test: add decline-token coverage to the verdict-parser contract (#88)#95
Merged
Conversation
The reference parser (tests/test_verdict_parser.py) modeled only the advisory (green|yellow|red) and binary (pass|fail) tokens, but gate1 also emits `decline` (the /ceo-escalation signal, start.md step 10) — so the contract test and the command had drifted: a structured `## Verdict: decline` was unguarded. Add a `gate1` parser mode whose pattern mirrors start.md step 10 (green|yellow|red|decline), with `decline` as a structured-only token (a free-form mention in the reasoning body must not trigger it; the no-structured fallback reuses the advisory classifier, which can never return decline). Three fixtures lock the behavior: structured-decline → decline; decline-then- green → green (last-wins, no escalation); decline-in-body-only → green. Scope note: the broader "#88 parser-contract tests" was found ~90% already covered by the existing 25 fixtures + enum-sync-check.sh + jq-sync-check.sh (gate1, yellow). This PR closes the one genuine remaining gap — `decline`. 28 fixtures pass (was 25). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
JFK
added a commit
that referenced
this pull request
Jun 13, 2026
The v0.14.0 "token + precision optimization" milestone was re-audited against the actual files (using the token-baseline tool from #94). The premise turned out to be largely invalid: - The "~28% / ~6,400 token" compression is not achievable: slash commands load whole (no runtime include / conditional load), so relocating sections to an appendix saves nothing, and the bulk of start.md/ship.md is load-bearing executable spec that must not be compressed. - The claimed precision bugs were phantom: step-18b precedence is already an If/Else-if chain; verdict last-wins is already explicit (and now test-guarded by #95); the propose.md "parallel Skill" instruction is correct (batched Skill calls are supported); the propose.md "regex mismatch" is a harmless subset, not a contradiction. This commit ships the ONLY verified-safe, genuinely-beneficial residue: - start.md: delete a verbatim-redundant `lang != "en"` localization line (649) that duplicated line 647. - goal.md: convert the red-verdict force-continue prose (phase-aware bullets) into a compact decision table, preserving every load-bearing detail (the gate2.binary_gate `fail` exception, phase routing, continue-to steps). Net effect (per tests/token-baseline.sh): TOTAL ~78,424 -> ~78,355 tokens (-69 tokens, -0.09%). The negligible number is itself the finding — it demonstrates the milestone's compression premise was unfounded, and the token-baseline tool (#87/#94) measuring it is working as intended. Closes #89 Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #88
Summary
Add
decline-token coverage to the verdict-parser contract test. The reference parser (tests/test_verdict_parser.py) modeled only advisory (green|yellow|red) and binary (pass|fail), but gate1 also emitsdecline(the/ceo-escalation signal,start.mdstep 10) — so the contract test and the command had drifted: a structured## Verdict: declinewas unguarded.Implementation notes
gate1parser mode;GATE1_PATTERNmirrorsstart.mdstep 10 exactly:^\s*##\s*Verdict:\s*(green|yellow|red|decline)\b.declineis structured-only — a free-form mention in the reasoning body must not trigger it; the no-structured-line fallback reuses the advisory classifier (which can never returndecline).gate1-decline-structured→decline;gate1-decline-then-green→green(last-wins, no escalation);gate1-decline-in-body→green.Gate1 (design review) found that #88 as written is ~90% already implemented:
tests/test_verdict_parser.py's existing 25 fixtures already cover last-wins (edge-multiple-verdicts), trailing punctuation (edge-verdict-with-punctuation), case-insensitive, leading-whitespace, structured-overrides-heuristic, and both advisory + binary modes;enum-sync-check.shandjq-sync-check.shalready cover enum + jq sync in CI. This PR therefore closes the one genuine remaining gap — thedeclinetoken — rather than re-adding existing coverage. The milestone issue was drafted before fully auditing existing test coverage.Pre-PR review summary
gate2.binary_gate= none)🤖 Generated via /gh-issue-driven:ship (autonomous=red-only, milestone v0.14.0)