Content
88%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.
A dense, highly actionable playbook with an excellent validated workflow from scoping through crash triage and evidence preservation. Its weaknesses are minor: a broken inline comment in one afl-fuzz example, an undefined 'validate_workspace_finding' tool reference, and a monolithic structure where reference tables could be split into bundle files.
Suggestions
Fix the afl-fuzz single-core example: move the '# timeout per testcase (ms)' comment above the command or drop it — as written, the comment after the backslash continuation breaks the shell command.
Define or point to 'validate_workspace_finding' (section 10) — a script path, tool name, or reference file — since it is currently invoked with no explanation of how to run it.
Move stable lookup material (Tools Quick Reference table, ASan bug-class/exploitability tables) into one-level-deep reference files under references/ to keep SKILL.md a lean overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely operational content — commands, tables, and harness code — with no explanation of concepts Claude already knows; even the prose asides carry non-obvious value ('Do NOT combine MSan with ASan — they conflict', '5–50 seeds beats 5000 random bytes'). Every token earns its place, matching the lean anchor. | 5 / 5 |
Actionability | Build, run, and triage commands plus complete C harness skeletons are largely copy-paste ready, but there are minor gaps: the inline '# timeout per testcase (ms)' comment after a backslash continuation in the afl-fuzz command is broken shell syntax, and 'validate_workspace_finding' in section 10 is invoked without being defined or located anywhere in the skill. | 4 / 5 |
Workflow Clarity | A clean ten-step sequence (scoping → fuzzer selection → instrumented build → harness → corpus → run → triage → exploitability → persist evidence) with explicit validation checkpoints: crash reproduction commands, afl-tmin minimization, backtrace-hash dedup, and a positive-signal/negative-control validation gate before reporting a finding. This matches the anchor with explicit validation steps and feedback loops. | 5 / 5 |
Progressive Disclosure | The single ~260-line file is well-sectioned and self-contained (no bundle files exist to reference), but content such as the Tools Quick Reference table, the ASan bug-class table, and the exploitability rubric is inline material that would fit one-level-deep reference files — good structure with minor organization gaps rather than the well-signaled-references anchor. | 4 / 5 |
Total | 18 / 20 Passed |