mirror of
https://github.com/trailofbits/skills.git
synced 2026-09-14 14:28:48 +08:00
Graham Sutherland
286b2fadb4
When a fuzz run reveals a source disclosure mechanism (LFI, path traversal, arbitrary file download, XXE, etc.), the skill now instructs Claude to: - Compile known filenames from stack traces, include() warnings, spec paths, and HTML attributes seen in fuzz results - Gate risky retrieval mechanisms (LOAD_FILE, SSRF) behind user confirmation; prohibit INTO OUTFILE unconditionally in line with the existing write-operation SQL safety constraint - Use urllib.request (not curl) to fetch raw source - Mine retrieved source for SQL sinks, hidden parameters, secondary sinks (exec/unserialize/eval), auth logic, and crypto weaknesses - Re-fuzz any newly discovered parameters before writing the report - Enforce that all follow-up fuzzing and exploitation complies with the safety constraints in fuzz-strategies.md Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>