[linkerlin]

Summary

未完成 run 的第一次 swarm_tick 重定向现在也写 retryable=false。把缺席的 retryable 当成「可以再 tick」、只在严格 false 时才停、同时不看 tick_again / next_action / ok 的宿主，会在第一次软返回上停下来；ok 仍是 true，不会再掉进保险丝的 ok=false。gate_soft_continue 仍然缺席，不把严格 false 写回那个会被读成整段会话失败的字段。

The first unfinished-run swarm_tick redirect now sets retryable=false as well. Hosts that treat an omitted retryable as permission to tick again, and that ignore tick_again / next_action / ok, stop on that first soft return while ok stays true, so they do not fall through to the fuse's ok=false. gate_soft_continue stays omitted.

What changed

• 未完成 run 的两次默认 tick（有软缺口与无软缺口）都写 retryable=false。第一次仍是 ok=true、没有 error、没有 loop_fuse，redirect_reason 仍是 pending_solidify。保险丝仍是 ok=false、loop_fuse=pending_redirect_loop、redirect_reason=pending_redirect_loop、没有 error。
• 软缺口上 gate_soft_continue 仍然缺席。缺席不是 false：只把严格 false 当成整段会话失败、且不读 session_not_failed 的宿主不会在第一次就结束会话。只认这一个字段、完全不读 retryable 的宿主仍可能再 tick；把该键写成 false 会重开 DEBUG #67。
• session_not_failed=true 与 preserve 三元组不动。boot / status 不写 retryable。不写 Parent baseline，不放宽固化门。DEBUG #61 的目录缓存没有做假修复。
• 仪器、swarm_tick 参数说明、README（英/中）、quickstart、AGENTS、DEBUG #69 写同一口径。常驻 instructions 未加长（仍小于 1200）。

How to test

1. uv run pytest tests/test_mcp_host_usability_pending_smoke.py tests/test_mcp_server.py tests/test_mcp_protocol.py tests/test_mcp_host_usability_p0.py tests/test_swarm.py -q --tb=line
   • Expected: 128 passed.
   • The new assertion (retryable is False on the first redirect, and a host that ticks unless retryable is strict false does not tick) fails on unmodified master.
2. uv run ruff check src tests && uv run ruff format --check src tests && uv run mypy src
   • Expected: ruff clean, mypy Success: no issues found in 353 source files.
3. Pre-existing, not part of this diff: uv run pytest tests/gep/test_material.py::TestWatermarkIngest::test_identical_rewrite_is_noop -q --tb=line still fails on unmodified master in this environment (skipped vs rewrite_noop). Identical write_text restores the previous mtime_ns, so the watermark fast-skip is doing what it says. Callers only read added. Not changed here.

Risk

Low. The unfinished-run tick envelope gains retryable=false on the first redirect, where it was previously absent. The fuse already had that value. ok on the first call stays true, so hosts that treat ok=false as session death do not die on the accidental first tick. gate_soft_continue is still omitted. Solidify gate, skip_validation, Parent baseline, and val seal are unchanged.

Harness/evaluator governance

Upstream governance surface: MCP swarm_tick unfinished-run redirect envelope (retryable only). Not solidify, selector, mutation, or the frozen gate.
Downstream EvoX impact: N/A -- existing tool return field, no bridge or contract change.
Rollout-local scope: host-visible tick redirect only; boot and status are unchanged.
Promotion boundary: N/A -- no proposal/rollout/default promotion.
Evaluator mismatch sets: N/A
Non-regression evidence: tests/test_mcp_host_usability_pending_smoke.py pins the first redirect (ok true, retryable false, gate_soft_continue omitted, session_not_failed true) and the fuse (ok false unchanged).
Fix-severity review: low
Owner approval: N/A
Security boundary: no new data, tool, host, network, or secret surface.
Rollback: revert this commit.
Live promotion: no
Autonomous evaluator self-editing: no

Self-check

• No new runtime dependencies without a clear justification in "What changed".
• Tests added or updated to cover the new behavior; gates pass locally: uv run pytest (modules listed above), uv run ruff check src tests, uv run ruff format --check src tests, uv run mypy src.

  

[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.