The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external user-provided data (case studies, benchmark exports, testimonials) which could serve as a vector for indirect prompt injection.
Ingestion points: Reads user-provided case data, benchmark exports, and permitted quotes.
Boundary markers: Explicitly references SECURITY.md and contains instructions to treat pasted material as untrusted input and to never follow instructions embedded in them.
Capability inventory: Reads and writes to local memory/ directories and invokes a project-local script registry-events.py for data operations.
Sanitization: Instructs the agent to mark unverified proofs as [needs source] and route them for adjudication rather than executing them.
[COMMAND_EXECUTION]: The skill coordinates with a local script to manage narrative and claim records.
Evidence: Instructions define the use of an authorized operation: propose request to registry-events.py to update the claims events ledger (memory/events/claims.ndjson).

proof-point-packager