The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a workflow where agents are instructed to read and adhere to guidelines found within the target repository's files.
Ingestion points: The agent is prompted to read AGENTS.md (root and module-level), directory-level README.md files, and the docs/ directory for standards and conventions.
Boundary markers: No explicit boundary markers or instructions to ignore malicious embedded commands in the ingested data are provided.
Capability inventory: The skill is designed to facilitate autonomous coding tasks which typically involve file system access, terminal command execution, and code modification.
Sanitization: There is no evidence of sanitization or validation of the instructions retrieved from the repository files before the agent processes them.

harness-engineering