v1.6.13

This release fixes how autopsy counts check runs. A foreground run-checks or task-checks call over two minutes was listed as a long_tool_call slowdown, although it is the check policy's own run, which the checks section already times. A helper started through a shell variable holding the interpreter ("$python" .../run-checks) was not counted as a check at all. And the chat summary's "full-suite runs" line showed the count of check calls with a test group, not the number of full-suite runs.

Built with subagents (a fresh adversarial review of the plan, an implementer and a reviewer per task, then a whole-branch review), without a parallel-lanes run, and released without a validation run on a real project.

What's fixed

• Helper check calls are check time. A foreground run-checks or task-checks call is no longer a long_tool_call. Its time is check time, a category of its own: it is charged before the lower rows, so an edit retry or a review round around it does not absorb it, and it is never listed as a slowdown. The unexplained list treats it as covered, as it already did for long checks run in the background. Those background checks (a start and await-checks wait calls) are check time the same way now, so a retried, escalated, redone or extra-review agent no longer carries its check time whichever way the check ran. A helper call that hits the Bash timeout still shows its timeout row. A test command run directly (npm test, node tests/run-tests.mjs) stays a long_tool_call, since that row's suggestion, a narrower command while iterating, is about exactly those runs.
• Helpers through a variable. A bare $NAME or ${NAME} in front of a helper script is read like an interpreter, so "$python" .../run-checks counts as a run-checks call.
• The summary names the project test runs. The chat summary's line reads project test runs: P; check calls with a test group: T (by role). history.json carries each run's project test runs (the executions known to have run) and how many could not be told, and history.md's table has a Project test runs column (N (+M unavailable) when some are unknown) beside the renamed Test-group calls.
• Retired: checks run through a script file. Some runs wrote a task-checks call into a script with the Write tool and ran bash <script>, a workaround for long command lines that 1.6.6's short task-checks command removed. Reading those scripts would need positional-argument substitution and splicing into the line, for a handful of archived runs. They stay uncounted.

Measured on the run history

autopsy 1.6.12 and 1.6.13 over the same copy of 38 archived runs, each run analysed once by each version:

• long_tool_call: 315 rows, 67,545.8 s before; 213 rows, 43,039.7 s after. The 102 rows that went were foreground helper calls, all but one from runs on engines before 1.6.1 (since 1.6.1 long checks run in the background and were already no slowdown).
• Check time inside other rows: escalation 2,414.1 -> 988.8 s, relaunch 14,457.6 -> 11,288.6 s, extra_review_round 12,550.4 -> 11,794.1 s, retry 2,188.6 -> 2,155.8 s (the foreground helper calls of any length and the background checks inside those agents are check time now; 4,098.4 s of it from background checks). No kind grows.
• All slowdowns: 134,458.9 s before, 104,569.4 s after (22.2% less); rows 740 -> 638.
• Unexplained: 150 spans (19,866.7 s) -> 143 (19,070.1 s), since helper calls now cover their own time.
• Check executions: two more counted (helper calls started through a variable).

Testing

• New tests, each failing before its change: helpers through $python and ${PY} (and variables that are no helper, unchanged); a 300 s foreground run-checks call with no slowdown row and no unexplained span, against a direct npm test that stays a long_tool_call; a run-checks call inside an edit retry's window and inside a fix agent's round, whose time neither absorbs; a run-checks call that times out; the summary line where the two counts differ; history's field and columns. The real fixtures from 1.6.11 now show their run-checks calls as check executions.
• Tests: full suite on Windows (Python 3.15) at the release head: 1935 tests (1919 pass, 16 skipped, 0 fail).
• CI: passes on Ubuntu (Node 24, and Node 24 with Python 3.15), macOS (Node 24) and Windows (Node 24, Python 3.12).
• Not run: a parallel-lanes run with this release installed.

Compatibility

• autopsy_version is 10; history reruns autopsy on every archived run with an older autopsy.json.
• Fewer long_tool_call rows on runs that called the helpers in the foreground; escalation, extra_review_round and relaunch rows can be smaller by the check time inside them.
• history.md's Full-suite runs column is now Test-group calls, with Project test runs beside it; history.json run entries gain project_test_runs.

Upgrade

In your clone:

git pull && bash install.sh