supplies the child's base identity even when `--system-prompt` was provided for the parent run. Global append addenda, including enabled Ponytail guidance, still follow their normal inclusion rules.
+### HTTP proxy
+To route zut-managed HTTP and HTTPS requests through one proxy, add `http_proxy` to `$ZUT_HOME/config.json`:
+```json
+{
+  "http_proxy": "http://127.0.0.1:7890"
+}
+```
+The setting is applied at startup to both HTTP and HTTPS traffic. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. `NO_PROXY` and `no_proxy` continue to control bypasses. Restart zut after changing the config file. If the URL contains proxy credentials, prefer protected environment variables because `config.json` is not a credential store.
 ## Persistent instructions (AGENTS.md)
 Use `AGENTS.md` to give zut standing instructions that layer **on top of** the default system prompt, without replacing it. This is the friendliest way to shape behavior (for example, taming local models that jump straight to code edits) because it adds guidance rather than taking over the base identity the way `SYSTEM.md` does.
 |---|---|---|
 | `$ZUT_HOME/AGENTS.md` | global, every run | appended to the default prompt |
 | `./AGENTS.md` (and parent dirs) | project | appended to the default prompt |
-| `$ZUT_HOME/SYSTEM.md` | global, every run | replaces the built-in identity; append addenda remain |
+| `$ZUT_HOME/SYSTEM.md` | global, every run | replaces the built-in identity and docs guidance; append addenda remain |
 | `--append-system-prompt <text>` | single run | appends for one invocation (repeatable) |
-| `--system-prompt <text>` | single run | replaces the built-in identity for one invocation; append addenda remain |
+| `--system-prompt <text>` | single run | replaces the built-in identity and docs guidance for one invocation; append addenda remain |
 > **Note:** zut does not read a `CLAUDE.md` instruction file. The only Claude-compatible thing it picks up is skills under `.claude/skills/`. If you are migrating from Claude Code, move that content into `AGENTS.md` (global or per-project) and zut will use it.
 zut                              # interactive tui
 zut "fix the failing test"       # tui, pre-filled prompt
 zut -p "list all go files"       # print final text, exit
+cat README.md | zut "summarize this text" # pipe input and infer print mode
 zut -p --stats stats.json "task" # print final text and write generation stats
 zut --json "refactor main.go"    # newline-delimited json events, exit
 zut -p --orchestrate --provider openai --model gpt-5 "implement and synthesize"
 | `-e`, `--ext <path>` | Load an extension from `<path>` for this run (repeatable; wins against installed extensions of the same name). |
 | `--no-ext` | Skip extension discovery for this run. `--ext` still works on top, so `--no-ext --ext ./x` runs only `x`. |
 | `--no-skill` | Disable all skills, including built-ins. No `skill` tool is registered and the system prompt has no skill manifest. |
+| `--no-context-files`, `-nc` | Disable discovery and loading of global and project `AGENTS.md` files. |
 | `--no-yolo` | Confirm every tool call before it runs (interactive TUI only). A dialog shows the tool name and a one-line preview of its args, while edit calls show the proposed diff in the tool panel, with four choices: yes, yes-always-this-tool-this-session, yes-always-this-session, no. Press `/` while the dialog is focused to run a slash command first; closing its input or child dialog returns to the pending confirmation. Ignored with a stderr warning in print / json / rpc modes, where tools still run freely so scripts and automation keep working. |
 ## Tools
 - **Interactive** (default): chat TUI with streaming output, spinner, cost meter, slash commands.
 - **Print**: `zut -p "prompt"` runs the agent to completion and writes only the final assistant text to stdout.
 - **Stream**: `zut --stream "prompt"` runs without the TUI and writes assistant text to stdout as it arrives. Tool activity goes to stderr.
+- **Piped input**: stdin implies print mode when no explicit mode is selected. Print, stream, and JSON modes prepend piped input to the positional prompt, separated by a newline. For example, `cat README.md | zut "summarize this text"`.
 - **JSON**: `zut --json "prompt"` emits one JSON object per agent event to stdout, newline-delimited. The schema is documented in [docs/rpc.md](docs/rpc.md).
 - **RPC**: `zut rpc` runs as a long-lived child process; commands in on stdin, events and responses out on stdout, both as NDJSON. Designed for embedding zut in third-party apps written in any language. See [docs/rpc.md](docs/rpc.md) for the wire schema and `examples/rpc/{python,node,shell,go}` for working clients.
 Supported fields per model: `id` (required), `name`, `reasoning`, `reasoningLevelMap`, `contextWindow`, `maxTokens`, `baseUrl`, `priceInput`, `priceOutput`, `priceCacheRead`, `priceCacheWrite`.
-`reasoningLevelMap` is optional. Protocol defaults apply when it is omitted. Add only model-specific exceptions using `minimum`, `low`, `medium`, `high`, `xhigh`, or `max` as keys. Map a key to another level when both inputs are equivalent, or to an empty string or `off` to remove it. The same effective mapping drives `/reasoning` and provider requests.
+`reasoningLevelMap` is optional. Protocol defaults apply when it is omitted. Add only model-specific exceptions using `minimum`, `low`, `medium`, `high`, `xhigh`, or `max` as keys. Map a key to another level when both inputs are equivalent, use an identity mapping such as `"max": "max"` to enable a level beyond the protocol default, or map it to an empty string or `off` to remove it. The same effective mapping drives `/reasoning` and provider requests.
 Provider keys are normalized: `openai-codex` and `openai-responses` map to `openai`, `anthropic-messages` maps to `anthropic`, `moonshot`, `moonshot-ai`, and `kimi-code` map to `kimi`, and `deepseek-chat` and `deepseek-ai` map to `deepseek`. Built-in provider ids such as `groq`, `openrouter`, `github-copilot`, `amazon-bedrock`, `google-vertex`, `azure-openai-responses`, `fireworks`, `vercel-ai-gateway`, `mistral`, and `xai` can also be used directly.

M docs/providers.md +14 -0


@@ -4,6 +4,20 @@ zut ships with built-in providers and a model catalog. You can select models
4	4		with `/model`, list them with `zut --list-models`, and add private models in
5	5		`$ZUT_HOME/models.json`.
6	6
	7	+	## HTTP proxy
	8	+
	9	+	Set one global proxy for zut-managed HTTP and HTTPS traffic with the `http_proxy` key in `$ZUT_HOME/config.json`:
	10	+
	11	+	```json
	12	+	{
	13	+	"http_proxy": "http://127.0.0.1:7890"
	14	+	}
	15	+	```
	16	+
	17	+	zut applies this setting at startup. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. Standard `NO_PROXY` and `no_proxy` bypass lists remain effective. Restart zut after changing the setting.
	18	+
	19	+	`config.json` is not a credential store. If the proxy URL contains a username or password, prefer setting the proxy environment variables in a protected shell or service configuration rather than saving those credentials in `config.json`.
	20	+
7	21		## Login methods
8	22
9	23		Use `/login` in interactive mode. Type in either provider picker to filter the list by provider ID or display name.


@@ -4,6 +4,20 @@ zut ships with built-in providers and a model catalog. You can select models
4	with `/model`, list them with `zut --list-models`, and add private models in	4		with `/model`, list them with `zut --list-models`, and add private models in
5	`$ZUT_HOME/models.json`.	5		`$ZUT_HOME/models.json`.
6		6
		7	+	## HTTP proxy
		8	+
		9	+	Set one global proxy for zut-managed HTTP and HTTPS traffic with the `http_proxy` key in `$ZUT_HOME/config.json`:
		10	+
		11	+	```json
		12	+	{
		13	+	"http_proxy": "http://127.0.0.1:7890"
		14	+	}
		15	+	```
		16	+
		17	+	zut applies this setting at startup. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. Standard `NO_PROXY` and `no_proxy` bypass lists remain effective. Restart zut after changing the setting.
		18	+
		19	+	`config.json` is not a credential store. If the proxy URL contains a username or password, prefer setting the proxy environment variables in a protected shell or service configuration rather than saving those credentials in `config.json`.
		20	+
7	## Login methods	21		## Login methods
8		22
9	Use `/login` in interactive mode. Type in either provider picker to filter the list by provider ID or display name.	23		Use `/login` in interactive mode. Type in either provider picker to filter the list by provider ID or display name.

M docs/rpc.md +2 -2


@@ -23,9 +23,9 @@ You'll see one JSON object per line on stdout: a response acknowledging the prom
23	23
24	24		## Flags
25	25
26		-	`zut rpc` accepts the same flags as the other modes: `--provider`, `--model`, `--cwd`, `--api-key`, `--base-url`, `--system-prompt`, `--append-system-prompt`, `--reasoning`, `--max-steps`, `--no-tools`, `--tools`. Sessions are disabled by default in RPC mode — the embedding application owns persistence.
	26	+	`zut rpc` accepts the same flags as the other modes: `--provider`, `--model`, `--cwd`, `--api-key`, `--base-url`, `--system-prompt`, `--append-system-prompt`, `--reasoning`, `--max-steps`, `--no-tools`, `--tools`, `--no-context-files`. Sessions are disabled by default in RPC mode — the embedding application owns persistence.
27	27
28		-	RPC agents use the same resolved system-prompt policy as other modes. Ponytail coding mode is enabled when `ponytail_enabled` is missing or `true` in `$ZUT_HOME/config.json`, and disabled when it is `false`. A custom `--system-prompt` replaces the built-in identity, while enabled append addenda remain present.
	28	+	RPC agents use the same resolved system-prompt policy as other modes. Ponytail coding mode is enabled when `ponytail_enabled` is missing or `true` in `$ZUT_HOME/config.json`, and disabled when it is `false`. A custom `--system-prompt` replaces the built-in identity and docs guidance, while enabled append addenda remain present.
29	29
30	30		## Auth
31	31


@@ -23,9 +23,9 @@ You'll see one JSON object per line on stdout: a response acknowledging the prom
23			23
24		## Flags	24		## Flags
25			25
26	-	`zut rpc` accepts the same flags as the other modes: `--provider`, `--model`, `--cwd`, `--api-key`, `--base-url`, `--system-prompt`, `--append-system-prompt`, `--reasoning`, `--max-steps`, `--no-tools`, `--tools`. Sessions are disabled by default in RPC mode — the embedding application owns persistence.	26	+	`zut rpc` accepts the same flags as the other modes: `--provider`, `--model`, `--cwd`, `--api-key`, `--base-url`, `--system-prompt`, `--append-system-prompt`, `--reasoning`, `--max-steps`, `--no-tools`, `--tools`, `--no-context-files`. Sessions are disabled by default in RPC mode — the embedding application owns persistence.
27			27
28	-	RPC agents use the same resolved system-prompt policy as other modes. Ponytail coding mode is enabled when `ponytail_enabled` is missing or `true` in `$ZUT_HOME/config.json`, and disabled when it is `false`. A custom `--system-prompt` replaces the built-in identity, while enabled append addenda remain present.	28	+	RPC agents use the same resolved system-prompt policy as other modes. Ponytail coding mode is enabled when `ponytail_enabled` is missing or `true` in `$ZUT_HOME/config.json`, and disabled when it is `false`. A custom `--system-prompt` replaces the built-in identity and docs guidance, while enabled append addenda remain present.
29			29
30		## Auth	30		## Auth
31			31

M go.mod +1 -0


@@ -9,6 +9,7 @@ require (
9	9		github.com/mattn/go-runewidth v0.0.16
10	10		github.com/sahilm/fuzzy v0.1.1
11	11		golang.org/x/image v0.43.0
	12	+	golang.org/x/mod v0.35.0
12	13		golang.org/x/net v0.57.0
13	14		)
14	15


@@ -9,6 +9,7 @@ require (
9	github.com/mattn/go-runewidth v0.0.16	9		github.com/mattn/go-runewidth v0.0.16
10	github.com/sahilm/fuzzy v0.1.1	10		github.com/sahilm/fuzzy v0.1.1
11	golang.org/x/image v0.43.0	11		golang.org/x/image v0.43.0
		12	+	golang.org/x/mod v0.35.0
12	golang.org/x/net v0.57.0	13		golang.org/x/net v0.57.0
13	)	14		)
14		15

M go.sum +2 -0


@@ -22,6 +22,8 @@ github.com/sahilm/fuzzy v0.1.1 h1:ceu5RHF8DGgoi+/dR5PsECjCDH1BE3Fnmpo7aVXOdRA=
22	22		github.com/sahilm/fuzzy v0.1.1/go.mod h1:VFvziUEIMCrT6A6tw2RFIXPXXmzXbOsSHF0DOI8ZK9Y=
23	23		golang.org/x/image v0.43.0 h1:FLxcP4ec2350nTfOC8ysKtqYSIFbk/QGjw1ZHNP4tsY=
24	24		golang.org/x/image v0.43.0/go.mod h1:rrpelvGFt+kLPAjPM4HeWPgrl0FtafueU//e5N0qk/Q=
	25	+	golang.org/x/mod v0.35.0 h1:Ww1D637e6Pg+Zb2KrWfHQUnH2dQRLBQyAtpr/haaJeM=
	26	+	golang.org/x/mod v0.35.0/go.mod h1:+GwiRhIInF8wPm+4AoT6L0FA1QWAad3OMdTRx4tFYlU=
25	27		golang.org/x/net v0.57.0 h1:K5+3DljvIuDG9/Jv9rvyMywYNFCQ9RSUY6OOTTkT+tE=
26	28		golang.org/x/net v0.57.0/go.mod h1:KpXc8iv+r3XplLAG/f7Jsf9RPszJzdR0f58q9vGOuEU=
27	29		golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=


@@ -22,6 +22,8 @@ github.com/sahilm/fuzzy v0.1.1 h1:ceu5RHF8DGgoi+/dR5PsECjCDH1BE3Fnmpo7aVXOdRA=
22	github.com/sahilm/fuzzy v0.1.1/go.mod h1:VFvziUEIMCrT6A6tw2RFIXPXXmzXbOsSHF0DOI8ZK9Y=	22		github.com/sahilm/fuzzy v0.1.1/go.mod h1:VFvziUEIMCrT6A6tw2RFIXPXXmzXbOsSHF0DOI8ZK9Y=
23	golang.org/x/image v0.43.0 h1:FLxcP4ec2350nTfOC8ysKtqYSIFbk/QGjw1ZHNP4tsY=	23		golang.org/x/image v0.43.0 h1:FLxcP4ec2350nTfOC8ysKtqYSIFbk/QGjw1ZHNP4tsY=
24	golang.org/x/image v0.43.0/go.mod h1:rrpelvGFt+kLPAjPM4HeWPgrl0FtafueU//e5N0qk/Q=	24		golang.org/x/image v0.43.0/go.mod h1:rrpelvGFt+kLPAjPM4HeWPgrl0FtafueU//e5N0qk/Q=
		25	+	golang.org/x/mod v0.35.0 h1:Ww1D637e6Pg+Zb2KrWfHQUnH2dQRLBQyAtpr/haaJeM=
		26	+	golang.org/x/mod v0.35.0/go.mod h1:+GwiRhIInF8wPm+4AoT6L0FA1QWAad3OMdTRx4tFYlU=
25	golang.org/x/net v0.57.0 h1:K5+3DljvIuDG9/Jv9rvyMywYNFCQ9RSUY6OOTTkT+tE=	27		golang.org/x/net v0.57.0 h1:K5+3DljvIuDG9/Jv9rvyMywYNFCQ9RSUY6OOTTkT+tE=
26	golang.org/x/net v0.57.0/go.mod h1:KpXc8iv+r3XplLAG/f7Jsf9RPszJzdR0f58q9vGOuEU=	28		golang.org/x/net v0.57.0/go.mod h1:KpXc8iv+r3XplLAG/f7Jsf9RPszJzdR0f58q9vGOuEU=
27	golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=	29		golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=

M packages/agent/args.go +7 -0


@@ -106,6 +106,10 @@ type Args struct {
106	106		// skill discovery, including built-ins.
107	107		WithSkills bool
108	108
	109	+	// NoContextFiles disables discovery and loading of AGENTS.md files
	110	+	// from $ZUT_HOME and the cwd's ancestor directories.
	111	+	NoContextFiles bool
	112	+
109	113		// InsecureTLS skips TLS verification for custom inference endpoints.
110	114		InsecureTLS bool
111	115

@@ -275,6 +279,8 @@ func ParseArgs(in []string) (Args, error) {
275	279		a.NoExt = true
276	280		case "--no-skill", "--no-skills":
277	281		a.NoSkill = true
	282	+	case "--no-context-files", "-nc":
	283	+	a.NoContextFiles = true
278	284		case "--with-skills", "--with-skill":
279	285		// Deprecated no-op: user skills are loaded by default.
280	286		a.WithSkills = true

@@ -575,6 +581,7 @@ func PrintHelp(version string) {
575	581		row{"-y, --yes", "accept zut run consent without prompting"},
576	582		row{"--no-ext", "skip extension discovery for this run"},
577	583		row{"--no-skill", "skip all skill discovery for this run"},
	584	+	row{"--no-context-files, -nc", "skip AGENTS.md discovery for this run"},
578	585		)
579	586		section("misc",
580	587		row{"--max-steps N", "agent loop iteration cap (default: unlimited)"},


@@ -106,6 +106,10 @@ type Args struct {
106	// skill discovery, including built-ins.	106		// skill discovery, including built-ins.
107	WithSkills bool	107		WithSkills bool
108		108
		109	+	// NoContextFiles disables discovery and loading of AGENTS.md files
		110	+	// from $ZUT_HOME and the cwd's ancestor directories.
		111	+	NoContextFiles bool
		112	+
109	// InsecureTLS skips TLS verification for custom inference endpoints.	113		// InsecureTLS skips TLS verification for custom inference endpoints.
110	InsecureTLS bool	114		InsecureTLS bool
111		115

@@ -275,6 +279,8 @@ func ParseArgs(in []string) (Args, error) {
275	a.NoExt = true	279		a.NoExt = true
276	case "--no-skill", "--no-skills":	280		case "--no-skill", "--no-skills":
277	a.NoSkill = true	281		a.NoSkill = true
		282	+	case "--no-context-files", "-nc":
		283	+	a.NoContextFiles = true
278	case "--with-skills", "--with-skill":	284		case "--with-skills", "--with-skill":
279	// Deprecated no-op: user skills are loaded by default.	285		// Deprecated no-op: user skills are loaded by default.
280	a.WithSkills = true	286		a.WithSkills = true

@@ -575,6 +581,7 @@ func PrintHelp(version string) {
575	row{"-y, --yes", "accept zut run consent without prompting"},	581		row{"-y, --yes", "accept zut run consent without prompting"},
576	row{"--no-ext", "skip extension discovery for this run"},	582		row{"--no-ext", "skip extension discovery for this run"},
577	row{"--no-skill", "skip all skill discovery for this run"},	583		row{"--no-skill", "skip all skill discovery for this run"},
		584	+	row{"--no-context-files, -nc", "skip AGENTS.md discovery for this run"},
578	)	585		)
579	section("misc",	586		section("misc",
580	row{"--max-steps N", "agent loop iteration cap (default: unlimited)"},	587		row{"--max-steps N", "agent loop iteration cap (default: unlimited)"},

M packages/agent/args_test.go +12 -0


@@ -188,6 +188,18 @@ func TestParseArgsStatsRequiresPrintMode(t *testing.T) {
188	188		}
189	189		}
190	190
	191	+	func TestParseArgsNoContextFiles(t *testing.T) {
	192	+	for _, flag := range []string{"--no-context-files", "-nc"} {
	193	+	args, err := ParseArgs([]string{flag})
	194	+	if err != nil {
	195	+	t.Fatalf("ParseArgs(%q): %v", flag, err)
	196	+	}
	197	+	if !args.NoContextFiles {
	198	+	t.Fatalf("ParseArgs(%q): NoContextFiles = false", flag)
	199	+	}
	200	+	}
	201	+	}
	202	+
191	203		func TestParseArgsStream(t *testing.T) {
192	204		args, err := ParseArgs([]string{"--stream", "hi"})
193	205		if err != nil {


@@ -188,6 +188,18 @@ func TestParseArgsStatsRequiresPrintMode(t *testing.T) {
188	}	188		}
189	}	189		}
190		190
		191	+	func TestParseArgsNoContextFiles(t *testing.T) {
		192	+	for _, flag := range []string{"--no-context-files", "-nc"} {
		193	+	args, err := ParseArgs([]string{flag})
		194	+	if err != nil {
		195	+	t.Fatalf("ParseArgs(%q): %v", flag, err)
		196	+	}
		197	+	if !args.NoContextFiles {
		198	+	t.Fatalf("ParseArgs(%q): NoContextFiles = false", flag)
		199	+	}
		200	+	}
		201	+	}
		202	+
191	func TestParseArgsStream(t *testing.T) {	203		func TestParseArgsStream(t *testing.T) {
192	args, err := ParseArgs([]string{"--stream", "hi"})	204		args, err := ParseArgs([]string{"--stream", "hi"})
193	if err != nil {	205		if err != nil {

M packages/agent/build.go +1 -1


@@ -842,7 +842,7 @@ func Resolve(args Args, requireCred bool) (Resolved, error) {
842	842		summaries := toolSummaries(reg, args)
843	843
844	844		contextFiles := []ContextFile(nil)
845		-	if selectedProfile == nil \|\| selectedProfile.InheritProjectContext == nil \|\| *selectedProfile.InheritProjectContext {
	845	+	if !args.NoContextFiles && (selectedProfile == nil \|\| selectedProfile.InheritProjectContext == nil \|\| *selectedProfile.InheritProjectContext) {
846	846		contextFiles = loadAgentsContext(args.CWD, ZutHome())
847	847		}
848	848		append_ := []string(nil)


@@ -842,7 +842,7 @@ func Resolve(args Args, requireCred bool) (Resolved, error) {
842		summaries := toolSummaries(reg, args)	842		summaries := toolSummaries(reg, args)
843			843
844		contextFiles := []ContextFile(nil)	844		contextFiles := []ContextFile(nil)
845	-	if selectedProfile == nil \|\| selectedProfile.InheritProjectContext == nil \|\| *selectedProfile.InheritProjectContext {	845	+	if !args.NoContextFiles && (selectedProfile == nil \|\| selectedProfile.InheritProjectContext == nil \|\| *selectedProfile.InheritProjectContext) {
846		contextFiles = loadAgentsContext(args.CWD, ZutHome())	846		contextFiles = loadAgentsContext(args.CWD, ZutHome())
847		}	847		}
848		append_ := []string(nil)	848		append_ := []string(nil)

M packages/agent/build_test.go +29 -0


@@ -708,6 +708,35 @@ func TestReadAgentsContextMissingFilesIsEmpty(t *testing.T) {
708	708		}
709	709		}
710	710
	711	+	func TestResolveNoContextFilesSkipsAgentsInstructions(t *testing.T) {
	712	+	zutHome := t.TempDir()
	713	+	cwd := t.TempDir()
	714	+	t.Setenv("ZUT_HOME", zutHome)
	715	+	t.Setenv("OPENAI_API_KEY", "test-key")
	716	+	if err := os.WriteFile(filepath.Join(zutHome, "AGENTS.md"), []byte("global rule"), 0o644); err != nil {
	717	+	t.Fatal(err)
	718	+	}
	719	+	if err := os.WriteFile(filepath.Join(cwd, "AGENTS.md"), []byte("project rule"), 0o644); err != nil {
	720	+	t.Fatal(err)
	721	+	}
	722	+
	723	+	r, err := Resolve(Args{
	724	+	Provider: "openai",
	725	+	Model: "gpt-5",
	726	+	CWD: cwd,
	727	+	NoContextFiles: true,
	728	+	}, false)
	729	+	if err != nil {
	730	+	t.Fatal(err)
	731	+	}
	732	+	if len(r.ContextFiles) != 0 {
	733	+	t.Fatalf("ContextFiles = %#v, want empty", r.ContextFiles)
	734	+	}
	735	+	if strings.Contains(r.SystemPrompt, "global rule") \|\| strings.Contains(r.SystemPrompt, "project rule") {
	736	+	t.Fatalf("system prompt contains disabled context files:\n%s", r.SystemPrompt)
	737	+	}
	738	+	}
	739	+
711	740		// TestResolveFallsBackWhenConfiguredModelIsGone reproduces the
712	741		// startup failure caught by the user's screenshot: the persisted
713	742		// config.json points at a model id that's no longer in the active


@@ -708,6 +708,35 @@ func TestReadAgentsContextMissingFilesIsEmpty(t *testing.T) {
708	}	708		}
709	}	709		}
710		710
		711	+	func TestResolveNoContextFilesSkipsAgentsInstructions(t *testing.T) {
		712	+	zutHome := t.TempDir()
		713	+	cwd := t.TempDir()
		714	+	t.Setenv("ZUT_HOME", zutHome)
		715	+	t.Setenv("OPENAI_API_KEY", "test-key")
		716	+	if err := os.WriteFile(filepath.Join(zutHome, "AGENTS.md"), []byte("global rule"), 0o644); err != nil {
		717	+	t.Fatal(err)
		718	+	}
		719	+	if err := os.WriteFile(filepath.Join(cwd, "AGENTS.md"), []byte("project rule"), 0o644); err != nil {
		720	+	t.Fatal(err)
		721	+	}
		722	+
		723	+	r, err := Resolve(Args{
		724	+	Provider: "openai",
		725	+	Model: "gpt-5",
		726	+	CWD: cwd,
		727	+	NoContextFiles: true,
		728	+	}, false)
		729	+	if err != nil {
		730	+	t.Fatal(err)
		731	+	}
		732	+	if len(r.ContextFiles) != 0 {
		733	+	t.Fatalf("ContextFiles = %#v, want empty", r.ContextFiles)
		734	+	}
		735	+	if strings.Contains(r.SystemPrompt, "global rule") \|\| strings.Contains(r.SystemPrompt, "project rule") {
		736	+	t.Fatalf("system prompt contains disabled context files:\n%s", r.SystemPrompt)
		737	+	}
		738	+	}
		739	+
711	// TestResolveFallsBackWhenConfiguredModelIsGone reproduces the	740		// TestResolveFallsBackWhenConfiguredModelIsGone reproduces the
712	// startup failure caught by the user's screenshot: the persisted	741		// startup failure caught by the user's screenshot: the persisted
713	// config.json points at a model id that's no longer in the active	742		// config.json points at a model id that's no longer in the active

M packages/agent/cli.go +31 -15


@@ -765,6 +765,10 @@ func fanoutAgentEvent(mgr *extensions.Manager, ev core.AgentEvent) {
765	765
766	766		// Run is the top-level entrypoint for the zut binary.
767	767		func Run(rawArgs []string, version string) error {
	768	+	// Apply network configuration before any subcommand can make an HTTP
	769	+	// request. Standard proxy environment variables retain precedence.
	770	+	applyConfiguredHTTPProxy()
	771	+
768	772		// Extension installs can invoke external build and clone processes.
769	773		// Give those processes a cancellation context without changing signal
770	774		// handling for the interactive modes below.

@@ -800,6 +804,9 @@ func runWithArgsRaw(rawArgs []string, version string) error {
800	804		rawArgs = append([]string{"--rpc"}, rawArgs[1:]...)
801	805		}
802	806
	807	+	if fi, err := os.Stdin.Stat(); err == nil {
	808	+	rawArgs = argsForStdin(rawArgs, fi.Mode())
	809	+	}
803	810		args, err := ParseArgs(rawArgs)
804	811		if err != nil {
805	812		PrintHelp(version)

@@ -1005,11 +1012,8 @@ func runPrintMode(ctx context.Context, args Args, version string) (runErr error)
1005	1012		defer func() { runErr = errors.Join(runErr, sess.FlushRetryLifecycle(ag.LastTurnUsage(), ag.Cost())) }()
1006	1013		}
1007	1014
1008		-	prompt := args.Prompt
1009		-	if prompt == "" {
1010		-	piped, _ := readAllStdin()
1011		-	prompt = strings.TrimSpace(piped)
1012		-	}
	1015	+	piped, _ := readAllStdin()
	1016	+	prompt := combinePromptInput(piped, args.Prompt)
1013	1017		if prompt == "" {
1014	1018		return fmt.Errorf("print mode requires a prompt (arg or stdin)")
1015	1019		}

@@ -1090,11 +1094,8 @@ func runStreamMode(ctx context.Context, args Args, version string) (runErr error
1090	1094		defer func() { runErr = errors.Join(runErr, sess.FlushRetryLifecycle(ag.LastTurnUsage(), ag.Cost())) }()
1091	1095		}
1092	1096
1093		-	prompt := args.Prompt
1094		-	if prompt == "" {
1095		-	piped, _ := readAllStdin()
1096		-	prompt = strings.TrimSpace(piped)
1097		-	}
	1097	+	piped, _ := readAllStdin()
	1098	+	prompt := combinePromptInput(piped, args.Prompt)
1098	1099		if prompt == "" {
1099	1100		return fmt.Errorf("stream mode requires a prompt (arg or stdin)")
1100	1101		}

@@ -1214,11 +1215,8 @@ func runJSONMode(ctx context.Context, args Args, version string) (runErr error)
1214	1215		defer func() { runErr = errors.Join(runErr, sess.FlushRetryLifecycle(ag.LastTurnUsage(), ag.Cost())) }()
1215	1216		}
1216	1217
1217		-	prompt := args.Prompt
1218		-	if prompt == "" {
1219		-	piped, _ := readAllStdin()
1220		-	prompt = strings.TrimSpace(piped)
1221		-	}
	1218	+	piped, _ := readAllStdin()
	1219	+	prompt := combinePromptInput(piped, args.Prompt)
1222	1220		if prompt == "" {
1223	1221		return fmt.Errorf("json mode requires a prompt (arg or stdin)")
1224	1222		}

@@ -2887,6 +2885,24 @@ func writeNewTranscriptLocked(ag core.Agent, sess core.Session, from int) (nex
2887	2885		return next, nil
2888	2886		}
2889	2887
	2888	+	func argsForStdin(rawArgs []string, mode os.FileMode) []string {
	2889	+	if mode&os.ModeCharDevice != 0 {
	2890	+	return rawArgs
	2891	+	}
	2892	+	return append([]string{"--print"}, rawArgs...)
	2893	+	}
	2894	+
	2895	+	func combinePromptInput(stdin, positional string) string {
	2896	+	parts := make([]string, 0, 2)
	2897	+	if stdin = strings.TrimSpace(stdin); stdin != "" {
	2898	+	parts = append(parts, stdin)
	2899	+	}
	2900	+	if positional = strings.TrimSpace(positional); positional != "" {
	2901	+	parts = append(parts, positional)
	2902	+	}
	2903	+	return strings.Join(parts, "\n")
	2904	+	}
	2905	+
2890	2906		func readAllStdin() (string, error) {
2891	2907		fi, err := os.Stdin.Stat()
2892	2908		if err != nil {

 // Run is the top-level entrypoint for the zut binary.
 func Run(rawArgs []string, version string) error {
+	// Apply network configuration before any subcommand can make an HTTP
+	// request. Standard proxy environment variables retain precedence.
+	applyConfiguredHTTPProxy()
 	// Extension installs can invoke external build and clone processes.
 	// Give those processes a cancellation context without changing signal
 	// handling for the interactive modes below.
 		rawArgs = append([]string{"--rpc"}, rawArgs[1:]...)
 	}
+	if fi, err := os.Stdin.Stat(); err == nil {
+		rawArgs = argsForStdin(rawArgs, fi.Mode())
+	}
 	args, err := ParseArgs(rawArgs)
 	if err != nil {
 		PrintHelp(version)
 		defer func() { runErr = errors.Join(runErr, sess.FlushRetryLifecycle(ag.LastTurnUsage(), ag.Cost())) }()
 	}
-	prompt := args.Prompt
+	piped, _ := readAllStdin()
-	if prompt == "" {
+	prompt := combinePromptInput(piped, args.Prompt)
-		piped, _ := readAllStdin()
-		prompt = strings.TrimSpace(piped)
-	}
 	if prompt == "" {
 		return fmt.Errorf("print mode requires a prompt (arg or stdin)")
 	}
 		defer func() { runErr = errors.Join(runErr, sess.FlushRetryLifecycle(ag.LastTurnUsage(), ag.Cost())) }()
 	}
-	prompt := args.Prompt
+	piped, _ := readAllStdin()
-	if prompt == "" {
+	prompt := combinePromptInput(piped, args.Prompt)
-		piped, _ := readAllStdin()
-		prompt = strings.TrimSpace(piped)
-	}
 	if prompt == "" {
 		return fmt.Errorf("stream mode requires a prompt (arg or stdin)")
 	}
 		defer func() { runErr = errors.Join(runErr, sess.FlushRetryLifecycle(ag.LastTurnUsage(), ag.Cost())) }()
 	}
-	prompt := args.Prompt
+	piped, _ := readAllStdin()
-	if prompt == "" {
+	prompt := combinePromptInput(piped, args.Prompt)
-		piped, _ := readAllStdin()
-		prompt = strings.TrimSpace(piped)
-	}
 	if prompt == "" {
 		return fmt.Errorf("json mode requires a prompt (arg or stdin)")
 	}
 	return next, nil
 }
+func argsForStdin(rawArgs []string, mode os.FileMode) []string {
+	if mode&os.ModeCharDevice != 0 {
+		return rawArgs
+	}
+	return append([]string{"--print"}, rawArgs...)
+}
+func combinePromptInput(stdin, positional string) string {
+	parts := make([]string, 0, 2)
+	if stdin = strings.TrimSpace(stdin); stdin != "" {
+		parts = append(parts, stdin)
+	}
+	if positional = strings.TrimSpace(positional); positional != "" {
+		parts = append(parts, positional)
+	}
+	return strings.Join(parts, "\n")
+}
 func readAllStdin() (string, error) {
 	fi, err := os.Stdin.Stat()
 	if err != nil {

M packages/agent/config.go +24 -0


@@ -181,6 +181,10 @@ type Config struct {
181	181		// Insecure skips TLS verification for custom inference endpoints.
182	182		Insecure bool `json:"insecure,omitempty"`
183	183
	184	+	// HTTPProxy is a global proxy URL used for HTTP and HTTPS requests when
	185	+	// the corresponding standard proxy environment variable is not already set.
	186	+	HTTPProxy string `json:"http_proxy,omitempty"`
	187	+
184	188		// LastChangelogShown is the version whose release-notes
185	189		// dialog the user has already seen. When the running binary's
186	190		// version differs, the next interactive run shows the

@@ -332,6 +336,26 @@ func SaveConfig(c Config) error {
332	336		return os.WriteFile(ConfigPath(), b, 0o644)
333	337		}
334	338
	339	+	// applyConfiguredHTTPProxy makes the persisted proxy available to Go's
	340	+	// standard HTTP transport and to child processes. Explicit environment
	341	+	// variables take precedence over config.json.
	342	+	func applyConfiguredHTTPProxy() {
	343	+	cfg, err := LoadConfig()
	344	+	if err != nil {
	345	+	return
	346	+	}
	347	+	proxy := strings.TrimSpace(cfg.HTTPProxy)
	348	+	if proxy == "" {
	349	+	return
	350	+	}
	351	+	if os.Getenv("HTTP_PROXY") == "" && os.Getenv("http_proxy") == "" {
	352	+	_ = os.Setenv("HTTP_PROXY", proxy)
	353	+	}
	354	+	if os.Getenv("HTTPS_PROXY") == "" && os.Getenv("https_proxy") == "" {
	355	+	_ = os.Setenv("HTTPS_PROXY", proxy)
	356	+	}
	357	+	}
	358	+
335	359		// AuthStoreFor returns the auth.Store backed by AuthPath().
336	360		func AuthStoreFor() *auth.Store { return auth.NewStore(AuthPath()) }
337	361


@@ -181,6 +181,10 @@ type Config struct {
181	// Insecure skips TLS verification for custom inference endpoints.	181		// Insecure skips TLS verification for custom inference endpoints.
182	Insecure bool `json:"insecure,omitempty"`	182		Insecure bool `json:"insecure,omitempty"`
183		183
		184	+	// HTTPProxy is a global proxy URL used for HTTP and HTTPS requests when
		185	+	// the corresponding standard proxy environment variable is not already set.
		186	+	HTTPProxy string `json:"http_proxy,omitempty"`
		187	+
184	// LastChangelogShown is the version whose release-notes	188		// LastChangelogShown is the version whose release-notes
185	// dialog the user has already seen. When the running binary's	189		// dialog the user has already seen. When the running binary's
186	// version differs, the next interactive run shows the	190		// version differs, the next interactive run shows the

@@ -332,6 +336,26 @@ func SaveConfig(c Config) error {
332	return os.WriteFile(ConfigPath(), b, 0o644)	336		return os.WriteFile(ConfigPath(), b, 0o644)
333	}	337		}
334		338
		339	+	// applyConfiguredHTTPProxy makes the persisted proxy available to Go's
		340	+	// standard HTTP transport and to child processes. Explicit environment
		341	+	// variables take precedence over config.json.
		342	+	func applyConfiguredHTTPProxy() {
		343	+	cfg, err := LoadConfig()
		344	+	if err != nil {
		345	+	return
		346	+	}
		347	+	proxy := strings.TrimSpace(cfg.HTTPProxy)
		348	+	if proxy == "" {
		349	+	return
		350	+	}
		351	+	if os.Getenv("HTTP_PROXY") == "" && os.Getenv("http_proxy") == "" {
		352	+	_ = os.Setenv("HTTP_PROXY", proxy)
		353	+	}
		354	+	if os.Getenv("HTTPS_PROXY") == "" && os.Getenv("https_proxy") == "" {
		355	+	_ = os.Setenv("HTTPS_PROXY", proxy)
		356	+	}
		357	+	}
		358	+
335	// AuthStoreFor returns the auth.Store backed by AuthPath().	359		// AuthStoreFor returns the auth.Store backed by AuthPath().
336	func AuthStoreFor() *auth.Store { return auth.NewStore(AuthPath()) }	360		func AuthStoreFor() *auth.Store { return auth.NewStore(AuthPath()) }
337		361

A packages/agent/config_proxy_test.go +92 -0

@@ -0,0 +1,92 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"os"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestApplyConfiguredHTTPProxy(t *testing.T) {
	9	+	t.Setenv("ZUT_HOME", t.TempDir())
	10	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	11	+	t.Setenv(key, "")
	12	+	}
	13	+	if err := SaveConfig(Config{HTTPProxy: " http://proxy.example:8080 "}); err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+
	17	+	applyConfiguredHTTPProxy()
	18	+
	19	+	if got := getenvForTest("HTTP_PROXY"); got != "http://proxy.example:8080" {
	20	+	t.Fatalf("HTTP_PROXY = %q", got)
	21	+	}
	22	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://proxy.example:8080" {
	23	+	t.Fatalf("HTTPS_PROXY = %q", got)
	24	+	}
	25	+	}
	26	+
	27	+	func TestApplyConfiguredHTTPProxyPreservesUppercaseEnvironment(t *testing.T) {
	28	+	t.Setenv("ZUT_HOME", t.TempDir())
	29	+	// Set lowercase first because environment variable names are
	30	+	// case-insensitive on Windows.
	31	+	t.Setenv("http_proxy", "")
	32	+	t.Setenv("https_proxy", "")
	33	+	t.Setenv("HTTP_PROXY", "http://http-env.example:8080")
	34	+	t.Setenv("HTTPS_PROXY", "http://https-env.example:8080")
	35	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	36	+	t.Fatal(err)
	37	+	}
	38	+
	39	+	applyConfiguredHTTPProxy()
	40	+
	41	+	if got := getenvForTest("HTTP_PROXY"); got != "http://http-env.example:8080" {
	42	+	t.Fatalf("HTTP_PROXY = %q", got)
	43	+	}
	44	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://https-env.example:8080" {
	45	+	t.Fatalf("HTTPS_PROXY = %q", got)
	46	+	}
	47	+	}
	48	+
	49	+	func TestApplyConfiguredHTTPProxyPreservesLowercaseEnvironment(t *testing.T) {
	50	+	t.Setenv("ZUT_HOME", t.TempDir())
	51	+	// Set uppercase first because environment variable names are
	52	+	// case-insensitive on Windows.
	53	+	t.Setenv("HTTP_PROXY", "")
	54	+	t.Setenv("HTTPS_PROXY", "")
	55	+	t.Setenv("http_proxy", "http://http-env.example:8080")
	56	+	t.Setenv("https_proxy", "http://https-env.example:8080")
	57	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	58	+	t.Fatal(err)
	59	+	}
	60	+
	61	+	applyConfiguredHTTPProxy()
	62	+
	63	+	if got := getenvForTest("http_proxy"); got != "http://http-env.example:8080" {
	64	+	t.Fatalf("http_proxy = %q", got)
	65	+	}
	66	+	if got := getenvForTest("https_proxy"); got != "http://https-env.example:8080" {
	67	+	t.Fatalf("https_proxy = %q", got)
	68	+	}
	69	+	}
	70	+
	71	+	func TestApplyConfiguredHTTPProxyIgnoresEmptySetting(t *testing.T) {
	72	+	t.Setenv("ZUT_HOME", t.TempDir())
	73	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	74	+	t.Setenv(key, "")
	75	+	}
	76	+	if err := SaveConfig(Config{HTTPProxy: " "}); err != nil {
	77	+	t.Fatal(err)
	78	+	}
	79	+
	80	+	applyConfiguredHTTPProxy()
	81	+
	82	+	if got := getenvForTest("HTTP_PROXY"); got != "" {
	83	+	t.Fatalf("HTTP_PROXY = %q", got)
	84	+	}
	85	+	if got := getenvForTest("HTTPS_PROXY"); got != "" {
	86	+	t.Fatalf("HTTPS_PROXY = %q", got)
	87	+	}
	88	+	}
	89	+
	90	+	func getenvForTest(key string) string {
	91	+	return os.Getenv(key)
	92	+	}

@@ -0,0 +1,92 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"os"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestApplyConfiguredHTTPProxy(t *testing.T) {
	9	+	t.Setenv("ZUT_HOME", t.TempDir())
	10	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	11	+	t.Setenv(key, "")
	12	+	}
	13	+	if err := SaveConfig(Config{HTTPProxy: " http://proxy.example:8080 "}); err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+
	17	+	applyConfiguredHTTPProxy()
	18	+
	19	+	if got := getenvForTest("HTTP_PROXY"); got != "http://proxy.example:8080" {
	20	+	t.Fatalf("HTTP_PROXY = %q", got)
	21	+	}
	22	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://proxy.example:8080" {
	23	+	t.Fatalf("HTTPS_PROXY = %q", got)
	24	+	}
	25	+	}
	26	+
	27	+	func TestApplyConfiguredHTTPProxyPreservesUppercaseEnvironment(t *testing.T) {
	28	+	t.Setenv("ZUT_HOME", t.TempDir())
	29	+	// Set lowercase first because environment variable names are
	30	+	// case-insensitive on Windows.
	31	+	t.Setenv("http_proxy", "")
	32	+	t.Setenv("https_proxy", "")
	33	+	t.Setenv("HTTP_PROXY", "http://http-env.example:8080")
	34	+	t.Setenv("HTTPS_PROXY", "http://https-env.example:8080")
	35	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	36	+	t.Fatal(err)
	37	+	}
	38	+
	39	+	applyConfiguredHTTPProxy()
	40	+
	41	+	if got := getenvForTest("HTTP_PROXY"); got != "http://http-env.example:8080" {
	42	+	t.Fatalf("HTTP_PROXY = %q", got)
	43	+	}
	44	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://https-env.example:8080" {
	45	+	t.Fatalf("HTTPS_PROXY = %q", got)
	46	+	}
	47	+	}
	48	+
	49	+	func TestApplyConfiguredHTTPProxyPreservesLowercaseEnvironment(t *testing.T) {
	50	+	t.Setenv("ZUT_HOME", t.TempDir())
	51	+	// Set uppercase first because environment variable names are
	52	+	// case-insensitive on Windows.
	53	+	t.Setenv("HTTP_PROXY", "")
	54	+	t.Setenv("HTTPS_PROXY", "")
	55	+	t.Setenv("http_proxy", "http://http-env.example:8080")
	56	+	t.Setenv("https_proxy", "http://https-env.example:8080")
	57	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	58	+	t.Fatal(err)
	59	+	}
	60	+
	61	+	applyConfiguredHTTPProxy()
	62	+
	63	+	if got := getenvForTest("http_proxy"); got != "http://http-env.example:8080" {
	64	+	t.Fatalf("http_proxy = %q", got)
	65	+	}
	66	+	if got := getenvForTest("https_proxy"); got != "http://https-env.example:8080" {
	67	+	t.Fatalf("https_proxy = %q", got)
	68	+	}
	69	+	}
	70	+
	71	+	func TestApplyConfiguredHTTPProxyIgnoresEmptySetting(t *testing.T) {
	72	+	t.Setenv("ZUT_HOME", t.TempDir())
	73	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	74	+	t.Setenv(key, "")
	75	+	}
	76	+	if err := SaveConfig(Config{HTTPProxy: " "}); err != nil {
	77	+	t.Fatal(err)
	78	+	}
	79	+
	80	+	applyConfiguredHTTPProxy()
	81	+
	82	+	if got := getenvForTest("HTTP_PROXY"); got != "" {
	83	+	t.Fatalf("HTTP_PROXY = %q", got)
	84	+	}
	85	+	if got := getenvForTest("HTTPS_PROXY"); got != "" {
	86	+	t.Fatalf("HTTPS_PROXY = %q", got)
	87	+	}
	88	+	}
	89	+
	90	+	func getenvForTest(key string) string {
	91	+	return os.Getenv(key)
	92	+	}

A packages/agent/prompt_input_test.go +73 -0

@@ -0,0 +1,73 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"os"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestArgsForStdinSelectsPrintModeForPipe(t *testing.T) {
	9	+	args, err := ParseArgs(argsForStdin([]string{"summarize"}, os.ModeNamedPipe))
	10	+	if err != nil {
	11	+	t.Fatal(err)
	12	+	}
	13	+	if args.Mode != ModePrint \|\| args.Prompt != "summarize" {
	14	+	t.Fatalf("Mode=%q Prompt=%q, want print mode with prompt", args.Mode, args.Prompt)
	15	+	}
	16	+	}
	17	+
	18	+	func TestArgsForStdinPreservesTerminalAndExplicitModes(t *testing.T) {
	19	+	tests := []struct {
	20	+	name string
	21	+	in []string
	22	+	mode os.FileMode
	23	+	want Mode
	24	+	}{
	25	+	{name: "terminal stays interactive", mode: os.ModeCharDevice, want: ModeInteractive},
	26	+	{name: "explicit stream overrides pipe default", in: []string{"--stream", "prompt"}, mode: os.ModeNamedPipe, want: ModeStream},
	27	+	{name: "explicit json overrides redirected file default", in: []string{"--json", "prompt"}, want: ModeJSON},
	28	+	{name: "explicit rpc overrides pipe default", in: []string{"--rpc"}, mode: os.ModeNamedPipe, want: ModeRPC},
	29	+	}
	30	+	for _, tc := range tests {
	31	+	t.Run(tc.name, func(t *testing.T) {
	32	+	args, err := ParseArgs(argsForStdin(tc.in, tc.mode))
	33	+	if err != nil {
	34	+	t.Fatal(err)
	35	+	}
	36	+	if args.Mode != tc.want {
	37	+	t.Fatalf("Mode=%q, want %q", args.Mode, tc.want)
	38	+	}
	39	+	})
	40	+	}
	41	+	}
	42	+
	43	+	func TestArgsForStdinAllowsPrintStats(t *testing.T) {
	44	+	args, err := ParseArgs(argsForStdin([]string{"--stats", "stats.json"}, os.ModeNamedPipe))
	45	+	if err != nil {
	46	+	t.Fatal(err)
	47	+	}
	48	+	if args.Mode != ModePrint \|\| args.StatsPath != "stats.json" {
	49	+	t.Fatalf("Mode=%q StatsPath=%q, want print mode with stats path", args.Mode, args.StatsPath)
	50	+	}
	51	+	}
	52	+
	53	+	func TestCombinePromptInput(t *testing.T) {
	54	+	tests := []struct {
	55	+	name string
	56	+	stdin string
	57	+	positional string
	58	+	want string
	59	+	}{
	60	+	{name: "stdin and instruction", stdin: "input text\n", positional: "summarize it", want: "input text\nsummarize it"},
	61	+	{name: "stdin only", stdin: " input text\n", want: "input text"},
	62	+	{name: "positional only", positional: " summarize it ", want: "summarize it"},
	63	+	{name: "empty", stdin: " \n", positional: " ", want: ""},
	64	+	}
	65	+
	66	+	for _, tc := range tests {
	67	+	t.Run(tc.name, func(t *testing.T) {
	68	+	if got := combinePromptInput(tc.stdin, tc.positional); got != tc.want {
	69	+	t.Fatalf("combinePromptInput(%q, %q) = %q, want %q", tc.stdin, tc.positional, got, tc.want)
	70	+	}
	71	+	})
	72	+	}
	73	+	}

@@ -0,0 +1,73 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"os"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestArgsForStdinSelectsPrintModeForPipe(t *testing.T) {
	9	+	args, err := ParseArgs(argsForStdin([]string{"summarize"}, os.ModeNamedPipe))
	10	+	if err != nil {
	11	+	t.Fatal(err)
	12	+	}
	13	+	if args.Mode != ModePrint \|\| args.Prompt != "summarize" {
	14	+	t.Fatalf("Mode=%q Prompt=%q, want print mode with prompt", args.Mode, args.Prompt)
	15	+	}
	16	+	}
	17	+
	18	+	func TestArgsForStdinPreservesTerminalAndExplicitModes(t *testing.T) {
	19	+	tests := []struct {
	20	+	name string
	21	+	in []string
	22	+	mode os.FileMode
	23	+	want Mode
	24	+	}{
	25	+	{name: "terminal stays interactive", mode: os.ModeCharDevice, want: ModeInteractive},
	26	+	{name: "explicit stream overrides pipe default", in: []string{"--stream", "prompt"}, mode: os.ModeNamedPipe, want: ModeStream},
	27	+	{name: "explicit json overrides redirected file default", in: []string{"--json", "prompt"}, want: ModeJSON},
	28	+	{name: "explicit rpc overrides pipe default", in: []string{"--rpc"}, mode: os.ModeNamedPipe, want: ModeRPC},
	29	+	}
	30	+	for _, tc := range tests {
	31	+	t.Run(tc.name, func(t *testing.T) {
	32	+	args, err := ParseArgs(argsForStdin(tc.in, tc.mode))
	33	+	if err != nil {
	34	+	t.Fatal(err)
	35	+	}
	36	+	if args.Mode != tc.want {
	37	+	t.Fatalf("Mode=%q, want %q", args.Mode, tc.want)
	38	+	}
	39	+	})
	40	+	}
	41	+	}
	42	+
	43	+	func TestArgsForStdinAllowsPrintStats(t *testing.T) {
	44	+	args, err := ParseArgs(argsForStdin([]string{"--stats", "stats.json"}, os.ModeNamedPipe))
	45	+	if err != nil {
	46	+	t.Fatal(err)
	47	+	}
	48	+	if args.Mode != ModePrint \|\| args.StatsPath != "stats.json" {
	49	+	t.Fatalf("Mode=%q StatsPath=%q, want print mode with stats path", args.Mode, args.StatsPath)
	50	+	}
	51	+	}
	52	+
	53	+	func TestCombinePromptInput(t *testing.T) {
	54	+	tests := []struct {
	55	+	name string
	56	+	stdin string
	57	+	positional string
	58	+	want string
	59	+	}{
	60	+	{name: "stdin and instruction", stdin: "input text\n", positional: "summarize it", want: "input text\nsummarize it"},
	61	+	{name: "stdin only", stdin: " input text\n", want: "input text"},
	62	+	{name: "positional only", positional: " summarize it ", want: "summarize it"},
	63	+	{name: "empty", stdin: " \n", positional: " ", want: ""},
	64	+	}
	65	+
	66	+	for _, tc := range tests {
	67	+	t.Run(tc.name, func(t *testing.T) {
	68	+	if got := combinePromptInput(tc.stdin, tc.positional); got != tc.want {
	69	+	t.Fatalf("combinePromptInput(%q, %q) = %q, want %q", tc.stdin, tc.positional, got, tc.want)
	70	+	}
	71	+	})
	72	+	}
	73	+	}

M packages/agent/systemprompt.go +2 -2


@@ -21,7 +21,7 @@ type ToolSummary struct {
21	21		type SystemPromptOpts struct {
22	22		CWD string
23	23		Tools []ToolSummary
24		-	Custom string // if set, replaces the default identity entirely
	24	+	Custom string // if set, replaces the built-in identity and docs guidance
25	25		Append []string // extra text appended at the end
26	26		Now time.Time
27	27		ZutDocsDir string

@@ -67,7 +67,7 @@ func BuildSystemPrompt(o SystemPromptOpts) string {
67	67		sb.WriteString("\n\n")
68	68		sb.WriteString(compactedSummaryHandoffInstruction)
69	69
70		-	if strings.TrimSpace(o.ZutDocsDir) != "" {
	70	+	if o.Custom == "" && strings.TrimSpace(o.ZutDocsDir) != "" {
71	71		sb.WriteString("\n\nZut's own docs are installed under ")
72	72		sb.WriteString(o.ZutDocsDir)
73	73		sb.WriteString("; use the read tool there when you need details about zut RPC, extensions, skills, or built-in behaviour.")


@@ -21,7 +21,7 @@ type ToolSummary struct {
21		type SystemPromptOpts struct {	21		type SystemPromptOpts struct {
22		CWD string	22		CWD string
23		Tools []ToolSummary	23		Tools []ToolSummary
24	-	Custom string // if set, replaces the default identity entirely	24	+	Custom string // if set, replaces the built-in identity and docs guidance
25		Append []string // extra text appended at the end	25		Append []string // extra text appended at the end
26		Now time.Time	26		Now time.Time
27		ZutDocsDir string	27		ZutDocsDir string

@@ -67,7 +67,7 @@ func BuildSystemPrompt(o SystemPromptOpts) string {
67		sb.WriteString("\n\n")	67		sb.WriteString("\n\n")
68		sb.WriteString(compactedSummaryHandoffInstruction)	68		sb.WriteString(compactedSummaryHandoffInstruction)
69			69
70	-	if strings.TrimSpace(o.ZutDocsDir) != "" {	70	+	if o.Custom == "" && strings.TrimSpace(o.ZutDocsDir) != "" {
71		sb.WriteString("\n\nZut's own docs are installed under ")	71		sb.WriteString("\n\nZut's own docs are installed under ")
72		sb.WriteString(o.ZutDocsDir)	72		sb.WriteString(o.ZutDocsDir)
73		sb.WriteString("; use the read tool there when you need details about zut RPC, extensions, skills, or built-in behaviour.")	73		sb.WriteString("; use the read tool there when you need details about zut RPC, extensions, skills, or built-in behaviour.")

M packages/agent/systemprompt_test.go +37 -0


@@ -3,6 +3,7 @@ package agent
3	3		import (
4	4		"strings"
5	5		"testing"
	6	+	"time"
6	7		)
7	8
8	9		func TestBuildSystemPromptAddsCompactionHandoffToCustomPrompt(t *testing.T) {

@@ -22,3 +23,39 @@ func TestBuildSystemPromptAddsCompactionHandoffToCustomPrompt(t *testing.T) {
22	23		}
23	24		}
24	25		}
	26	+
	27	+	func TestBuildSystemPromptCustomOmitsBuiltInDocs(t *testing.T) {
	28	+	got := BuildSystemPrompt(SystemPromptOpts{
	29	+	CWD: "/workspace",
	30	+	Custom: "Custom instructions",
	31	+	Append: []string{"Additional context"},
	32	+	Now: time.Date(2026, time.August, 6, 0, 0, 0, 0, time.UTC),
	33	+	ZutDocsDir: "/zut/docs",
	34	+	})
	35	+
	36	+	if strings.Contains(got, "Zut's own docs") \|\| strings.Contains(got, "/zut/docs") {
	37	+	t.Fatalf("custom prompt includes built-in docs guidance:\n%s", got)
	38	+	}
	39	+	for _, want := range []string{
	40	+	"Custom instructions",
	41	+	"Additional context",
	42	+	"Current date: 2026-08-06",
	43	+	"Current working directory: /workspace",
	44	+	} {
	45	+	if !strings.Contains(got, want) {
	46	+	t.Errorf("custom prompt missing %q:\n%s", want, got)
	47	+	}
	48	+	}
	49	+	}
	50	+
	51	+	func TestBuildSystemPromptDefaultIncludesBuiltInDocs(t *testing.T) {
	52	+	got := BuildSystemPrompt(SystemPromptOpts{
	53	+	CWD: "/workspace",
	54	+	Now: time.Date(2026, time.August, 6, 0, 0, 0, 0, time.UTC),
	55	+	ZutDocsDir: "/zut/docs",
	56	+	})
	57	+
	58	+	if !strings.Contains(got, "Zut's own docs are installed under /zut/docs") {
	59	+	t.Fatalf("default prompt missing built-in docs guidance:\n%s", got)
	60	+	}
	61	+	}


@@ -3,6 +3,7 @@ package agent
3	import (	3		import (
4	"strings"	4		"strings"
5	"testing"	5		"testing"
		6	+	"time"
6	)	7		)
7		8
8	func TestBuildSystemPromptAddsCompactionHandoffToCustomPrompt(t *testing.T) {	9		func TestBuildSystemPromptAddsCompactionHandoffToCustomPrompt(t *testing.T) {

@@ -22,3 +23,39 @@ func TestBuildSystemPromptAddsCompactionHandoffToCustomPrompt(t *testing.T) {
22	}	23		}
23	}	24		}
24	}	25		}
		26	+
		27	+	func TestBuildSystemPromptCustomOmitsBuiltInDocs(t *testing.T) {
		28	+	got := BuildSystemPrompt(SystemPromptOpts{
		29	+	CWD: "/workspace",
		30	+	Custom: "Custom instructions",
		31	+	Append: []string{"Additional context"},
		32	+	Now: time.Date(2026, time.August, 6, 0, 0, 0, 0, time.UTC),
		33	+	ZutDocsDir: "/zut/docs",
		34	+	})
		35	+
		36	+	if strings.Contains(got, "Zut's own docs") \|\| strings.Contains(got, "/zut/docs") {
		37	+	t.Fatalf("custom prompt includes built-in docs guidance:\n%s", got)
		38	+	}
		39	+	for _, want := range []string{
		40	+	"Custom instructions",
		41	+	"Additional context",
		42	+	"Current date: 2026-08-06",
		43	+	"Current working directory: /workspace",
		44	+	} {
		45	+	if !strings.Contains(got, want) {
		46	+	t.Errorf("custom prompt missing %q:\n%s", want, got)
		47	+	}
		48	+	}
		49	+	}
		50	+
		51	+	func TestBuildSystemPromptDefaultIncludesBuiltInDocs(t *testing.T) {
		52	+	got := BuildSystemPrompt(SystemPromptOpts{
		53	+	CWD: "/workspace",
		54	+	Now: time.Date(2026, time.August, 6, 0, 0, 0, 0, time.UTC),
		55	+	ZutDocsDir: "/zut/docs",
		56	+	})
		57	+
		58	+	if !strings.Contains(got, "Zut's own docs are installed under /zut/docs") {
		59	+	t.Fatalf("default prompt missing built-in docs guidance:\n%s", got)
		60	+	}
		61	+	}

M packages/agent/update.go +22 -30


@@ -7,9 +7,10 @@ import (
7	7		"net/http"
8	8		"os"
9	9		"path/filepath"
10		-	"strconv"
11	10		"strings"
12	11		"time"
	12	+
	13	+	"golang.org/x/mod/semver"
13	14		)
14	15
15	16		// updateCheckTTL is how often we hit the GitHub API to look for a new

@@ -132,40 +133,31 @@ func buildInfo(current, latest, url string) UpdateInfo {
132	133		return info
133	134		}
134	135
135		-	// versionLess returns a < b for dotted semver-ish tags like "0.0.4".
136		-	// Non-numeric components compare as zero, which is fine for zut's
137		-	// x.y.z-only scheme.
	136	+	// versionLess reports whether a is an older semantic version than b.
	137	+	// Invalid versions return false so update checks never offer a downgrade.
138	138		func versionLess(a, b string) bool {
139		-	as := splitVersion(a)
140		-	bs := splitVersion(b)
141		-	for i := 0; i < 3; i++ {
142		-	av, bv := 0, 0
143		-	if i < len(as) {
144		-	av = as[i]
145		-	}
146		-	if i < len(bs) {
147		-	bv = bs[i]
148		-	}
149		-	if av != bv {
150		-	return av < bv
151		-	}
152		-	}
153		-	return false
	139	+	comparison, err := compareVersions(a, b)
	140	+	return err == nil && comparison < 0
154	141		}
155	142
156		-	func splitVersion(s string) []int {
157		-	s = strings.TrimPrefix(s, "v")
158		-	// Strip build-metadata suffix like "(abc1234, 2026-04-18)".
159		-	if i := strings.IndexAny(s, " ("); i > 0 {
160		-	s = s[:i]
	143	+	func compareVersions(a, b string) (int, error) {
	144	+	a = normalizedVersion(a)
	145	+	b = normalizedVersion(b)
	146	+	if !semver.IsValid(a) {
	147	+	return 0, fmt.Errorf("invalid semantic version %q", strings.TrimPrefix(a, "v"))
	148	+	}
	149	+	if !semver.IsValid(b) {
	150	+	return 0, fmt.Errorf("invalid semantic version %q", strings.TrimPrefix(b, "v"))
161	151		}
162		-	parts := strings.Split(s, ".")
163		-	out := make([]int, 0, len(parts))
164		-	for _, p := range parts {
165		-	n, _ := strconv.Atoi(p)
166		-	out = append(out, n)
	152	+	return semver.Compare(a, b), nil
	153	+	}
	154	+
	155	+	func normalizedVersion(v string) string {
	156	+	v = versionOnly(v)
	157	+	if !strings.HasPrefix(v, "v") {
	158	+	v = "v" + v
167	159		}
168		-	return out
	160	+	return v
169	161		}
170	162
171	163		// fetchLatestRelease queries the GitHub API for the latest published


@@ -7,9 +7,10 @@ import (
7		"net/http"	7		"net/http"
8		"os"	8		"os"
9		"path/filepath"	9		"path/filepath"
10	-	"strconv"
11		"strings"	10		"strings"
12		"time"	11		"time"
			12	+
			13	+	"golang.org/x/mod/semver"
13		)	14		)
14			15
15		// updateCheckTTL is how often we hit the GitHub API to look for a new	16		// updateCheckTTL is how often we hit the GitHub API to look for a new

@@ -132,40 +133,31 @@ func buildInfo(current, latest, url string) UpdateInfo {
132		return info	133		return info
133		}	134		}
134			135
135	-	// versionLess returns a < b for dotted semver-ish tags like "0.0.4".	136	+	// versionLess reports whether a is an older semantic version than b.
136	-	// Non-numeric components compare as zero, which is fine for zut's	137	+	// Invalid versions return false so update checks never offer a downgrade.
137	-	// x.y.z-only scheme.
138		func versionLess(a, b string) bool {	138		func versionLess(a, b string) bool {
139	-	as := splitVersion(a)	139	+	comparison, err := compareVersions(a, b)
140	-	bs := splitVersion(b)	140	+	return err == nil && comparison < 0
141	-	for i := 0; i < 3; i++ {
142	-	av, bv := 0, 0
143	-	if i < len(as) {
144	-	av = as[i]
145	-	}
146	-	if i < len(bs) {
147	-	bv = bs[i]
148	-	}
149	-	if av != bv {
150	-	return av < bv
151	-	}
152	-	}
153	-	return false
154		}	141		}
155			142
156	-	func splitVersion(s string) []int {	143	+	func compareVersions(a, b string) (int, error) {
157	-	s = strings.TrimPrefix(s, "v")	144	+	a = normalizedVersion(a)
158	-	// Strip build-metadata suffix like "(abc1234, 2026-04-18)".	145	+	b = normalizedVersion(b)
159	-	if i := strings.IndexAny(s, " ("); i > 0 {	146	+	if !semver.IsValid(a) {
160	-	s = s[:i]	147	+	return 0, fmt.Errorf("invalid semantic version %q", strings.TrimPrefix(a, "v"))
			148	+	}
			149	+	if !semver.IsValid(b) {
			150	+	return 0, fmt.Errorf("invalid semantic version %q", strings.TrimPrefix(b, "v"))
161		}	151		}
162	-	parts := strings.Split(s, ".")	152	+	return semver.Compare(a, b), nil
163	-	out := make([]int, 0, len(parts))	153	+	}
164	-	for _, p := range parts {	154	+
165	-	n, _ := strconv.Atoi(p)	155	+	func normalizedVersion(v string) string {
166	-	out = append(out, n)	156	+	v = versionOnly(v)
			157	+	if !strings.HasPrefix(v, "v") {
			158	+	v = "v" + v
167		}	159		}
168	-	return out	160	+	return v
169		}	161		}
170			162
171		// fetchLatestRelease queries the GitHub API for the latest published	163		// fetchLatestRelease queries the GitHub API for the latest published

M packages/agent/update_test.go +32 -0


@@ -52,3 +52,35 @@ func TestCheckForUpdateSkipsDevVersionWithoutCacheOrNetwork(t *testing.T) {
52	52		t.Fatalf("development check made %d network requests, want 0", transport.calls)
53	53		}
54	54		}
	55	+
	56	+	func TestVersionLess(t *testing.T) {
	57	+	pseudo := "0.3.35-0.20260809165004-9e3c28d4b65"
	58	+	cases := []struct {
	59	+	name string
	60	+	a, b string
	61	+	want bool
	62	+	}{
	63	+	{name: "older release", a: "0.3.33", b: "0.3.34", want: true},
	64	+	{name: "newer release", a: "0.3.35", b: "0.3.34", want: false},
	65	+	{name: "pseudo after prior release", a: pseudo, b: "0.3.34", want: false},
	66	+	{name: "pseudo before final release", a: pseudo, b: "0.3.35", want: true},
	67	+	{name: "prior release before pseudo", a: "0.3.34", b: pseudo, want: true},
	68	+	{name: "decorated build", a: "0.3.33 (abc1234, 2026-08-09)", b: "0.3.34", want: true},
	69	+	{name: "invalid current", a: "not-a-version", b: "0.3.34", want: false},
	70	+	}
	71	+	for _, tc := range cases {
	72	+	t.Run(tc.name, func(t *testing.T) {
	73	+	if got := versionLess(tc.a, tc.b); got != tc.want {
	74	+	t.Fatalf("versionLess(%q, %q) = %v, want %v", tc.a, tc.b, got, tc.want)
	75	+	}
	76	+	})
	77	+	}
	78	+	}
	79	+
	80	+	func TestBuildInfoDoesNotOfferPseudoVersionDowngrade(t *testing.T) {
	81	+	current := "0.3.35-0.20260809165004-9e3c28d4b65"
	82	+	info := buildInfo(current, "v0.3.34", "https://example.com/release")
	83	+	if info.Available {
	84	+	t.Fatalf("buildInfo(%q, %q) offered a downgrade", current, info.Latest)
	85	+	}
	86	+	}


@@ -52,3 +52,35 @@ func TestCheckForUpdateSkipsDevVersionWithoutCacheOrNetwork(t *testing.T) {
52	t.Fatalf("development check made %d network requests, want 0", transport.calls)	52		t.Fatalf("development check made %d network requests, want 0", transport.calls)
53	}	53		}
54	}	54		}
		55	+
		56	+	func TestVersionLess(t *testing.T) {
		57	+	pseudo := "0.3.35-0.20260809165004-9e3c28d4b65"
		58	+	cases := []struct {
		59	+	name string
		60	+	a, b string
		61	+	want bool
		62	+	}{
		63	+	{name: "older release", a: "0.3.33", b: "0.3.34", want: true},
		64	+	{name: "newer release", a: "0.3.35", b: "0.3.34", want: false},
		65	+	{name: "pseudo after prior release", a: pseudo, b: "0.3.34", want: false},
		66	+	{name: "pseudo before final release", a: pseudo, b: "0.3.35", want: true},
		67	+	{name: "prior release before pseudo", a: "0.3.34", b: pseudo, want: true},
		68	+	{name: "decorated build", a: "0.3.33 (abc1234, 2026-08-09)", b: "0.3.34", want: true},
		69	+	{name: "invalid current", a: "not-a-version", b: "0.3.34", want: false},
		70	+	}
		71	+	for _, tc := range cases {
		72	+	t.Run(tc.name, func(t *testing.T) {
		73	+	if got := versionLess(tc.a, tc.b); got != tc.want {
		74	+	t.Fatalf("versionLess(%q, %q) = %v, want %v", tc.a, tc.b, got, tc.want)
		75	+	}
		76	+	})
		77	+	}
		78	+	}
		79	+
		80	+	func TestBuildInfoDoesNotOfferPseudoVersionDowngrade(t *testing.T) {
		81	+	current := "0.3.35-0.20260809165004-9e3c28d4b65"
		82	+	info := buildInfo(current, "v0.3.34", "https://example.com/release")
		83	+	if info.Available {
		84	+	t.Fatalf("buildInfo(%q, %q) offered a downgrade", current, info.Latest)
		85	+	}
		86	+	}

M packages/agent/zutfile.go +5 -1


@@ -765,7 +765,11 @@ func checkZutfileRequirements(zf zutfileLoaded, version string) error {
765	765		if isDevVersion(version) {
766	766		return fmt.Errorf("agent requires zut %s or newer; unversioned development builds cannot satisfy min_zut", min)
767	767		}
768		-	if versionLess(version, min) {
	768	+	comparison, err := compareVersions(version, min)
	769	+	if err != nil {
	770	+	return fmt.Errorf("validate min_zut requirement: %w", err)
	771	+	}
	772	+	if comparison < 0 {
769	773		return fmt.Errorf("agent requires zut %s or newer; running %s", min, versionOnly(version))
770	774		}
771	775		}

 		if isDevVersion(version) {
 			return fmt.Errorf("agent requires zut %s or newer; unversioned development builds cannot satisfy min_zut", min)
 		}
-		if versionLess(version, min) {
+		comparison, err := compareVersions(version, min)
+		if err != nil {
+			return fmt.Errorf("validate min_zut requirement: %w", err)
+		}
+		if comparison < 0 {
 			return fmt.Errorf("agent requires zut %s or newer; running %s", min, versionOnly(version))
 		}
 	}

M packages/agent/zutfile_test.go +8 -0


@@ -246,6 +246,14 @@ func TestCheckZutfileMinVersion(t *testing.T) {
246	246		if err := checkZutfileRequirements(zf, "0.3.0"); err != nil {
247	247		t.Fatalf("minimum version rejected: %v", err)
248	248		}
	249	+	if err := checkZutfileRequirements(zf, "not-a-version"); err == nil {
	250	+	t.Fatal("invalid zut version accepted")
	251	+	}
	252	+
	253	+	zf.Manifest.Runtime.MinZut = "not-a-version"
	254	+	if err := checkZutfileRequirements(zf, "0.3.0"); err == nil {
	255	+	t.Fatal("invalid min_zut requirement accepted")
	256	+	}
249	257		}
250	258
251	259		func TestApplyZutfileModelRequirementsRejectsUnsupportedFields(t *testing.T) {


@@ -246,6 +246,14 @@ func TestCheckZutfileMinVersion(t *testing.T) {
246	if err := checkZutfileRequirements(zf, "0.3.0"); err != nil {	246		if err := checkZutfileRequirements(zf, "0.3.0"); err != nil {
247	t.Fatalf("minimum version rejected: %v", err)	247		t.Fatalf("minimum version rejected: %v", err)
248	}	248		}
		249	+	if err := checkZutfileRequirements(zf, "not-a-version"); err == nil {
		250	+	t.Fatal("invalid zut version accepted")
		251	+	}
		252	+
		253	+	zf.Manifest.Runtime.MinZut = "not-a-version"
		254	+	if err := checkZutfileRequirements(zf, "0.3.0"); err == nil {
		255	+	t.Fatal("invalid min_zut requirement accepted")
		256	+	}
249	}	257		}
250		258
251	func TestApplyZutfileModelRequirementsRejectsUnsupportedFields(t *testing.T) {	259		func TestApplyZutfileModelRequirementsRejectsUnsupportedFields(t *testing.T) {

M packages/provider/catalog_builtin.go +1 -0


@@ -308,6 +308,7 @@ var builtinCatalog = []Model{
308	308		{Provider: "opencode-go", ID: "glm-5", DisplayName: "GLM-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},
309	309		{Provider: "opencode-go", ID: "glm-5.1", DisplayName: "GLM-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},
310	310		{Provider: "opencode-go", ID: "glm-5.2", DisplayName: "GLM-5.2", ContextWindow: 1024000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.91, PriceOutput: 2.86, PriceCacheRead: 0.169, BaseURL: "https://opencode.ai/zen/go/v1"},
	311	+	{Provider: "opencode-go", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses, ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, PriceCacheWrite: 0.25, PriceTierInputTokens: 272000, PriceInputAbove: 0.4, PriceOutputAbove: 1.8, PriceCacheReadAbove: 0.04, PriceCacheWriteAbove: 0.5, BaseURL: "https://opencode.ai/zen/go/v1"},
311	312		{Provider: "opencode-go", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 500000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen/go/v1"},
312	313		{Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "Kimi K2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},
313	314		{Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "Kimi K2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},


@@ -308,6 +308,7 @@ var builtinCatalog = []Model{
308	{Provider: "opencode-go", ID: "glm-5", DisplayName: "GLM-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},	308		{Provider: "opencode-go", ID: "glm-5", DisplayName: "GLM-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},
309	{Provider: "opencode-go", ID: "glm-5.1", DisplayName: "GLM-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},	309		{Provider: "opencode-go", ID: "glm-5.1", DisplayName: "GLM-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},
310	{Provider: "opencode-go", ID: "glm-5.2", DisplayName: "GLM-5.2", ContextWindow: 1024000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.91, PriceOutput: 2.86, PriceCacheRead: 0.169, BaseURL: "https://opencode.ai/zen/go/v1"},	310		{Provider: "opencode-go", ID: "glm-5.2", DisplayName: "GLM-5.2", ContextWindow: 1024000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.91, PriceOutput: 2.86, PriceCacheRead: 0.169, BaseURL: "https://opencode.ai/zen/go/v1"},
		311	+	{Provider: "opencode-go", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses, ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, PriceCacheWrite: 0.25, PriceTierInputTokens: 272000, PriceInputAbove: 0.4, PriceOutputAbove: 1.8, PriceCacheReadAbove: 0.04, PriceCacheWriteAbove: 0.5, BaseURL: "https://opencode.ai/zen/go/v1"},
311	{Provider: "opencode-go", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 500000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen/go/v1"},	312		{Provider: "opencode-go", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 500000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen/go/v1"},
312	{Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "Kimi K2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},	313		{Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "Kimi K2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},
313	{Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "Kimi K2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},	314		{Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "Kimi K2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},

M packages/provider/extra_providers.go +7 -2


@@ -124,9 +124,14 @@ func NewOpenCode(apiKey, baseURL string) Client {
124	124		return NewOpenAICompat("opencode", apiKey, baseURL, "https://opencode.ai/zen/v1")
125	125		}
126	126
127		-	// NewOpenCodeGo is the opencode-go variant.
	127	+	// NewOpenCodeGo routes each model through the wire API exposed by OpenCode Go.
128	128		func NewOpenCodeGo(apiKey, baseURL string) Client {
129		-	return NewOpenAICompat("opencode-go", apiKey, baseURL, "https://opencode.ai/zen/go/v1")
	129	+	const defaultBaseURL = "https://opencode.ai/zen/go/v1"
	130	+	return NewModelRouter("opencode-go",
	131	+	NewOpenAICompat("opencode-go", apiKey, baseURL, defaultBaseURL),
	132	+	map[string]Client{
	133	+	APIResponses: NewOpenAIResponsesNamed(apiKey, firstNonEmptyString(baseURL, defaultBaseURL), "opencode-go"),
	134	+	})
130	135		}
131	136
132	137		// NewMinimaxOpenAI is the OpenAI-completions flavor of MiniMax, in case

 	return NewOpenAICompat("opencode", apiKey, baseURL, "https://opencode.ai/zen/v1")
 }
-// NewOpenCodeGo is the opencode-go variant.
+// NewOpenCodeGo routes each model through the wire API exposed by OpenCode Go.
 func NewOpenCodeGo(apiKey, baseURL string) Client {
-	return NewOpenAICompat("opencode-go", apiKey, baseURL, "https://opencode.ai/zen/go/v1")
+	const defaultBaseURL = "https://opencode.ai/zen/go/v1"
+	return NewModelRouter("opencode-go",
+		NewOpenAICompat("opencode-go", apiKey, baseURL, defaultBaseURL),
+		map[string]Client{
+			APIResponses: NewOpenAIResponsesNamed(apiKey, firstNonEmptyString(baseURL, defaultBaseURL), "opencode-go"),
+		})
 }
 // NewMinimaxOpenAI is the OpenAI-completions flavor of MiniMax, in case

M packages/provider/llamacpp.go +16 -2


@@ -47,6 +47,10 @@ type LlamaCPPModel struct {
47	47		Size int64 `json:"size,omitempty"`
48	48		FileType string `json:"ftype,omitempty"`
49	49		} `json:"meta,omitempty"`
	50	+	// Progress is emitted at the model's top level by current llama.cpp
	51	+	// routers. Status.Progress supports versions that follow the documented
	52	+	// nested shape.
	53	+	Progress map[string]LlamaCPPBytes `json:"progress,omitempty"`
50	54		}
51	55
52	56		// LlamaCPPBytes describes byte progress for one downloaded file.

@@ -204,7 +208,13 @@ func (c *LlamaCPPClient) watch(ctx context.Context, events chan<- llamaCPPEvent)
204	208		if c.APIKey != "" {
205	209		req.Header.Set("authorization", "Bearer "+c.APIKey)
206	210		}
207		-	resp, err := c.HTTP.Do(req)
	211	+	// http.Client.Timeout covers the complete response body lifetime, which is
	212	+	// appropriate for ordinary management requests but would terminate this
	213	+	// long-lived SSE stream. Preserve the configured transport and other client
	214	+	// behavior while letting ctx control the stream lifetime.
	215	+	streamClient := *c.HTTP
	216	+	streamClient.Timeout = 0
	217	+	resp, err := streamClient.Do(req)
208	218		if err != nil {
209	219		return err
210	220		}

@@ -408,7 +418,11 @@ func (c *LlamaCPPClient) DownloadAndWait(ctx context.Context, model string, upda
408	418		}
409	419		if entry.Status.Value == "downloading" {
410	420		sawDownload = true
411		-	if progress, ok := downloadProgress(entry.Status.Progress); ok {
	421	+	files := entry.Status.Progress
	422	+	if len(files) == 0 {
	423	+	files = entry.Progress
	424	+	}
	425	+	if progress, ok := downloadProgress(files); ok {
412	426		update(progress)
413	427		}
414	428		} else if sawDownload \|\| polls >= 2 {

 		Size         int64  `json:"size,omitempty"`
 		FileType     string `json:"ftype,omitempty"`
 	} `json:"meta,omitempty"`
+	// Progress is emitted at the model's top level by current llama.cpp
+	// routers. Status.Progress supports versions that follow the documented
+	// nested shape.
+	Progress map[string]LlamaCPPBytes `json:"progress,omitempty"`
 }
 // LlamaCPPBytes describes byte progress for one downloaded file.
 	if c.APIKey != "" {
 		req.Header.Set("authorization", "Bearer "+c.APIKey)
 	}
-	resp, err := c.HTTP.Do(req)
+	// http.Client.Timeout covers the complete response body lifetime, which is
+	// appropriate for ordinary management requests but would terminate this
+	// long-lived SSE stream. Preserve the configured transport and other client
+	// behavior while letting ctx control the stream lifetime.
+	streamClient := *c.HTTP
+	streamClient.Timeout = 0
+	resp, err := streamClient.Do(req)
 	if err != nil {
 		return err
 	}
 				}
 				if entry.Status.Value == "downloading" {
 					sawDownload = true
-					if progress, ok := downloadProgress(entry.Status.Progress); ok {
+					files := entry.Status.Progress
+					if len(files) == 0 {
+						files = entry.Progress
+					}
+					if progress, ok := downloadProgress(files); ok {
 						update(progress)
 					}
 				} else if sawDownload || polls >= 2 {

M packages/provider/llamacpp_test.go +56 -0


@@ -32,6 +32,62 @@ func TestLlamaCPPLoadProgressStages(t *testing.T) {
32	32		}
33	33		}
34	34
	35	+	func TestLlamaCPPWatchOutlivesRequestTimeout(t *testing.T) {
	36	+	requestStarted := make(chan struct{})
	37	+	releaseEvent := make(chan struct{})
	38	+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	39	+	w.Header().Set("content-type", "text/event-stream")
	40	+	w.WriteHeader(http.StatusOK)
	41	+	w.(http.Flusher).Flush()
	42	+	close(requestStarted)
	43	+	select {
	44	+	case <-releaseEvent:
	45	+	fmt.Fprint(w, "data: {\"model\":\"test\",\"event\":\"download_progress\",\"data\":{}}\n\n")
	46	+	w.(http.Flusher).Flush()
	47	+	case <-r.Context().Done():
	48	+	}
	49	+	}))
	50	+	defer server.Close()
	51	+
	52	+	client, err := NewLlamaCPPClient(server.URL, "")
	53	+	if err != nil {
	54	+	t.Fatal(err)
	55	+	}
	56	+	client.HTTP.Timeout = 20 * time.Millisecond
	57	+	ctx, cancel := context.WithCancel(context.Background())
	58	+	defer cancel()
	59	+	events := make(chan llamaCPPEvent, 1)
	60	+	watchDone := make(chan error, 1)
	61	+	go func() { watchDone <- client.watch(ctx, events) }()
	62	+	<-requestStarted
	63	+
	64	+	select {
	65	+	case err := <-watchDone:
	66	+	t.Fatalf("watch stopped at ordinary request timeout: %v", err)
	67	+	case <-time.After(3 * client.HTTP.Timeout):
	68	+	}
	69	+	close(releaseEvent)
	70	+	select {
	71	+	case event := <-events:
	72	+	if event.Model != "test" \|\| event.Event != "download_progress" {
	73	+	t.Fatalf("event = %#v", event)
	74	+	}
	75	+	case <-time.After(time.Second):
	76	+	t.Fatal("timed out waiting for delayed SSE event")
	77	+	}
	78	+	}
	79	+
	80	+	func TestLlamaCPPModelAcceptsTopLevelDownloadProgress(t *testing.T) {
	81	+	var model LlamaCPPModel
	82	+	if err := json.Unmarshal([]byte(`{"id":"test","status":{"value":"downloading"},"progress":{"model.gguf":{"done":512,"total":1024}}}`), &model); err != nil {
	83	+	t.Fatal(err)
	84	+	}
	85	+	progress, ok := downloadProgress(model.Progress)
	86	+	if !ok \|\| progress.Ratio != 0.5 {
	87	+	t.Fatalf("progress = %#v, ok = %v", progress, ok)
	88	+	}
	89	+	}
	90	+
35	91		func TestLlamaCPPClientLoadAndDownloadProgress(t *testing.T) {
36	92		var mu sync.Mutex
37	93		status := "unloaded"


@@ -32,6 +32,62 @@ func TestLlamaCPPLoadProgressStages(t *testing.T) {
32	}	32		}
33	}	33		}
34		34
		35	+	func TestLlamaCPPWatchOutlivesRequestTimeout(t *testing.T) {
		36	+	requestStarted := make(chan struct{})
		37	+	releaseEvent := make(chan struct{})
		38	+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		39	+	w.Header().Set("content-type", "text/event-stream")
		40	+	w.WriteHeader(http.StatusOK)
		41	+	w.(http.Flusher).Flush()
		42	+	close(requestStarted)
		43	+	select {
		44	+	case <-releaseEvent:
		45	+	fmt.Fprint(w, "data: {\"model\":\"test\",\"event\":\"download_progress\",\"data\":{}}\n\n")
		46	+	w.(http.Flusher).Flush()
		47	+	case <-r.Context().Done():
		48	+	}
		49	+	}))
		50	+	defer server.Close()
		51	+
		52	+	client, err := NewLlamaCPPClient(server.URL, "")
		53	+	if err != nil {
		54	+	t.Fatal(err)
		55	+	}
		56	+	client.HTTP.Timeout = 20 * time.Millisecond
		57	+	ctx, cancel := context.WithCancel(context.Background())
		58	+	defer cancel()
		59	+	events := make(chan llamaCPPEvent, 1)
		60	+	watchDone := make(chan error, 1)
		61	+	go func() { watchDone <- client.watch(ctx, events) }()
		62	+	<-requestStarted
		63	+
		64	+	select {
		65	+	case err := <-watchDone:
		66	+	t.Fatalf("watch stopped at ordinary request timeout: %v", err)
		67	+	case <-time.After(3 * client.HTTP.Timeout):
		68	+	}
		69	+	close(releaseEvent)
		70	+	select {
		71	+	case event := <-events:
		72	+	if event.Model != "test" \|\| event.Event != "download_progress" {
		73	+	t.Fatalf("event = %#v", event)
		74	+	}
		75	+	case <-time.After(time.Second):
		76	+	t.Fatal("timed out waiting for delayed SSE event")
		77	+	}
		78	+	}
		79	+
		80	+	func TestLlamaCPPModelAcceptsTopLevelDownloadProgress(t *testing.T) {
		81	+	var model LlamaCPPModel
		82	+	if err := json.Unmarshal([]byte(`{"id":"test","status":{"value":"downloading"},"progress":{"model.gguf":{"done":512,"total":1024}}}`), &model); err != nil {
		83	+	t.Fatal(err)
		84	+	}
		85	+	progress, ok := downloadProgress(model.Progress)
		86	+	if !ok \|\| progress.Ratio != 0.5 {
		87	+	t.Fatalf("progress = %#v, ok = %v", progress, ok)
		88	+	}
		89	+	}
		90	+
35	func TestLlamaCPPClientLoadAndDownloadProgress(t *testing.T) {	91		func TestLlamaCPPClientLoadAndDownloadProgress(t *testing.T) {
36	var mu sync.Mutex	92		var mu sync.Mutex
37	status := "unloaded"	93		status := "unloaded"

M packages/provider/model_router_test.go +31 -0


@@ -48,3 +48,34 @@ func TestModelRouterRejectsMissingAPIClient(t *testing.T) {
48	48		t.Fatal("expected missing API client error")
49	49		}
50	50		}
	51	+
	52	+	func TestOpenCodeGoRoutesLunaToResponses(t *testing.T) {
	53	+	router := NewOpenCodeGo("token", "https://example.com/go/v1").(*modelRouter)
	54	+	if got := router.fallback.(*openaiClient).baseURL; got != "https://example.com/go/v1" {
	55	+	t.Fatalf("Completions base URL = %q", got)
	56	+	}
	57	+	responses := router.byAPI[APIResponses].(renamedClient).inner.(codexClient)
	58	+	if got := responses.baseURL; got != "https://example.com/go/v1/responses" {
	59	+	t.Fatalf("Responses base URL = %q", got)
	60	+	}
	61	+
	62	+	completionsCapture := &routeCaptureClient{name: "opencode-go"}
	63	+	responsesCapture := &routeCaptureClient{name: "opencode-go"}
	64	+	router.fallback = completionsCapture
	65	+	router.byAPI[APIResponses] = responsesCapture
	66	+	for _, model := range []string{"gpt-5.6-luna", "kimi-k3"} {
	67	+	stream, err := router.Stream(context.Background(), Request{Model: model})
	68	+	if err != nil {
	69	+	t.Fatal(err)
	70	+	}
	71	+	for range stream {
	72	+	}
	73	+	}
	74	+
	75	+	if len(responsesCapture.models) != 1 \|\| responsesCapture.models[0] != "gpt-5.6-luna" {
	76	+	t.Fatalf("Responses models = %v", responsesCapture.models)
	77	+	}
	78	+	if len(completionsCapture.models) != 1 \|\| completionsCapture.models[0] != "kimi-k3" {
	79	+	t.Fatalf("Completions models = %v", completionsCapture.models)
	80	+	}
	81	+	}


@@ -48,3 +48,34 @@ func TestModelRouterRejectsMissingAPIClient(t *testing.T) {
48	t.Fatal("expected missing API client error")	48		t.Fatal("expected missing API client error")
49	}	49		}
50	}	50		}
		51	+
		52	+	func TestOpenCodeGoRoutesLunaToResponses(t *testing.T) {
		53	+	router := NewOpenCodeGo("token", "https://example.com/go/v1").(*modelRouter)
		54	+	if got := router.fallback.(*openaiClient).baseURL; got != "https://example.com/go/v1" {
		55	+	t.Fatalf("Completions base URL = %q", got)
		56	+	}
		57	+	responses := router.byAPI[APIResponses].(renamedClient).inner.(codexClient)
		58	+	if got := responses.baseURL; got != "https://example.com/go/v1/responses" {
		59	+	t.Fatalf("Responses base URL = %q", got)
		60	+	}
		61	+
		62	+	completionsCapture := &routeCaptureClient{name: "opencode-go"}
		63	+	responsesCapture := &routeCaptureClient{name: "opencode-go"}
		64	+	router.fallback = completionsCapture
		65	+	router.byAPI[APIResponses] = responsesCapture
		66	+	for _, model := range []string{"gpt-5.6-luna", "kimi-k3"} {
		67	+	stream, err := router.Stream(context.Background(), Request{Model: model})
		68	+	if err != nil {
		69	+	t.Fatal(err)
		70	+	}
		71	+	for range stream {
		72	+	}
		73	+	}
		74	+
		75	+	if len(responsesCapture.models) != 1 \|\| responsesCapture.models[0] != "gpt-5.6-luna" {
		76	+	t.Fatalf("Responses models = %v", responsesCapture.models)
		77	+	}
		78	+	if len(completionsCapture.models) != 1 \|\| completionsCapture.models[0] != "kimi-k3" {
		79	+	t.Fatalf("Completions models = %v", completionsCapture.models)
		80	+	}
		81	+	}

M packages/provider/openai.go +5 -0


@@ -264,6 +264,11 @@ func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {
264	264		// don't clamp zut's "maximum" to "high" for those models.
265	265		effort = OpenAICompatAnthropicEffort(reasoning)
266	266		}
	267	+	if hasReasoningLevelOverride(m, req.Reasoning) {
	268	+	// Explicit model mappings describe the endpoint's actual effort
	269	+	// values and take precedence over conservative protocol defaults.
	270	+	effort = reasoning
	271	+	}
267	272		if effort != "" {
268	273		out.ReasoningEffort = effort
269	274		}


@@ -264,6 +264,11 @@ func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {
264	// don't clamp zut's "maximum" to "high" for those models.	264		// don't clamp zut's "maximum" to "high" for those models.
265	effort = OpenAICompatAnthropicEffort(reasoning)	265		effort = OpenAICompatAnthropicEffort(reasoning)
266	}	266		}
		267	+	if hasReasoningLevelOverride(m, req.Reasoning) {
		268	+	// Explicit model mappings describe the endpoint's actual effort
		269	+	// values and take precedence over conservative protocol defaults.
		270	+	effort = reasoning
		271	+	}
267	if effort != "" {	272		if effort != "" {
268	out.ReasoningEffort = effort	273		out.ReasoningEffort = effort
269	}	274		}

M packages/provider/openai_codex.go +5 -1


@@ -206,7 +206,11 @@ func (c codexClient) buildRequest(req Request) (codexRequest, error) {
206	206		body.ServiceTier = fastModeServiceTier
207	207		}
208	208		if m.Reasoning {
209		-	if effort := OpenAICodexReasoningEffort(reasoning, req.Model); effort != "" {
	209	+	effort := OpenAICodexReasoningEffort(reasoning, req.Model)
	210	+	if hasReasoningLevelOverride(m, req.Reasoning) {
	211	+	effort = reasoning
	212	+	}
	213	+	if effort != "" {
210	214		body.Reasoning = &codexReasoningConfig{Effort: effort}
211	215		}
212	216		}

 		body.ServiceTier = fastModeServiceTier
 	}
 	if m.Reasoning {
-		if effort := OpenAICodexReasoningEffort(reasoning, req.Model); effort != "" {
+		effort := OpenAICodexReasoningEffort(reasoning, req.Model)
+		if hasReasoningLevelOverride(m, req.Reasoning) {
+			effort = reasoning
+		}
+		if effort != "" {
 			body.Reasoning = &codexReasoningConfig{Effort: effort}
 		}
 	}

M packages/provider/openai_codex_test.go +21 -0


@@ -126,6 +126,27 @@ func TestGPT56UsesNativeMaxReasoningEffort(t *testing.T) {
126	126		}
127	127		}
128	128
	129	+	func TestResponsesRequestUsesExplicitReasoningLevelMap(t *testing.T) {
	130	+	SetLiveModels([]Model{{
	131	+	Provider: "custom-responses",
	132	+	ID: "reasoning-model",
	133	+	API: APIResponses,
	134	+	Reasoning: true,
	135	+	ReasoningLevelMap: map[string]string{"max": "max"},
	136	+	}})
	137	+	t.Cleanup(func() { SetLiveModels(nil) })
	138	+
	139	+	named := NewOpenAIResponsesNamed("token", "https://example.test/v1", "custom-responses").(*renamedClient)
	140	+	c := named.inner.(*codexClient)
	141	+	wire, err := c.buildRequest(Request{Model: "reasoning-model", Reasoning: "max"})
	142	+	if err != nil {
	143	+	t.Fatal(err)
	144	+	}
	145	+	if wire.Reasoning == nil \|\| wire.Reasoning.Effort != "max" {
	146	+	t.Fatalf("reasoning = %+v", wire.Reasoning)
	147	+	}
	148	+	}
	149	+
129	150		func TestXAIResponsesUsesNamedProviderCatalogAndEndpoint(t *testing.T) {
130	151		named := NewOpenAIResponsesNamed("token", "https://api.x.ai/v1", "xai").(*renamedClient)
131	152		c := named.inner.(*codexClient)


@@ -126,6 +126,27 @@ func TestGPT56UsesNativeMaxReasoningEffort(t *testing.T) {
126	}	126		}
127	}	127		}
128		128
		129	+	func TestResponsesRequestUsesExplicitReasoningLevelMap(t *testing.T) {
		130	+	SetLiveModels([]Model{{
		131	+	Provider: "custom-responses",
		132	+	ID: "reasoning-model",
		133	+	API: APIResponses,
		134	+	Reasoning: true,
		135	+	ReasoningLevelMap: map[string]string{"max": "max"},
		136	+	}})
		137	+	t.Cleanup(func() { SetLiveModels(nil) })
		138	+
		139	+	named := NewOpenAIResponsesNamed("token", "https://example.test/v1", "custom-responses").(*renamedClient)
		140	+	c := named.inner.(*codexClient)
		141	+	wire, err := c.buildRequest(Request{Model: "reasoning-model", Reasoning: "max"})
		142	+	if err != nil {
		143	+	t.Fatal(err)
		144	+	}
		145	+	if wire.Reasoning == nil \|\| wire.Reasoning.Effort != "max" {
		146	+	t.Fatalf("reasoning = %+v", wire.Reasoning)
		147	+	}
		148	+	}
		149	+
129	func TestXAIResponsesUsesNamedProviderCatalogAndEndpoint(t *testing.T) {	150		func TestXAIResponsesUsesNamedProviderCatalogAndEndpoint(t *testing.T) {
130	named := NewOpenAIResponsesNamed("token", "https://api.x.ai/v1", "xai").(*renamedClient)	151		named := NewOpenAIResponsesNamed("token", "https://api.x.ai/v1", "xai").(*renamedClient)
131	c := named.inner.(*codexClient)	152		c := named.inner.(*codexClient)

M packages/provider/provider_test.go +14 -4


@@ -649,10 +649,8 @@ func TestGPT56CatalogEntries(t *testing.T) {
649	649		}
650	650		}
651	651
652		-	// TestOpenCodeGoCatalog pins the Grok 4.5 and Kimi K3 entries added to
653		-	// the OpenCode Go provider. Both are listed in the vendor's current Go
654		-	// model lineup (https://opencode.ai/docs/go) and were missing from the
655		-	// baked-in catalog.
	652	+	// TestOpenCodeGoCatalog pins entries added from the vendor's current Go
	653	+	// model lineup (https://opencode.ai/docs/go).
656	654		func TestOpenCodeGoCatalog(t *testing.T) {
657	655		cases := []struct {
658	656		id string

@@ -662,6 +660,7 @@ func TestOpenCodeGoCatalog(t *testing.T) {
662	660		priceOut float64
663	661		cacheRead float64
664	662		}{
	663	+	{"gpt-5.6-luna", 1050000, 128000, 0.2, 1.2, 0.02},
665	664		{"grok-4.5", 500000, 500000, 2, 6, 0.3},
666	665		{"kimi-k3", 1048576, 131072, 3, 15, 0.3},
667	666		}

@@ -680,6 +679,17 @@ func TestOpenCodeGoCatalog(t *testing.T) {
680	679		t.Fatalf("opencode-go/%s baseURL: %s", tc.id, m.BaseURL)
681	680		}
682	681		}
	682	+
	683	+	luna, err := FindModel("opencode-go", "gpt-5.6-luna")
	684	+	if err != nil {
	685	+	t.Fatal(err)
	686	+	}
	687	+	if luna.API != APIResponses \|\| luna.PriceCacheWrite != 0.25 {
	688	+	t.Fatalf("opencode-go/gpt-5.6-luna API or cache-write price: %+v", luna)
	689	+	}
	690	+	if luna.PriceTierInputTokens != 272000 \|\| luna.PriceInputAbove != 0.4 \|\| luna.PriceOutputAbove != 1.8 \|\| luna.PriceCacheReadAbove != 0.04 \|\| luna.PriceCacheWriteAbove != 0.5 {
	691	+	t.Fatalf("opencode-go/gpt-5.6-luna long-context prices: %+v", luna)
	692	+	}
683	693		}
684	694
685	695		// TestOpenAIOmitsZeroMaxTokens guards against sending max_tokens: 0 for

 	}
 }
-// TestOpenCodeGoCatalog pins the Grok 4.5 and Kimi K3 entries added to
+// TestOpenCodeGoCatalog pins entries added from the vendor's current Go
-// the OpenCode Go provider. Both are listed in the vendor's current Go
+// model lineup (https://opencode.ai/docs/go).
-// model lineup (https://opencode.ai/docs/go) and were missing from the
-// baked-in catalog.
 func TestOpenCodeGoCatalog(t *testing.T) {
 	cases := []struct {
 		id        string
 		priceOut  float64
 		cacheRead float64
 	}{
+		{"gpt-5.6-luna", 1050000, 128000, 0.2, 1.2, 0.02},
 		{"grok-4.5", 500000, 500000, 2, 6, 0.3},
 		{"kimi-k3", 1048576, 131072, 3, 15, 0.3},
 	}
 			t.Fatalf("opencode-go/%s baseURL: %s", tc.id, m.BaseURL)
 		}
 	}
+	luna, err := FindModel("opencode-go", "gpt-5.6-luna")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if luna.API != APIResponses || luna.PriceCacheWrite != 0.25 {
+		t.Fatalf("opencode-go/gpt-5.6-luna API or cache-write price: %+v", luna)
+	}
+	if luna.PriceTierInputTokens != 272000 || luna.PriceInputAbove != 0.4 || luna.PriceOutputAbove != 1.8 || luna.PriceCacheReadAbove != 0.04 || luna.PriceCacheWriteAbove != 0.5 {
+		t.Fatalf("opencode-go/gpt-5.6-luna long-context prices: %+v", luna)
+	}
 }
 // TestOpenAIOmitsZeroMaxTokens guards against sending max_tokens: 0 for

M packages/provider/reasoning.go +7 -5


@@ -5,8 +5,8 @@ import "strings"
5	5		var reasoningLevelOrder = []string{"", "minimum", "low", "medium", "high", "xhigh", "max"}
6	6
7	7		// AvailableReasoningLevels returns the distinct reasoning levels supported by
8		-	// a model. Protocol defaults are adjusted by optional per-model overrides. The
9		-	// empty string represents off.
	8	+	// a model. Optional per-model overrides can remove, remap, or extend protocol
	9	+	// defaults. The empty string represents off.
10	10		func AvailableReasoningLevels(model Model) []string {
11	11		defaults := defaultReasoningLevels(model)
12	12		if !model.Reasoning \|\| len(model.ReasoningLevelMap) == 0 {

@@ -24,9 +24,6 @@ func AvailableReasoningLevels(model Model) []string {
24	24		if mapped, overridden := model.ReasoningLevelMap[level]; overridden {
25	25		effective = NormalizeReasoning(mapped)
26	26		}
27		-	if effective != "" && !containsReasoningLevel(defaults, effective) {
28		-	effective = nearestReasoningLevel(defaults, effective)
29		-	}
30	27		if reasoningLevelRank(effective) > 0 {
31	28		available[effective] = true
32	29		}

@@ -90,6 +87,11 @@ func containsReasoningLevel(levels []string, target string) bool {
90	87		return false
91	88		}
92	89
	90	+	func hasReasoningLevelOverride(model Model, level string) bool {
	91	+	_, ok := model.ReasoningLevelMap[NormalizeReasoning(level)]
	92	+	return ok
	93	+	}
	94	+
93	95		// ClampReasoningForModel maps a configured level to the nearest level exposed
94	96		// for the active model. Ties prefer the higher level.
95	97		func ClampReasoningForModel(model Model, level string) string {

 var reasoningLevelOrder = []string{"", "minimum", "low", "medium", "high", "xhigh", "max"}
 // AvailableReasoningLevels returns the distinct reasoning levels supported by
-// a model. Protocol defaults are adjusted by optional per-model overrides. The
+// a model. Optional per-model overrides can remove, remap, or extend protocol
-// empty string represents off.
+// defaults. The empty string represents off.
 func AvailableReasoningLevels(model Model) []string {
 	defaults := defaultReasoningLevels(model)
 	if !model.Reasoning || len(model.ReasoningLevelMap) == 0 {
 		if mapped, overridden := model.ReasoningLevelMap[level]; overridden {
 			effective = NormalizeReasoning(mapped)
 		}
-		if effective != "" && !containsReasoningLevel(defaults, effective) {
-			effective = nearestReasoningLevel(defaults, effective)
-		}
 		if reasoningLevelRank(effective) > 0 {
 			available[effective] = true
 		}
 	return false
 }
+func hasReasoningLevelOverride(model Model, level string) bool {
+	_, ok := model.ReasoningLevelMap[NormalizeReasoning(level)]
+	return ok
+}
 // ClampReasoningForModel maps a configured level to the nearest level exposed
 // for the active model. Ties prefer the higher level.
 func ClampReasoningForModel(model Model, level string) string {

M packages/provider/reasoning_test.go +34 -10


@@ -29,7 +29,17 @@ func TestAvailableReasoningLevels(t *testing.T) {
29	29		Provider: "anthropic", Reasoning: true,
30	30		ReasoningLevelMap: map[string]string{"minimum": "low", "high": "", "max": "max"},
31	31		},
32		-	want: []string{"", "low", "medium", "xhigh"},
	32	+	want: []string{"", "low", "medium", "xhigh", "max"},
	33	+	},
	34	+	{
	35	+	name: "explicit levels extend generic defaults",
	36	+	model: Model{
	37	+	Provider: "custom", Reasoning: true,
	38	+	ReasoningLevelMap: map[string]string{
	39	+	"minimum": "", "low": "", "medium": "", "high": "high", "xhigh": "xhigh", "max": "max",
	40	+	},
	41	+	},
	42	+	want: []string{"", "high", "xhigh", "max"},
33	43		},
34	44		}
35	45		for _, tt := range tests {

@@ -59,6 +69,12 @@ func TestClampReasoningForModel(t *testing.T) {
59	69		level: "minimum",
60	70		want: "high",
61	71		},
	72	+	{
	73	+	name: "explicit max extends generic defaults",
	74	+	model: Model{Reasoning: true, ReasoningLevelMap: map[string]string{"max": "max"}},
	75	+	level: "max",
	76	+	want: "max",
	77	+	},
62	78		}
63	79		for _, tt := range tests {
64	80		t.Run(tt.name, func(t *testing.T) {

@@ -71,20 +87,28 @@ func TestClampReasoningForModel(t *testing.T) {
71	87
72	88		func TestOpenAIRequestUsesReasoningLevelMap(t *testing.T) {
73	89		SetLiveModels([]Model{{
74		-	Provider: "openai",
	90	+	Provider: "custom",
75	91		ID: "mapped-reasoning-model",
76	92		Reasoning: true,
77		-	ReasoningLevelMap: map[string]string{"high": "low"},
	93	+	ReasoningLevelMap: map[string]string{"high": "low", "max": "max"},
78	94		}})
79	95		t.Cleanup(func() { SetLiveModels(nil) })
80	96
81		-	client := NewOpenAI("test", "").(*openaiClient)
82		-	request, err := client.buildRequest(Request{Model: "mapped-reasoning-model", Reasoning: "high"})
83		-	if err != nil {
84		-	t.Fatal(err)
85		-	}
86		-	if request.ReasoningEffort != "low" {
87		-	t.Fatalf("reasoning effort = %q, want low", request.ReasoningEffort)
	97	+	client := NewOpenAICompat("custom", "test", "", "").(*openaiClient)
	98	+	for _, tt := range []struct {
	99	+	requested string
	100	+	want string
	101	+	}{
	102	+	{requested: "high", want: "low"},
	103	+	{requested: "max", want: "max"},
	104	+	} {
	105	+	request, err := client.buildRequest(Request{Model: "mapped-reasoning-model", Reasoning: tt.requested})
	106	+	if err != nil {
	107	+	t.Fatal(err)
	108	+	}
	109	+	if request.ReasoningEffort != tt.want {
	110	+	t.Errorf("reasoning effort for %q = %q, want %q", tt.requested, request.ReasoningEffort, tt.want)
	111	+	}
88	112		}
89	113		}
90	114

 				Provider: "anthropic", Reasoning: true,
 				ReasoningLevelMap: map[string]string{"minimum": "low", "high": "", "max": "max"},
 			},
-			want: []string{"", "low", "medium", "xhigh"},
+			want: []string{"", "low", "medium", "xhigh", "max"},
+		},
+		{
+			name: "explicit levels extend generic defaults",
+			model: Model{
+				Provider: "custom", Reasoning: true,
+				ReasoningLevelMap: map[string]string{
+					"minimum": "", "low": "", "medium": "", "high": "high", "xhigh": "xhigh", "max": "max",
+				},
+			},
+			want: []string{"", "high", "xhigh", "max"},
 		},
 	}
 	for _, tt := range tests {
 			level: "minimum",
 			want:  "high",
 		},
+		{
+			name:  "explicit max extends generic defaults",
+			model: Model{Reasoning: true, ReasoningLevelMap: map[string]string{"max": "max"}},
+			level: "max",
+			want:  "max",
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 func TestOpenAIRequestUsesReasoningLevelMap(t *testing.T) {
 	SetLiveModels([]Model{{
-		Provider:          "openai",
+		Provider:          "custom",
 		ID:                "mapped-reasoning-model",
 		Reasoning:         true,
-		ReasoningLevelMap: map[string]string{"high": "low"},
+		ReasoningLevelMap: map[string]string{"high": "low", "max": "max"},
 	}})
 	t.Cleanup(func() { SetLiveModels(nil) })
-	client := NewOpenAI("test", "").(*openaiClient)
+	client := NewOpenAICompat("custom", "test", "", "").(*openaiClient)
-	request, err := client.buildRequest(Request{Model: "mapped-reasoning-model", Reasoning: "high"})
+	for _, tt := range []struct {
-	if err != nil {
+		requested string
-		t.Fatal(err)
+		want      string
-	}
+	}{
-	if request.ReasoningEffort != "low" {
+		{requested: "high", want: "low"},
-		t.Fatalf("reasoning effort = %q, want low", request.ReasoningEffort)
+		{requested: "max", want: "max"},
+	} {
+		request, err := client.buildRequest(Request{Model: "mapped-reasoning-model", Reasoning: tt.requested})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if request.ReasoningEffort != tt.want {
+			t.Errorf("reasoning effort for %q = %q, want %q", tt.requested, request.ReasoningEffort, tt.want)
+		}
 	}
 }