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.
A tightly engineered instruction-only skill: clear scope boundaries, receipt-bound close semantics, explicit labeling rules (Measured / User-provided / Estimated), and well-signaled external references. Its weaknesses are density and repetition rather than missing substance, plus the absence of any example connector invocation for a skill whose whole job is polling via four scripts.
Suggestions
Add one example invocation per connector in Data Sources (e.g., `python scripts/connectors/hn.py <story-url>` and its output shape) so the polling steps are copy-paste executable instead of by-filename reference.
State the launch-day-conductor / retro-analyzer routing once in the Scope guard and reference it from steps 4, 7, and 8 instead of restating the full handoff each time.
Split the 90-word single-sentence steps (especially step 1) into short sub-bullets for mode/receipt/targets so the receipt rules scan without re-reading.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows and every section is functional, but it is dense to a fault — steps 1 and the opening paragraph are 90-100-word single sentences — and cross-skill handoffs repeat (launch-day-conductor appears in the Scope guard, step 4, Reference Materials, and Next Best Skill; launch-retro-analyzer appears four times), so it is not 'every token earns its place'. Not 3: there is no truly unnecessary explanation, only tightening opportunities; not 5: the repetition and compound-sentence density exceed the lean anchor. | 4 / 5 |
Actionability | Guidance is concrete throughout — named connector scripts ('scripts/connectors/hn.py (keyless Algolia + Firebase — rank, points, comments)'), explicit cadence limits ('gdelt.py needs ≥5s between calls; keep HN/PH polling to a few reads per hour'), a save-path template ('memory/launch/launch-monitor/YYYY-MM-DD-<topic>.md'), and alert fields (metric, threshold, KPI target). Not 5: for a script-driven skill, no connector is ever shown being invoked (no example call to hn.py or producthunt.py with arguments), leaving the polling steps one level short of copy-paste executable; not 3: the guidance is specific and directive, not pseudocode or high-level hints. | 4 / 5 |
Workflow Clarity | Eight numbered steps are clearly sequenced (confirm mode → verify instrumentation → set cadence → watch signals → snapshot → read retention → alert → close), with explicit validation (per-surface pass/fail in step 2, 'No targets means NEEDS_INPUT' in step 1, a 'Done when' checklist) and a real feedback loop: close only when 'every required current-manifest action has a terminal matching receipt', otherwise 'emit window_status: OPEN with the missing receipt IDs' and re-check. Not 4: checkpoints and error-recovery loops are explicit rather than merely present; this is a read/watch skill, so the destructive/batch cap does not apply. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview and detail is offloaded to clearly signaled, one-level-deep references, each annotated with its purpose in Reference Materials ('CONNECTORS.md — connector setup for scripts/connectors/hn.py, producthunt.py...', 'Launch Action Control — receipt-bound snapshots, required-action joins, and close-window semantics'). No bundle files ship with this skill, and the inline content (the 8-step procedure, Skill Contract) is core operating material that belongs in SKILL.md, not material that should have been split out. Not 4: navigation is easy, references are labeled, and nothing inlined obviously belongs in a separate file. | 5 / 5 |
Total | 18 / 20 Passed |