The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of "requests" or "faults" and is instructed to preserve and incorporate the user's original words during refinement. This input is processed in the same context where the agent is authorized to execute local shell and Python scripts, creating a surface for potential injection.
Ingestion points: Processes user-provided requests and fault reports.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the prompt instructions.
Capability inventory: Shell script execution (sh .agents/tools/plan-refresh.sh, records/scripts/mark-ready.sh) and Python script execution (quality.py).
Sanitization: No sanitization or validation logic is specified for the ingested user input.
[COMMAND_EXECUTION]: The skill's operational logic depends on the execution of multiple local scripts and CLI tools.
Evidence: Explicit instructions to run sh .agents/tools/plan-refresh.sh, quality.py reach --paths, and records/scripts/mark-ready.sh using the agent's environment capabilities.

shape