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.
A well-structured, highly actionable instruction skill with a clear eight-step workflow and strong completion criteria. Its main weaknesses are systematic repetition of the proxy-labeling and registry-submission rules (each stated 3–4 times) and the absence of executable connector invocation examples or output templates.
Suggestions
State the proxy-labeling red line and the registry propose-only rule once each (e.g. in Data Sources and Instructions step 7) and reference them elsewhere instead of repeating the full rule in the Scope guard, Writes, step 6, 'Done when', and Save Results sections.
Add concrete connector invocation examples (e.g. 'python scripts/connectors/hn.py "<brand query>"') and compact templates for the triaged-mention table and coverage-disclosure table so the expected output is copy-paste reproducible.
Trim the Scope guard paragraph, which restates sibling-skill disambiguation already fully covered in the description's 'Not for...' clause.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence, but key rules are stated repeatedly: the proxy-labeling red line appears four times (Scope guard, Data Sources, Instructions step 6, 'Done when') and the registry propose-only path also appears four times (Scope guard, Writes, step 7, Save Results). Mostly efficient, but systematically tightened-able through deduplication. | 3 / 5 |
Actionability | Concrete throughout — named connectors with rate limits ('gdelt.py (≥5s between calls)'), a specific spike threshold ('current day > 2× the 7-day median'), routing targets with SLAs ('bug → the user's issue tracker (within 1 business day)'), and an exact save path. Falls short of 5 because no actual connector invocation commands or triage/coverage-table templates are shown. | 4 / 5 |
Workflow Clarity | Eight numbered steps in a coherent build→sweep→triage→baseline→watchlist→disclose→submit→report sequence, with a 'Done when' checklist and gated writes ('ask before writing', propose-only events). Checkpoints are explicit, but there are no validate→fix→retry feedback loops, keeping it below 5. | 4 / 5 |
Progressive Disclosure | Good structure with a Reference Materials section giving one-level-deep, clearly signaled references with one-line descriptions (e.g. '[CONNECTORS.md] — keyless listening connector recipes and rate limits'). Not 5: all referenced files live outside this bundle and none ship in references/scripts/assets, and step 3's full routing/SLA table is inlined detail that could live in a reference file. | 4 / 5 |
Total | 15 / 20 Passed |