The Agent Skills Directory

[PROMPT_INJECTION]: The skill establishes an autonomous loop that depends on ingesting state and history from local files, creating a vulnerability surface for indirect prompt injection.\n- Ingestion points: The agent reads the controller document and the append-only worklog file (DOC_PATH and WORKLOG_PATH) at the start of every iteration to determine its strategy.\n- Boundary markers: No delimiters or safety instructions are defined to separate the data in the worklog from the instructions governing the agent's behavior.\n- Capability inventory: During the 'Iterate' phase, the agent is empowered to perform code changes and run tests or 'signals' to evaluate its progress.\n- Sanitization: The skill lacks any mechanism for validating or sanitizing the content retrieved from the external files before the agent processes it.

goal-loop