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.
A tightly written operational skill with excellent gated workflow logic, copy-paste commands, and strong validation checkpoints. Its weaknesses are external vocabulary used without pointers (rows, proceed_manifest, Phase 5/6, escalate) and slightly indirect navigation to the type-specific implement prompts.
Suggestions
Replace the glob 'assets/implement-*.md' with the four explicit links (implement-bug.md, implement-docs.md, implement-feature.md, implement-refactor.md) so navigation does not require a directory listing.
Give 'escalate' a concrete mechanism (where to post, what to include) in the Gate and ORIGIN sections, since it is the prescribed outcome for several STOP paths but has no defined action.
Add a pointer for the Return-section vocabulary ('row', 'driver inputs', 'proceed_manifest', 'Phase 5/6') to the scheduler skill or a short manifest reference, so the handoff steps are executable without assumed parent context.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and telegraphic with no padding or explanations of concepts Claude already knows — every section carries operational rules. Minor trimming is possible: the reviewer-request prohibition appears twice ('Do not request that actor as a reviewer' and 'Do not request yourself as reviewer'), and the one-issue rule is restated in Activation card, ORIGIN, and Procedure sections. This fits 'Efficient; minor instances of over-explanation that could be trimmed'. | 4 / 5 |
Actionability | Strong concrete commands throughout: the eligibility invocation ('node scripts/governance/eligibility.cjs --repo microsoft/apm --issue N --approval-url URL'), assignee commands ('gh issue edit --add-assignee @me'), ORIGIN detection ('gh api user --jq .login'), and two copy-paste card templates. Gaps keep it below 5: 'escalate' has no defined mechanism, and the Return section leans on undefined external terms ('persist the row's blocked status', 'driver inputs', 'proceed_manifest', 'Phase 5/6') without a pointer to where those live. | 4 / 5 |
Workflow Clarity | The workflow is clearly sequenced with abundant explicit validation and feedback loops: the 4-step ORIGIN resolution ladder (fail-closed to unattended), the 7-step assignment gate with immediate re-reads and a lost-race recovery path ('remove-assignee @me and STOP as blocked'), and STOP conditions on every disallowed gate state. This matches 'Clear sequence with explicit validation steps; feedback loops for error recovery' — the destructive/batch validation cap does not apply because validation is pervasive. | 5 / 5 |
Progressive Disclosure | The SKILL.md is an operational overview linking one-level-deep assets that all exist in the bundle ([assets/triage-prompt.md], [assets/fix-prompt.md], [assets/solution-pipeline-prompt.md], [assets/strategic-alignment-prompt.md]). Minor gaps: 'type-specific assets/implement-*.md' is a glob rather than naming the four files (bug/docs/feature/refactor), 'PRINCIPLES alignment' and 'pr-description-skill' are referenced without paths, and references/strategic-alignment-gate.md is present but never linked from the body. This fits 'Good structure; most content is appropriately placed; references mostly clear; minor organization gaps'. | 4 / 5 |
Total | 17 / 20 Passed |