The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files and remote task trackers to generate documentation and rules. This creates an attack surface where instructions embedded in the analyzed content could manipulate the agent's drafting process.
Ingestion points: Project source code, existing convention files (conventions.md), requirement documents, and remote backlogs referenced in AGENTS.md.
Boundary markers: The skill instructions do not specify the use of delimiters or explicit safety prompts to ensure that the agent ignores instructions found within the processed files.
Capability inventory: The skill is authorized to modify critical steering files including AGENTS.md, which controls agent behavior, as well as conventions.md, requirements.md, and decision records.
Sanitization: No mechanisms are defined for sanitizing or escaping content retrieved from external or local environment sources before it is interpreted by the agent.

codify