fix: give the post-release triage sweep a scheduler-level trigger (#1386)

[fdaviddpt]

What

A post-release triage sweep is a stated cadence rule (three to four ticks, then a release, then a
triage sweep, #855) that has never fired on its own -- the first sweep this loop ever recorded was
run by hand, across 31 tagged releases. scripts/oss_state.py --triage-recorded/--last-triage
already record and read the last sweep in three states; nothing consumed the answer, because the
cadence rule lives in skills/manager/phases/accounting.md, read only by a sub-manager that dies
with its own context before a release and cannot count ticks or releases across its own spawns.

Closes #1386.

How

scripts/triage_trigger.py follows the same shape scripts/release_trigger.py (#966) already
gives its own cadence step: a real threshold in config (release.triggers.triage_after_release,
absent means not wanted, the same rule curate_route_threshold already uses), evaluated
mechanically against the most recent tag's own commit date and the last recorded sweep, in three
states -- due / not-due / could-not-tell, never folding an unreadable input into not-due.

commands/tick.md reads it at the RELEASE: released handback -- the one point in the scheduler
that spans ticks -- and dispatches oss:triager on due. It then records the sweep's completion
with oss_state.py --triage-recorded once the triager's own final message confirms the sweep ran,
closing the loop back to not-due; without that step the trigger would fire on every release
forever (a self-review finding, see below).

scripts/oss_config.py gains TRIGGER_BOOLEAN_KEYS to validate the new boolean trigger key
alongside the existing numeric merged_prs/soak_hours ones. skills/manager/phases/ accounting.md's Cadence section and docs/overview.md's "What is not true yet" section (the exact
worked example this issue's own body is built from) are both updated to describe the mechanism
rather than the gap.

Out of scope, on purpose

The label-coverage measurement (how much of the open board carries no priority label) that
accounting.md itself still calls "still unbuilt" is a separate, second half of the same defect
this issue's body names, and this lane does not touch it -- confirmed against the maintainer's own
issue comment narrowing the fix to the trigger alone.

Self-review

Two spawns (Explore, oss:auditor) against the committed diff. Both findings fixed in place:

• docs/overview.md's "What is not true yet" section still described the exact defect this diff
  closes as current. Reworded to the past tense.
• The automated oss:triager dispatch never advanced --last-triage, so due would never clear.
  commands/tick.md now records the sweep once the triager confirms it ran.
• A third finding (the new TRIGGER_BOOLEAN_KEYS validator branch was correct but had no
  exercising test) closed with two new tests in tests/test_oss_config.py.

Tree-snapshot compare showed mutated, attributable entirely to this session's own legitimate
self-review fix commits made after the pre-review snapshot -- both reviewers explicitly reported
performing no mutation of their own.

Tests

Red: before scripts/triage_trigger.py existed, tests/test_triage_trigger_1386.py failed at
import (ModuleNotFoundError: No module named 'triage_trigger').

Green: python3 -m pytest tests/test_triage_trigger_1386.py -q --no-cov -> 8 passed.

Guards derived via scripts/lane_setup.py 1386 --lane <each touched file> (six named:
test_bare_gh_git_spawn_sweep_1165, test_claude_md_currency, test_command_references,
test_content_invariants, test_gate_state_consumers_328, test_unwired_scripts_253), plus the
byte-budget and config test files directly relevant to the touched files -- 787 tests total,
all green. Full pytest tests/ -q was not run locally, per this repo's own rule that CI's 13-leg
matrix is the authority and a local full run answers a weaker question than CI does.

🤖 Generated with Claude Code

https://claude.ai/code/session_012UYdzDHX3k3C2Eh7B5hRVH

[AI-generated]