The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by reading project documentation files and explicitly instructing the agent to "trust it over the chat" during session resumes. This instruction could lead the agent to follow malicious or adversarial instructions embedded within a doc file rather than its primary guidelines.\n
Ingestion points: The agent reads the local template.md and project-specific documentation files located in the docs/ directory to establish context and state.\n
Boundary markers: The instructions do not define clear delimiters or specify that the content of the documentation files should be treated exclusively as data, which may lead to instruction confusion.\n
Capability inventory: The agent has the capability to write files to the docs/ folder and create or modify the AGENTS.md file in the project root to enforce rules.\n
Sanitization: There is no evidence of sanitization, filtering, or validation of the content read from the documentation files before the agent processes it.

project-doc