feat: v1.2 Diagnostics — step-precision, thin-coverage, round-emission, self-cycle wording - #81
Merged
Conversation
…ssion + self-cycle wording
Fix B (walker.py): generate_step_precision_concerns — four structural patterns
per step action: pip install without version pin, python -m <pkg> with no
subcommand, bare URL with no version token, bare model ID alongside LLM vendor SDK.
Each emits an edge-case concern with a per-step idempotency guard.
Fix C (llm_judge.py, findings.py): tier3-negative-paths-thin-coverage — when a
negative-path-omission finding is present and the step's negative-paths count is
< 3, emit an alongside warn finding. Threshold is >= 3 = documented. New kind
registered in KNOWN_KINDS and TIER3_CONTRADICTION_SEVERITY.
Fix D (walker.py): after each round_count increment in record_answer, emit
_status.emit("info", "walker.round", round=N, pending=K) for per-round operator
visibility. Wrapped in try/except so test contexts without _status can't break it.
Fix M (spec_ast.py): rewrite self-cycle-produces message and suggested_fix.
Message now says "references X which it also declares in produces (possible
self-cycle)." Suggested fix offers the two-branch operator choice: remove from
produces if action only names the path, or assert idempotency if it reads it.
Glossary (docs/glossary.md): added tier3-negative-paths-thin-coverage (finding)
and walker.round (status) entries with full dev/pm/triggered_by/user_action.
Tests: 1896 passed (+28 over 1868 baseline).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The 'Tier-3 demotion alongside-finding' framing was inaccurate. negative-path-omission is info-severity and never enters the faithfulness demotion path (_BLOCK_CONTRADICTION_KINDS in llm_judge.py only contains block-severity kinds). The new check fires alongside the original finding when negative-paths count < 3 — co-occurrence, not demotion. Rewriting dev:, pm:, triggered_by:, and dropping the misleading tier3-unfaithful-contradiction related: entry. Caught in Wave 4 Opus aggregate review. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Joncik91
added a commit
that referenced
this pull request
May 15, 2026
13 fixes across 5 classes surfaced by Vidence's v1.1 dogfooding cycle: - L: sanitized-input field with per-artifact-version taint tracking (#80) - E: gated Python-stack walker detection (#78) - H/I/N: --project flag + wizard placeholders + skill doc fix (#79) - A: calibrated-for exemplar audit (#77) - O/P/Q: excessive-post-ship-iteration wiring + tier3 run fingerprint + ADR extractor extensions (#82) - B/C/D/M: step-precision + thin-coverage + walker round + self-cycle wording (#81) No spec contract changes — v1.0/v1.1/v1.1.1 locked specs remain valid. Tests: 1842 → 1923 (+81). Five new finding kinds / status codes, all documented in docs/glossary.md. Co-authored-by: Copilot <copilot@github.com> Co-authored-by: Claude Opus 4.7 <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.
Summary
Four diagnostic-quality fixes for Spectre v1.2:
Fix B — Step action precision concerns (
bin/walker.py)generate_step_precision_concerns()function checks four structural vague-action patterns per step:pip installwithout version pin,python -m <pkg>with no subcommand, bare URL with no version token, bare model ID alongside LLM vendor SDK.edge-caseconcerns with per-step idempotency guard.tests/test_walker_step_precision.py.Fix C — Tier-3 thin-coverage finding (
bin/llm_judge.py,bin/findings.py)tier3-negative-paths-thin-coveragewarn finding: emitted alongside anynegative-path-omissionfinding when the step'snegative-paths:count is < 3 (threshold: ≥ 3 = "documented").KNOWN_KINDSandTIER3_CONTRADICTION_SEVERITY.tests/test_llm_judge_thin_coverage.py.Fix D — Walker round-count visibility (
bin/walker.py)record_answeremits_status.emit("info", "walker.round", round=N, pending=K)after eachround_countincrement.tests/test_walker_round_emission.py.Fix M — Self-cycle-produces wording (
bin/spec_ast.py)message: now says "references X which it also declares in produces (possible self-cycle)" — removes the incorrect "consumes" framing.suggested_fix: offers two branches — remove from produces if action only names the path; assert idempotency if action reads it.tests/test_spec_ast_self_cycle.py.Test plan
pytest tests/— 1896 passed (+28 over 1868 baseline)pytest tests/test_walker_step_precision.py— 17 tests, all 4 vague shapes + controlspytest tests/test_llm_judge_thin_coverage.py— 8 tests, thin/thick/non-omission casespytest tests/test_walker_round_emission.py— 2 tests, emit + field formatpytest tests/test_spec_ast_self_cycle.py— extends with 2 wording tests🤖 Generated with Claude Code