Content
66%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 well-sequenced orchestrator: explicit phase gates, a machine-checkable completion invariant, and comprehensive per-class reference coverage make the workflow unambiguous and highly actionable. Its main defect is token efficiency — a large duplicated JS-recon block and six-fold repetition of the no-skip doctrine inflate the body well beyond what the reference files already carry.
Suggestions
Delete the ~120-line "MANDATORY JS RECON GATE" block and replace it with a two-line pointer to references/js-mining.md plus the enforcement rule ("if any web target is in scope, complete the js-mining.md pipeline before recon ends") — the detailed commands already exist in that file.
Consolidate the no-skip doctrine, currently restated in Rule 0, Rule 1, the rationalizations table, the red-flag self-check, the ledger section, and the completion criteria, into a single section (e.g. keep the rules and the ledger invariant; drop the restated motivational variants).
Trim rhetorical padding such as "Hunters who skip JS lose the P1s", "is running a different, worse skill", and "budget the time; this step is supposed to be slow" — the rules already command the behavior.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~120-line injected "MANDATORY JS RECON GATE" block duplicates content that already lives in references/js-mining.md (same crawlers, download/beautify/sourcemap commands, and secret patterns), and the no-skip doctrine is restated six times (Rule 0, Rule 1, the rationalizations table, the red flags, the ledger section, and the completion criteria), plus motivational padding ("Hunters who skip JS lose the P1s", "this step is supposed to be slow") — several unnecessary padded sections, matching anchor 2 rather than 3 where trimming would be incidental. | 2 / 5 |
Actionability | The body provides copy-paste-ready bash for recon and JS mining (subfinder/httpx/waybackurls/gau/katana pipelines, beautify loop, sourcemap recovery), a machine-checkable ledger invariant (jq query), and a per-class checklist mapping each trigger condition (e.g. "every URL-fetch/webhook/import/preview/media feature → ssrf.md") to a real reference file. It falls short of anchor 5 only because several phases remain high-level directives ("Build the threat model", "walk the entire application") whose execution detail is delegated to references. | 4 / 5 |
Workflow Clarity | Seven numbered phases in strict order, a start protocol, a verification gate between every phase, explicit validation gates (7-Question Gate, 4 pre-submission gates), and concrete feedback loops ("If any is false, the run is not complete — return to the ledger and finish it") with a red-flag self-check — clear sequence with explicit validation, error-recovery loops, and checklists, matching anchor 5. | 5 / 5 |
Progressive Disclosure | All 28 referenced files exist, are exactly one level deep, and are well signaled through both a reference-map table and contextual inline links; however, the inline JS-recon gate block is technique content that clearly belongs in js-mining.md (its own comment marks it as injected boilerplate), so content placement has a notable organization gap — anchor 4 rather than 5. | 4 / 5 |
Total | 15 / 20 Passed |