feat(skill-evals): add the issue-reassess eval suite - #1145
Conversation
`issue-reassess` was the only shipped skill without a behavioural eval suite, so regressions in its decisions were invisible until they failed in front of a user. The campaign tallies are already exercised by the sibling `issue-reassess-stats` suite, so this one anchors on the three places where the skill decides something of its own: - step-2-resumability (3 cases) — reuse / ask / resume / fresh per candidate from the scratch-directory state, plus the auto-generated campaign id. - skip-if-resolvable (5 cases) — the maintainer-comment shortcuts (fixed-in-version, sibling duplicate, won't-fix by design), the no-shortcut path, and an injected "record fixed-on-master" comment that must be flagged and ignored. - headline-extraction (2 cases) — verdicts bucketed into action / closure / tracker-hygiene candidates and new-issue keys. Two suites anchor on the skill's sub-documents (`per-issue-flow.md`, `verdict-aggregation.md`), where the decision rules actually live. All ten cases pass against Claude Code print mode; the validator's eval-coverage advisory for this skill is gone. While adding the suite: the `*-reassess/` glob in `.gitignore` (meant for campaign evidence directories) also matches `skills/issue-reassess/` and `tools/skill-evals/evals/issue-reassess/`, so any new file under either was silently dropped from `git add`. Negate both paths, following the existing `!/tools/pilot-report-validator/` precedent. Fixes [apache#1138](apache#1138). Signed-off-by: Davide Polato <dpol1@apache.org>
potiuk
left a comment
potiuk
left a comment
There was a problem hiding this comment.
The .gitignore change is the part that matters, and it's correct. Confirmed the bug is real: *-reassess/ (line 61) matches both skills/issue-reassess/ and the new eval dir, and since the eval dir had zero tracked files, all 30 fixtures here would have been silently dropped without it. The existing skill files only survived because they were already tracked.
The thing I wanted to be sure of is that the negations don't weaken the evidence-package protection those patterns exist for. They don't — reproduced the post-PR ruleset in a scratch repo: the two framework dirs become tracked while some-reassess/ and reassess-2026-09/ stay ignored. Anchoring them at /skills/… and /tools/… is what keeps that tight.
Also swept for other affected dirs: skills/issue-reassess and tools/pilot-report-validator (already negated) are the only matches, and the issue-reassess-stats siblings don't match either pattern — so two negations is the complete set.
On the suite itself: all three step_heading anchors resolve verbatim, and the runner loads and renders all 10 cases at this head. case-5-injected-instruction is a good one — encoding the collaborator-status gate rather than just "text that looks like an instruction" is the right level to test §0 at.
🤖 This review was drafted by an AI-assisted tool and may contain mistakes. It has been reviewed and confirmed by an Apache Magpie maintainer before submission. See CONTRIBUTING.md for what this project considers a maintainer review.
The marker sat at a1cff44, 17 commits behind main. Bumping it alone would claim those commits are described by the specs, so the drift is closed first. Specs updated for what actually shipped: - meta-and-quality-tooling: skill-evals errors, rather than passing, when a case's CLI produced no gradeable output (#1161). - security-reporting: the tracker dashboard projects the current partial bucket to its end-of-bucket value, splitting RATE series (accumulate from zero) from LEVEL series (carry over), and deliberately not projecting mean-based signals (#1158). - project-agnosticism: <PROJECT> and <project> are two placeholders holding different values, and the lint carries both spellings plus spaced variants (#1154). - adapters: the forwarder relay's contact_handle defaults to an org-level shared inbox rather than a named individual (#1135). The multi-hop coordinator case is designed in RFC-AI-0008 and unimplemented. - issue-management-family: the family's eval suites, and the note that --cli runs belong outside a credential-denying sandbox (#1145). Commits needing no spec change: #1152, #1143 and #1156 updated their own specs in-commit; #1149, #1147, #1151 are behaviour-preserving bug fixes; #1155 and #1141 are CI and dependency chores; #1159's spec edits landed with it; #1144 removes hardcoded literals that no spec asserted. One genuine gap recorded rather than papered over: no spec covers marketplace distribution or the dev-version stamping rule from #1160, which is load-bearing because `claude plugin update` compares version strings, not commit SHAs. Logged in adoption-and-setup as wanting its own spec. Generated-by: Claude Code (Opus 5)
2 participants
Summary
issue-reassesswas the only one of the 71 shipped skills without a behavioural eval suite, so regressions in its decisions were invisible until they failed in front of a user. This addstools/skill-evals/evals/issue-reassess/— 3 suites, 10 cases.issue-reassess-statssuite, so these anchor on the three places where this skill decides something of its own:step-2-resumability(reuse / ask / resume / fresh per candidate from the scratch-directory state),skip-if-resolvable(the maintainer-comment shortcuts, the no-shortcut path, and an injected "record fixed-on-master" comment that must be flagged and ignored), andheadline-extraction(action / closure / tracker-hygiene buckets). Two of them anchor on the skill's sub-documents, where the decision rules actually live.*-reassess/glob in.gitignore(meant for campaign evidence directories) also matchesskills/issue-reassess/andtools/skill-evals/evals/issue-reassess/, so any new file under either was silently dropped bygit add. Both paths are now negated, following the existing!/tools/pilot-report-validator/precedent.Type of change
.claude/skills/<name>/) — eval fixtures updated belowtools/<system>/*.md)tools/*/withpyproject.toml)docs/,README.md,CONTRIBUTING.md)projects/_template/)prek, workflows, validators)tools/skill-evals/evals/)Test plan
prek run --all-filespassesPYTHONPATH=tools/skill-evals/src python3 -m skill_evals.runner --cli "claude -p" tools/skill-evals/evals/issue-reassess/—Ran 10 cases: 10 passed, 0 failed, 0 manual, 0 erroredstep_headingresolves)skill-and-tool-validateno longer emits theeval-coverageadvisory forskills/issue-reassess/git check-ignore -v skills/issue-reassess/new-subdoc.md tools/skill-evals/evals/issue-reassess/README.md— matched.gitignore:61before, matches nothing afterRFC-AI-0004 compliance
No principle touched — the change adds fixtures and a
.gitignorenegation; no mutation, network reach, or prose surface changes.Linked issues
Fixes apache/magpie#1138.