Content
73%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 teaches a genuinely actionable, well-sequenced methodology with validation checkpoints and a feedback loop, and it is mostly token-efficient. Its one serious defect is progressive disclosure: it references METHODOLOGY.md and an entire resources/ tree of CodeQL/Semgrep templates that are absent from the bundle, leaving the skill's most concrete deliverables unreachable.
Suggestions
Add the referenced bundle files — METHODOLOGY.md and the resources/ templates (codeql/*.ql, semgrep/*.yaml, variant-report-template.md) — to the skill directory, or remove the references if those materials are not shipped.
Inline one small worked example of an exact-match pattern and its generalized Semgrep or CodeQL form, so the core generalize-and-verify loop is executable even when the template files are unavailable.
Tighten the 'Critical Pitfalls' section by collapsing the repeated Example/Mitigation boilerplate into a compact table, and drop the 'You are a variant analysis expert' role-play opener.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense and efficient — decision tables for abstraction points and tool selection, a numbered generalization loop, and concrete stop conditions ('Stop when false positive rate exceeds ~50%'). Minor trimming opportunities exist: the role-play opener ('You are a variant analysis expert. Your role is to...') and the four-fold repeated Example/Mitigation boilerplate in the pitfalls section add tokens without proportionate value. This matches anchor 4 ('efficient; minor instances of over-explanation') rather than anchor 5's every-token-earns-its-place. | 4 / 5 |
Actionability | Provides mostly concrete, executable guidance: an exact-match ripgrep command, a verification criterion ('Does it match exactly ONE location (the original)?'), a specific FP-rate threshold, and an element-by-element abstraction decision table. The gap keeping it below anchor 5 is that the sole code block is a placeholder ('rg -n "exact_vulnerable_code_here"') and the real CodeQL/Semgrep query templates are deferred to external files, so the body itself lacks a copy-paste-ready worked example. It clearly exceeds anchor 3's pseudocode level. | 4 / 5 |
Workflow Clarity | The five-step process is clearly sequenced with explicit validation checkpoints and a genuine feedback loop: verify the initial pattern matches exactly one location, review ALL new matches after each single-element generalization, classify true/false positives, and 'If FP rate too high, revert and try different abstraction' — an explicit error-recovery loop with a defined stop condition. This matches anchor 5 ('clear sequence with explicit validation steps; feedback loops for error recovery'). The destructive/batch-operation cap does not apply since pattern searches are read-only. | 5 / 5 |
Progressive Disclosure | References are clearly signaled and one level deep ('For deeper strategic guidance, see METHODOLOGY.md'; 'resources/codeql/': python.ql, javascript.ql, etc.), but none of these files exist in the bundle — there is no METHODOLOGY.md and no resources/ directory, nor any references/, scripts/, or assets/ folders. Navigation to the deeper content fails entirely, which is a worse structural failure than the unclear signaling of anchor 3; it matches anchor 2's 'minimal structure' in effect, since the disclosure promise is unfulfilled and the listed templates (a key part of the skill's value) are inaccessible. | 2 / 5 |
Total | 15 / 20 Passed |