The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is instructed to prioritize repository-local instruction files such as AGENTS.md and CLAUDE.md to guide its behavior and workflow intent. Because these files are part of the scanned repository and are treated as authoritative by the skill, they represent a vector for indirect prompt injection where an attacker could provide instructions that override the agent's core constraints or bias its project audit findings.
Ingestion points: The skill reads AGENTS.md, CLAUDE.md, and path-scoped instruction files (e.g., Assets/CLAUDE.md) from the local environment.
Boundary markers: Absent; the content of these files is incorporated into the agent's context without explicit delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill possesses Read, Glob, Grep, and Write capabilities, allowing it to modify project documentation and configuration files based on its analysis.
Sanitization: Absent; the skill does not perform validation or escaping of the instructions found in the repository files.

adopt