The Agent Skills Directory

[SAFE]: The skill operates as a knowledge-management tool within a restricted scope. It instructs the agent to synthesize infrastructure information from existing project repositories and local configuration files while maintaining clear boundaries.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection, as its primary function is to ingest and synthesize data from external project repositories, manifest files, and CI workflows.
Ingestion points: Data is sourced from external repositories, package.json manifests, Docker/compose files, and CI workflow definitions.
Boundary markers: The skill enforces the use of a specific reader tool (bun run context:map) to access the synthesized map, providing a structured interface for the data.
Capability inventory: The skill's write capabilities are strictly limited to the references/ subdirectory, and it utilizes a diagrams capability for architectural visualization.
Sanitization: The instructions include a mandatory rule against storing credentials, tokens, or passwords in the synthesized map, directing the agent to reference .env keys instead.
[DATA_EXPOSURE]: The skill correctly identifies .env and .agents/project.yaml as the authoritative sources for secrets and environment URLs, respectively, and provides explicit instructions to prevent these sensitive values from being copied into its internal maps or evidence logs.

infra-context