The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process raw, untrusted data from various external sources such as customer quotes, social mentions, and reviews.\n- Ingestion points: Untrusted text is captured from external sources and entered into the agent context via the feedback-capture process defined in SKILL.md.\n- Boundary markers: The skill instructions suggest using markdown blockquotes ('>') for raw feedback in the output template, which provides some structural separation but is not a robust security boundary against adversarial instructions within the text.\n- Capability inventory: The agent has the capability to write files to the local filesystem ('context/feedback/'), update JSON indices, and execute local shell commands ('python tools/allocate-id.py').\n- Sanitization: There is no requirement or guidance for sanitizing the feedback input before it is analyzed by the agent or used to generate arguments for the ID allocation script.\n- [COMMAND_EXECUTION]: The skill requires the execution of a local utility script 'python tools/allocate-id.py'.\n- The skill instructs the agent to run this script to obtain unique IDs, passing a '--note' argument derived from the feedback summary. Passing arguments derived from untrusted input without explicit validation presents a minor risk of command argument manipulation.

feedback-capture