Content
81%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.
This is a strong operational skill: fully executable commands with exact flags and JSON field sets, explicitly sequenced workflows with genuine validation gates (repo gate, CI-green-before-land, clean-worktree-before-next-item, blocked-state reporting), and a correctly referenced one-level-deep bundle script. Its main weakness is conciseness — the scope rule, owner defaults, and URL-first rules are each stated multiple times across the ~310 lines.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — almost no space is spent explaining concepts Claude already knows, and commands are given as runnable blocks rather than prose. However, at ~310 lines it repeats itself: the scope rule appears in '## Scope Rule' and again at the end of '## Autonomous Work Mode' ('Autonomous work is still bounded by scope...'), URL-first output rules and owner defaults ('steipete, openclaw', broad/all/everything triggers) each appear twice, and environment-specific fallback paths pad several sections. Not 4 because the repetition is more than 'minor instances'; not 2 because nothing here is filler or tutorial-style padding — nearly every line is instruction. | 3 / 5 |
Actionability | Guidance is copy-paste ready throughout: a working `repobar_cmd` shim with binary/SwiftPM fallbacks, exact `gh issue/pr list/view` invocations with the precise `--json` field sets, `gh pr diff --patch`, ready-to-use `repobar_cmd repos` invocations with flags, a jq summary pipeline, a concrete trust-output template, and two literal output-shape templates. Concrete examples cover the common cases (current-project triage, broad scan, autonomous work). Not 4 because there are no real gaps — even edge handling (missing repo remote fallback via `git remote get-url origin` sed parsing) is fully specified. | 5 / 5 |
Workflow Clarity | Multi-step processes are explicitly sequenced with real validation checkpoints and feedback loops: the '## Local Repo Gate' (branch must be main, pull --ff-only succeeds, worktree clean, else stop and ask), the 7-step autonomous loop with 'Verify locally and live end-to-end', 'Ensure CI is green... verify a clean worktree before selecting the next autonomous item', the numbered decide-if-autonomous go/ask-first rules, and blocked-state reporting with exact contents. Destructive/batch operations (comment, close, merge, rerun, patch) are guarded by 'Only comment, close, merge, rerun, or patch with strong evidence' and proof-before-done rules. Not 4 because checkpoints are explicit and comprehensive, not merely 'most present'. | 5 / 5 |
Progressive Disclosure | Structure is good: clear `##` sections in logical order (Setup, Local Repo Gate, Scope Rule, Triage Output, Autonomous Work Mode, Trust Signals, Item Evaluation, Fast Queue Map, Detail Pass, Triage Heuristics, Output Shape), and the one bundle file — `scripts/github-activity.sh` — exists on disk and is correctly referenced one level deep with its exact invocation ('skills/github-project-triage/scripts/github-activity.sh --repo <owner/repo> --global <login>') plus a fallback path. Not 5 because a ~310-line SKILL.md keeps several sizable, separable sub-workflows inline (the full broad-scan RepoBar command catalog in 'Fast Queue Map', the trust-signal tooling, the item-evaluation rubric) that would fit naturally in reference files; not 3 because what is inline is cohesive, well-signaled, and the existing reference is clean. | 4 / 5 |
Total | 17 / 20 Passed |