The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions (Step 1) direct subagents to conduct qualitative investigations by reading source code, documentation, and configurations from the repository being audited. This processes untrusted data which could contain malicious instructions designed to manipulate the subagents' analysis.
Ingestion points: The entire target repository, including files like README.md, AGENTS.md, and source code, serves as the ingestion point.
Boundary markers: The skill lacks explicit instructions or markers for subagents to ignore potential instructions embedded within the audited content.
Capability inventory: Subagents have the capability to read any file within the provided repository path. The orchestrator synthesizes these findings into a report stored locally or in a knowledge base.
Sanitization: There is no mention of sanitizing the content retrieved from the audited codebase.
[COMMAND_EXECUTION]: The workflow involves executing scripts/readiness.sh, a local shell script that uses tools like find and grep to perform an initial inventory of the codebase.
[DATA_EXFILTRATION]: The skill is designed to access sensitive data paths, including searching for committed secrets (Pillar 7) and checking environment variables and templates (Pillar 5). While this access is fundamental to the skill's purpose of auditing security and environment readiness, it involves access to sensitive project files.

agent-readiness