bamdadd/leakgauge

Test the hijack-vs-leak gap at the scorer level with multi-secret traces about 21 hours ago

good first issue type: test area: agent-safety size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Scorer: strip zero-width/format characters before canary matching 1 day ago

good first issue type: test area: agent-safety size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Add a metrics glossary with a worked synthetic trace 1 day ago

good first issue type: docs area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Reject --max-steps <= 0 (silent empty run reads as a clean pass) 16 days ago

good first issue mentored type: bug area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Clean error (not a traceback) for bad paths in 'leakgauge report' 16 days ago

good first issue mentored type: bug area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Add direct unit tests for the _try_b64 canary decoder 16 days ago

good first issue mentored type: test area: agent-safety size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Harden _try_hex odd-length handling: dropping the last nibble can trust a misaligned decode 19 days ago

good first issue type: bug area: agent-safety size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Warn upfront when a real model has no price entry 20 days ago

good first issue type: feature area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Validate `--k` and `--seed` in the CLI 20 days ago

good first issue type: feature area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Document the `report` subcommand in `--help` 20 days ago

good first issue type: docs area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Add a worked "author a new case" walkthrough to CONTRIBUTING 20 days ago

good first issue type: docs area: infra size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Pin scorer encoding-normalization behavior on adversarial edge cases 20 days ago

good first issue type: test area: agent-safety size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Show each model's rank number and rank shift on the reorder plot 20 days ago

good first issue type: feature area: llm size: S

bamdadd/leakgauge

0

A leakage-verified prompt-injection benchmark for measuring and improving agent robustness (defensive).

Python

Open Issues Need Help