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.
This is a strong operational skill body: a clearly phased workflow with genuine validation checkpoints, feedback loops, and safety gates, expressed in concrete commands rather than abstraction. The main defects are packaging-level — three referenced template files do not exist in the bundle, and the delegation tool syntax is assumed rather than shown.
Suggestions
Ship the missing templates/ files (authorization.md, pentest-report.md, exploitation-queue.json) or remove the references and inline their schemas — Phases 0, 3, and 5 currently point at files that do not exist in the bundle.
Include one concrete delegate_task invocation example (arguments, expected subagent output) so the Phase 3 delegation is copy-paste ready.
Trim meta-commentary such as the "Adapted from Shannon's pipeline" attribution and tool-rationale asides into a single short credits/notes line to tighten token budget.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — it assumes Claude's competence (never defines SQLi, XSS, or nmap) and every section carries procedure rather than background. Minor trim opportunities exist (the "Adapted from Shannon's pipeline" attribution line and rationale sentences like the `-T3` IDS/IPS justification), keeping it below anchor 5's "every token earns its place". | 4 / 5 |
Actionability | It provides concrete executable commands ("nmap -sT -T3 --top-ports 100 -oN evidence/nmap.txt $TARGET", the whatweb/curl lines, the browser_exec JS snippet) and specific witness payloads ("' AND 1=1--", "<svg/onload=console.log(\"HERMES-PENTEST-XSS\")>") with exact output paths. Not anchor 5 because delegate_task/browser_exec call syntax is assumed rather than shown, and three referenced templates/ files are absent from the bundle, leaving the Phase 0/3/5 artifacts under-specified. | 4 / 5 |
Workflow Clarity | Phases 0-5 are explicitly sequenced with validation checkpoints throughout: the authorization gate ("Wait for explicit `authorized` response"), pre-send checks, witness-payload verification, and bypass exhaustion before any false-positive verdict. Feedback loops for error recovery (STOP on off-scope redirects, back off on 503/429 storms) and stop-condition checklists are present, and destructive operations have approval gates, so the destructive/batch cap at 3 does not apply. | 5 / 5 |
Progressive Disclosure | The overview body stays lean and pushes per-class detail to four real one-level-deep references (scope-enforcement.md, vuln-taxonomy.md, exploitation-techniques.md, bypass-techniques.md) plus scripts/recon-scan.sh, each clearly mapped in the "Further Reading" section. Not anchor 5 because three referenced paths (templates/authorization.md, templates/pentest-report.md, templates/exploitation-queue.json) are missing from the bundle entirely, leaving dangling navigation — though their content is partially inlined, so the damage is bounded. | 4 / 5 |
Total | 17 / 20 Passed |