fix(runtime): pass option-like Grep patterns after a -- separator

[heeoneie]

Summary

The sandboxed filesystem worker appended the Grep pattern as a bare positional, so ripgrep parsed a pattern starting with - as flags. -webkit-box made rg exit 1, which operations.ts maps to { kind: 'grep', matches: [] } — the model was told the string is absent from a file that contains it, with no error and no diagnostic.

The host-local sibling already handles this at workspace-executor.ts:455; this change mirrors it in the worker.

Fixes #3733

Verification

Confirmed the underlying behaviour against ripgrep directly:

rg -n --no-heading --max-count=50 '-webkit-box' file.ts      -> exit 1, no output
rg -n --no-heading --max-count=50 -- '-webkit-box' file.ts   -> exit 0, 1:const style = "-webkit-box";

Ran locally against packages/runtime:

• npm --workspace @maka/core run build, @maka/storage, @maka/runtime — all clean
• node --test dist/__tests__/filesystem-worker.test.js — the new test passes; pass count goes 19 -> 20 with no new failures
• biome lint and biome format on both changed files — clean

Not run: the full workspace suite and the desktop app. This sandbox could not complete a root npm install (the registry mirror blocks @xterm/xterm, and the Electron postinstall cannot reach its download host), so only the core / storage / runtime workspaces were built and exercised.

One pre-existing failure in that file — reports no diff — not a new-file diff — when an existing file cannot be read — fails on unmodified main in this environment too, because the test runs as root and chmod 000 does not deny root. Unrelated to this change.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude (Anthropic) — drafted the one-line fix and the regression test, and ran the local verification above. Reviewed and submitted by me. Generated-by: Claude trailers are on both commits.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[M4n5ter]

Relationship note: PRs #3734, #3735, and #3903 all close #3733, touch the same two runtime files, and apply the same -- separator fix to the sandboxed Grep invocation. I could not find a canonical/supersede statement in these PR bodies or comments. At the current snapshot, this is the latest-created candidate, but the owner should decide the canonical PR. Please document that relationship and mark the other attempts as superseded/close them, or explain the distinct scope, so the issue is not merged three times.

This comment records the relationship only; it is not a merge decision.

---

Automated review note posted by @未开智选手. This is not an independent human review; a human should verify the conclusion.

[M4n5ter]

I reviewed exact head 7e7521c2fc6a6ac90e664da292d7f1fc56f6de07 and found no P0–P3 issues.

This fixes the root cause rather than masking the result: the user-controlled pattern now follows ripgrep's -- option terminator, so a leading dash cannot be parsed as flags before the existing exit-code mapping runs. It also removes the behavioral divergence from the host-local executor without adding a parser, state, or another execution path.

I compared the equivalent fixes in #3734 and #3735. I selected this version because its regression pins the complete terminal argv contract (['--', pattern, path]) and the full worker response with the smallest production patch. Removing only the separator made that regression fail; restoring it passed, as did all 21 filesystem-worker tests. A direct ripgrep 15.2.0 check reproduced exit 1/no output without -- and the expected matches with it. The current-main merge tree is clean and preserves the reviewed two-file patch.

---

Posted by an automated review agent operated by @M4n5ter. This is not an
independent human review and does not satisfy the committer review required by
CONTRIBUTING.md. A human is accountable for this comment — please push back if
anything here is wrong.

简体中文

本条评论由 @M4n5ter 运行的自动化审查程序发出。它不构成 CONTRIBUTING.md
所要求的独立人类审查，也不能替代人类审查。有人类对本条评论负责，如有错误请直接指出。