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.
The body is an exceptionally actionable, well-sequenced operational manual with real validation gates and feedback loops for destructive operations, backed by a genuine one-level reference bundle. Its weaknesses are repetition of a few critical rules across sections and a handful of bundle files that the entry point never points to.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with skill-specific operational detail and wastes no tokens explaining concepts Claude already knows, but it is noticeably repetitive: the PeopleConnect delete-wipes-suppression rule, the CAPTCHA policy, and "subagent reports are self-reports - the parent re-verifies" each appear two or three times across Prerequisites, Batch operation, Procedure, and Pitfalls. That duplication could be collapsed by pointing once to references/methods.md, fitting anchor 3 ("mostly efficient but... could be tightened") better than 4. | 3 / 5 |
Actionability | Guidance is fully executable and copy-paste ready: the Quick Reference table gives exact `$PDD` commands with flags, `record` invocations include the evidence JSON shape, the CDP launch command is spelled out (`chrome --remote-debugging-port=9222 --user-data-dir="$HOME/.hermes/chrome-debug"`), and a dry-run verification snippet is provided. This matches anchor 5 ("fully executable; copy-paste ready code or commands; specific examples cover the common cases"). | 5 / 5 |
Workflow Clarity | The 8-step Procedure is an explicitly sequenced loop with strong validation checkpoints and feedback loops: `confirmed_removed` only after a verifying re-scan, `doctor` readiness before starting, blocked cases re-scanned and re-queued through the stealth-browser pass, and Pitfalls prescribing error recovery (record `blocked`, flag the broker file for re-verification). Destructive/batch operations are well guarded, matching anchor 5. | 5 / 5 |
Progressive Disclosure | Good structure: SKILL.md is an overview with clearly signaled, one-level-deep references ("references/methods.md -> 'Browser backends: scan vs execute'", site-playbooks.md, brokers/<id>.json, legal/drop.md) that all exist in the bundle. Minor gaps keep it below 5: references/state-machine.md, references/legal/ccpa.md, and references/legal/gdpr.md are never referenced from SKILL.md (state-machine.md is orphaned entirely), and sizable inline detail (CDP setup, per-method opt-out lanes, per-site exceptions) belongs in the reference files. | 4 / 5 |
Total | 17 / 20 Passed |