llms.txt · diff

git:20260531.9a8dfe7 to git:20260718.a80684f

10 added, 6 removed. Audit A to A.

# little-canary

Prompt injection detection library for LLM apps and agents.

Use this repo when you need to:

- screen inbound untrusted text before it reaches a main model

- combine structural pattern checks with sacrificial-canary behavior checks

- - return a simple decision: block, flag, or pass

+ - return a routing decision plus explicit behavioral-coverage state

Primary entry points:

- Python API via `SecurityPipeline`

+ - explicit replay-admission and live mechanism gates via `little-canary demo --replay|--live`

- local server via `little-canary serve`

Expected output:

- - verdict object with safety, action, summary, and optional advisory text

+ - verdict object with safety, degradation, canary/analysis status, summary, risk, and optional advisory text

Do not use this repo as:

- a guarantee that prompt injection is impossible

- a replacement for runtime containment controls

- a benchmark suite

- Key success condition:

- - the same configured pipeline and mocked backend produce the same verdict behavior across runs

+ Evidence rules:

+ - `REPLAY` verifies analyzer behavior only when admitted response bytes are packaged; otherwise it exits unavailable

+ - replay does not call a model, and unavailable replay is never evidence that input is safe

+ - `LIVE` binds the observed result to one backend, model digest, runtime, and configuration

+ - `MOCK` and `STATIC_ONLY` validate logic but do not prove live integration

+ - remote backends receive raw input; a configured judge receives raw input plus canary output

## About Hermes Labs

- Hermes Labs is an independent AI-reliability lab building open-source tools that catch silent failure modes in production AI. More at https://hermes-labs.ai.

+ Hermes Labs is an independent AI-reliability lab building open-source tools for inspecting silent failure modes. More at https://hermes-labs.ai.