Content
52%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 well-structured with a clear when-to-use/when-not-to-use split, explicit safety constraints, and a properly gated mutation workflow. However, its central orchestration leans on seven advisor files that are absent from the bundle, and the same orchestration is restated three times, leaving the skill only partially executable as shipped. The existing references and assets are real and well-signaled, which keeps the structure salvageable.
Suggestions
Bundle the seven `agents/*-advisor.md` files (target-pr, research, security, code-quality, impact, verdict, source-control) or replace the file references with inline guidance — every numbered workflow step currently points at files that do not exist.
Merge the three overlapping orchestration sections ("Preferred advisor orchestration", "Workflow summary", and "Advisors") into a single section, and deduplicate the "Always" list (e.g., the shared-research-packet items appear twice).
Inline the key lens criteria and decision-gate details into SKILL.md or into the existing `references/instructions.md` so the review is actionable when the runtime does not support skill-local advisors.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is operational with no concept explanations Claude already knows, but the advisor orchestration is described three times ("Preferred advisor orchestration" with per-advisor roles, the numbered "Workflow summary", and the "Advisors" section repeating the same roles), and the 18-item "Always" list contains internal duplicates ("Reuse one shared research packet..." vs "Do not re-fetch PR comments or review threads independently..."). Mostly efficient but noticeably redundant — fits 'includes some unnecessary explanation or could be tightened' rather than level 2's 'several unnecessary explanations', since all content is skill-specific rather than padded. | 3 / 5 |
Actionability | Concrete elements exist (explicit status "Awaiting user PR selection", named templates like `assets/research-template.md.hbs`, specific confirmation gates), but the primary executable path delegates its substance to `agents/research-advisor.md`, `agents/security-advisor.md`, `agents/code-quality-advisor.md`, `agents/impact-advisor.md`, and `agents/verdict-advisor.md` — none of which exist in the bundle. The inline fallback ('research first, lenses next, verdict after that') gives only high-level sequencing without the lens criteria or research detail, matching 'some concrete guidance but incomplete; missing key details' rather than level 4's 'minor gaps'. | 3 / 5 |
Workflow Clarity | The sequence is clear and well-ordered: resolve target → shared research packet → parallel lenses → verdict → conditional source-control, with explicit checkpoints for a mutation-capable skill ("Post the research checkpoint comment to the PR before any branch mutation", "Post the final verdict comment... before any approval or merge", "Never merge or approve a dependency PR without explicit user confirmation"). It is not capped at 3 because validation/verification gates are present rather than missing; it falls short of level 5 only because the detailed decision-gate mechanics live in `references/instructions.md` rather than feedback loops fully specified in the body. | 4 / 5 |
Progressive Disclosure | Scoring against the actual bundle: of the paths referenced in the body, `references/instructions.md`, `references/questions.md`, the three `assets/` files, and `assets/partials/agent-content/README.md` all exist — but all seven `agents/*-advisor.md` files that steps 1–5 of the core workflow are built on are missing from the bundle. The majority of the body's referenced paths are broken, so navigation fails for the primary execution path. This is worse than level 3 ('references present but not clearly signaled' — these are clearly signaled but nonexistent) and comparable in defect severity to level 2's structural failure. | 2 / 5 |
Total | 12 / 20 Passed |