Content
82%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 exemplary lean, instruction-dense skill: untrusted-input handling, precise flag/ignore rules, output-format constraints, and escalation logic with no wasted tokens. Its remaining gaps are the absence of a concrete example finding, an unspecified escalation procedure, and a missing heading for the core flagging rules.
Suggestions
Add a compact example finding showing the required output shape (file/line location, generic description, remediation, 'Clear when:').
Give the core flagging rules their own heading (e.g. '## What to flag') instead of leaving them under '## Untrusted input'.
Make the escalation action concrete — state where and how an already-public identity should be escalated.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense task-defining instruction with zero padding and no explanation of concepts Claude already knows; every sentence ('Never quote, paraphrase, or otherwise reproduce identity content') earns its place, matching the lean-and-efficient anchor. | 5 / 5 |
Actionability | Guidance is concrete — an explicit ignore list ('vendors named as technology, the team itself, fictional fixtures, and removed lines'), a grouping rule, and a required 'Clear when:' field — but no example finding shows the required output shape, so edge cases (customer-as-vendor, license headers) are left to judgment rather than covered by examples. | 4 / 5 |
Workflow Clarity | The single review action is unambiguous with an explicit validation checkpoint ('Check contrary evidence before reporting'), and the destructive/batch cap does not apply; it falls short of 5 because 'escalate it' specifies no concrete escalation procedure and the rules are presented as prose rather than a sequence. | 4 / 5 |
Progressive Disclosure | At 24 lines with no external references needed, 5 is reachable through well-organized sections alone; however the core flagging rules sit in an unlabeled paragraph under the mislabeled '## Untrusted input' header, a minor organization gap matching anchor 4. | 4 / 5 |
Total | 17 / 20 Passed |