Content
85%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.
An unusually actionable, well-structured instruction skill: concrete commands, exact paths, an explicit validation step, and clean one-level reference files. Its one real weakness is redundancy -- the Anti-Patterns section largely restates the Rules and inlines time-sensitive repo state, adding significant token cost without new guidance.
Suggestions
Consolidate Rules and Anti-Patterns: four of the five anti-patterns (never-mutate, empty-funnel fall-through, no-invented-value-ranking, freshness check) restate Rules 1-4. Merge each rule with its anti-pattern, or move the BAD/GOOD/Consequence examples into references/backlog-triage-and-ranking.md.
Remove time-sensitive repo state from the body ('all 15 filed follow-ups are currently DONE', and the '38 open register rows, 7 open tech-debt rows' counts in the GOOD example) or relocate it to references/worked-example.md, which already flags its counts as hypothetical -- stale counts erode trust and cost tokens.
Tighten the Mindset section, which substantially overlaps Rules 2 and 5 and the first anti-pattern; a single sentence pointing at the fall-through rule would preserve the guidance at a fraction of the length.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient project-specific guidance, but four of the five Anti-Patterns restate Rules 1-4 at length (e.g. 'NEVER promote, close, or edit a row or follow-up from inside this skill' vs Rule 1 'This skill never mutates the register, the tech-debt list, or a follow-up's status'), the Mindset section overlaps Rules 2 and 5, and time-sensitive repo state ('where all 15 filed follow-ups are currently DONE') is inlined. It sits between the 'noticeably verbose' and 'mostly efficient' anchors -- the BAD/GOOD/Consequence framing adds some value beyond the duplication, but the file could be roughly a third shorter. | 3 / 5 |
Actionability | Every workflow step is backed by a concrete executable command ('find .context -name '*.md' | wc -l', 'grep -l 'status: ACTIVE' .context/follow-ups/*.md', 'grep -c '| Open ' docs/RISK_REGISTER.md docs/TECH_DEBT.md'), exact file paths in Prerequisites, defined entry statuses, and a worked-example template for the output format -- fully actionable for an instruction-only skill. | 5 / 5 |
Workflow Clarity | Five clearly sequenced steps with an explicit validation checkpoint (step 1's index-freshness comparison with a regenerate-on-mismatch recovery action via context-index), explicit ordering rules for the recommendation (Rule 5), and a hard stop with handoff at step 5. The skill is read-only so the destructive-operation cap does not apply; only the vague 'diverges meaningfully' threshold is a soft spot. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview that links two real one-level-deep references (references/backlog-triage-and-ranking.md and references/worked-example.md, both verified present and substantive) at their point of use in the Workflow, plus a References table with a 'When to Use' column for discovery. Neither reference file nests further content references, so navigation stays flat. | 5 / 5 |
Total | 18 / 20 Passed |