The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external report files which could potentially contain malicious instructions intended to influence the agent's behavior.\n
Ingestion points: The skill reads report files provided by the user in Step 2 of the workflow.\n
Boundary markers: The skill does not explicitly define delimiters for untrusted content, but it requires a structured extraction process.\n
Capability inventory: The skill uses Write, Edit, and Bash tools to modify project files on a feature branch.\n
Sanitization: No specific sanitization techniques are detailed, but the skill mandates a human review phase.\n
Mitigation: The risk is significantly mitigated by a mandatory human approval gate in Step 5. The agent is required to present findings in a jargon-free, observable-change format, ensuring the user can verify the proposed actions before they are written to the file system.

plan-intake