The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of runtime logs from the PS2Recomp execution environment.
Ingestion points: The stubs.py script reads logs from console/runs/*.log and source files from a provided repository path.
Boundary markers: The script uses specific regular expressions to parse known log formats rather than treating the entire log content as instructions.
Capability inventory: The included Python script is read-only and does not possess file-writing, network-access, or subprocess-execution capabilities.
Sanitization: Data is validated against strict regex patterns (e.g., matching PC/RA hex addresses and specific error strings) before being reported to the agent.
[SAFE]: The skill's behavior matches its stated purpose as a backend engineering aid. It adheres to best practices by maintaining read-only access to the user's repository and logs, and it includes explicit guardrails against processing or requesting copyrighted game assets.

console-backend