Content
67%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 delivers a clear, gated investigation workflow with concrete, executable commands and well-structured sections. Its weaknesses are token redundancy — duplicated command blocks, scripted dialogue, and a repetitive closing line — plus slightly vague discovery guidance for log and database locations.
Suggestions
Remove the Quick Reference section's duplicated commands (git, sqlite3, ps/lsof appear verbatim in the Step 2 task prompts) or keep only the Quick Reference and strip them from the task blocks.
Replace the two scripted 'Initial Response' dialogue blocks with a short 2-3 line summary of what to ask the user in each invocation mode.
Delete the closing 'Remember:' paragraph, which restates the intro, and add guidance for handling inconclusive task results (e.g., cross-check logs against git timing before presenting a root cause).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core process guidance is lean and free of concept over-explanation, but the Quick Reference repeats commands verbatim from the Step 2 task blocks, the 'Initial Response' section scripts ~25 lines of dialogue, and the closing 'Remember:' paragraph restates the intro. Mostly efficient but could be tightened, matching anchor 3. | 3 / 5 |
Actionability | Commands like 'ls -t ./logs/*.log | head -1', 'sqlite3 {database_path} ".tables"', 'git log --oneline -10', and 'lsof -i :{port}' are copy-paste ready, and each task prompt has numbered steps plus an explicit 'Return:' contract. Minor gaps remain: log and database locations are deferred to 'check project-specific locations'/'check project config', keeping it just short of fully executable. | 4 / 5 |
Workflow Clarity | A clear three-step sequence (understand, investigate, report) with an explicit gate ('Always require problem description', Step 1 fires only after the user describes the issue) and fallback guidance ('If That Doesn't Work'). Validation gaps are minor — no instruction to cross-check task findings against each other or handle empty/inconclusive task results — matching anchor 4 rather than the explicit-feedback-loop anchor 5. | 4 / 5 |
Progressive Disclosure | No bundle files exist and none are referenced; the ~190-line body is well-organized into clear sections (Initial Response, Environment, Process Steps, Important Notes, Quick Reference) with no nested or buried references. Structure is good with minor gaps from inter-section duplication; it exceeds the under-50-line simple-skill exception, so not a 5. | 4 / 5 |
Total | 15 / 20 Passed |