Content
65%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exceptionally concrete, well-sourced set of security invariants with exact file/function/test anchors — highly actionable for its genre. Its weaknesses are token efficiency and structure: enormous run-on bullets with historical rationale, and monolithic single-file density where per-area reference files would ease navigation.
Suggestions
Split each major area (Fork Routing, Forge Identity, Redaction, Trust Boundary, Gate Neutralization, Extra Gates) into its own reference file under references/ and keep SKILL.md as a short overview of the invariants, which would address both progressive disclosure and token load.
Break the mega-bullets into short imperative lines (invariant → anchor file/function → regression tests) and trim historical narrative such as issue/PR triage backstory and tool version notes to a compact 'history' note per section.
Add a short per-area checklist (make the change → run the named regression tests → confirm fail-closed behavior) so the implicit validation checkpoints become an explicit workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body contains no beginner-concept padding and every fact is project-specific, but single bullets run ten or more lines and embed historical narrative ("per the maintainer's triage on #1178", "Greptile finding on PR #1183", "empirically verified against omp 18.2.0") that is not needed to apply the invariant. This fits anchor 3 (mostly efficient but could be tightened) better than anchor 4, which expects only minor trims. | 3 / 5 |
Actionability | Directives are imperative and anchored to exact symbols, files, and flags ("resolve the push URL via `resolvePushURL` (`internal/pipeline/steps/common_git.go`)", "Reuse `internal/safeurl` for new redaction sites rather than adding a git-local helper", "never pass `<owner>:<branch>` to `gh pr list --head`"), with named regression tests per area. For an instruction-only skill this is fully concrete guidance covering the common cases, matching anchor 5; the rubric notes say absence of code is not penalized when guidance is actionable. | 5 / 5 |
Workflow Clarity | The content is organized by subsystem rather than as a sequenced process — there are no ordered steps and no explicit validate→fix→retry loops; the extensive regression-test lists serve only as implicit checkpoints. This matches anchor 3 (checkpoints missing or implicit); it is a multi-area invariants skill, so the simple-skill exception to score 5 does not apply, and no destructive/batch cap is triggered. | 3 / 5 |
Progressive Disclosure | Six clearly bold-headed sections give real structure, but roughly 20 KB of dense invariants live in one file with no bundle files — the repo-config trust boundary section alone is a wall of text that clearly belongs in a separate reference. This fits anchor 3 (some structure, but content that should be separate is inline) rather than anchor 4's appropriately split layout. | 3 / 5 |
Total | 14 / 20 Passed |