claude-leak-bench

1 Commit

1 Branch

0 Tags

111 KiB

Commit Graph

Author	SHA1	Message	Date
Nicolas Massé	c4f4171302	Initial end-to-end secret leak bench for Claude Code pytest suite that runs claude -p behind a recording API proxy and checks whether a canary secret reaches the wire, the transcript or the answer. Cases: cat, Read, env, unlabeled random value, known value, tail -f in foreground/background, Monitor. Subjects: bare Claude Code (control) and the redacted PostToolUse hook. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	4 weeks ago

Author

SHA1

Message

Date

Nicolas Massé

c4f4171302

Initial end-to-end secret leak bench for Claude Code

pytest suite that runs claude -p behind a recording API proxy and checks
whether a canary secret reaches the wire, the transcript or the answer.
Cases: cat, Read, env, unlabeled random value, known value, tail -f in
foreground/background, Monitor. Subjects: bare Claude Code (control) and
the redacted PostToolUse hook.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

4 weeks ago

1 Commits (c4f4171302ef3887bff0d3fcd70e22c60bee78fa) All Branches Search

1 Commits (c4f4171302ef3887bff0d3fcd70e22c60bee78fa)

All Branches