The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files and command outputs, creating a surface for indirect prompt injection where malicious content in the codebase could influence the agent's actions.\n- Ingestion points: The skill instructions in SKILL.md direct the agent to read .agent/context/loops.md, .agent/context/project-context.md, and STOP-RULES.md, and to observe/summarize verifier command outputs.\n- Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions when reading these files or command outputs.\n- Capability inventory: The skill allows the agent to execute shell commands (verifiers) and perform file modifications across the codebase during the 'Act' and 'Observe' phases.\n- Sanitization: No explicit validation or escaping of the external data is performed before processing by the agent.

agent-loop