[remote-controlled-man]

Description

Paper2Code counted only tool names, so reading five different modules or writing several planned files consecutively could abort as a loop. Compare the tool name and a stable snapshot of its JSON arguments instead. Changed paths or content can proceed; repeated calls with identical arguments still stop the workflow, regardless of object key order.

Closes #294.

Changes Made

• Pass tool arguments from the shared implementation-workflow wrapper into loop detection.
• Preserve status probes, tool-name diagnostics, timeout, stall and error safeguards.
• Add regression tests for distinct reads, writes across turns, changed content, mutable argument snapshots and nested key ordering; retain genuine-repeat coverage.
• Update the English and Chinese changelogs.

Validation

Validated at 728e90f4fa969b2f0ac3435bf95bc26abc892a30:

• Base reproduction: the fifth read of a different file aborts the workflow; the new regression passes with the fix.
• Complete Python suite, Ubuntu 24.04 aarch64 / Python 3.12.3 with scripts/ci/requirements.lock: 3238 passed, 43 skipped.
• pre-commit run --all-files --show-diff-on-failure and python -m pip check: passed.
• Desktop/Web on Node 24.21.0: lint, 440 unit tests, typecheck, protocol check and npm run build (including version and Tauri-config checks): passed.
• npm run test:web, using Chromium and the real local service: 2 passed.
• Full-history Gitleaks scan: 754 commits, no leaks found.
• Python wheel/sdist build, Twine checks and clean-environment distribution verification: passed. Both artifacts contain the changed runtime files byte-for-byte; their file lists contain no local test tools or environment directories.

Not run locally: Rust fmt/clippy/tests, Windows lifecycle checks, or the Python 3.13/3.14 matrix. These still need their remote CI results. No live provider calls were used. The existing large-chunk build warnings remain.

Checklist

• Changes tested locally; platform limits are listed above.
• Independent read-only code and requirements review completed.
• Relevant regression tests added.
• Documentation updated.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.