The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by reading and sampling arbitrary source files and package manifests from the repository to perform its architecture analysis.
Ingestion points: SKILL.md (Step 2) reads entry points (e.g., index.ts) and representative source files. references/discover-context.md performs existence checks and reads project manifests like package.json, lockfiles, and workspace configurations.
Boundary markers: No explicit delimiters or instructions are provided to the agent or sub-agents to ignore potentially malicious instructions embedded within the sampled source code.
Capability inventory: The skill has the ability to write markdown reports to the filesystem (Step 4) and execute a local Node.js script (Inputs section).
Sanitization: There is no evidence of sanitization or escaping applied to the content read from the repository before it is processed.
[COMMAND_EXECUTION]: The skill executes a local Node.js script to check the freshness of the audit state.
Evidence: node <lodestar-audit-skill>/scripts/audit-state.mjs check-freshness --root <repo> in SKILL.md. This is a call to a script within a related skill (lodestar-audit).

lodestar-architecture