xonecas.io
64875
Mc

mini-coder / commit 56d1062

fix: remove shell tool timeout limits that killed long-running commands

Root cause: AI SDK streamText `chunkMs: 120_000` timeout aborted the
entire stream when tool execution took >120s with no chunks arriving.
The shell tool schema also capped explicit timeouts at 5 min.

A coding agent should not artificially limit subprocess duration.
Users can interrupt with ESC at any time.

Changes:
- Remove `timeout: { chunkMs: 120_000 }` from streamText options
- Remove `.max(300_000)` cap from shell tool schema
- Remove resolved known issue from KNOWN_ISSUES.md

xonecas2026-03-21 56d10627693842dc40c6a0b28699192005a56bf7 parent e3b3a1a

3 files changed

docs/KNOWN_ISSUES.md+0−1
@@ -4,7 +4,6 @@
44
55 - `/model` doesn't autocomplete models provider/name pattern.
66 - AI SDK expands `claude-3-5-haiku` → dated variant that Zen doesn't serve (404).
7-- Shell tool max-timeout leaves the parent terminal in a broken state.
87
98 ## Features
109
src/llm-api/turn-request.ts+0−1
@@ -137,6 +137,5 @@ export function buildStreamTextRequest(
137137 // The AI SDK logs errors to stderr by default. We surface failures through
138138 // streamed turn events so CLI output stays compact and consistent.
139139 },
140- timeout: { chunkMs: 120_000 },
141140 } as StreamTextOptions;
142141 }
src/tools/shell.ts+1−2
@@ -10,10 +10,9 @@ const ShellSchema = z.object({
1010 .number()
1111 .int()
1212 .min(1000)
13- .max(300_000)
1413 .optional()
1514 .describe(
16- "Timeout in milliseconds (max: 5min). If omitted, the command runs until it exits.",
15+ "Timeout in milliseconds. If omitted, the command runs until it exits.",
1716 ),
1817 env: z
1918 .record(z.string(), z.string())