The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and execute untrusted data in the form of reproduction packages and test harnesses provided for security analysis.
Ingestion points: The agent ingests external harness source code and planned inputs (SKILL.md).
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the ingested test harness data.
Capability inventory: The skill has the capability to execute shell commands via uv run and arbitrary reproduction harnesses (SKILL.md).
Sanitization: No sanitization of the external test harness code or inputs is described.
[COMMAND_EXECUTION]: The skill invokes various local scripts and development tools, including python, pytest, and uv, to perform build checks and verification steps (SKILL.md).
[DYNAMIC_EXECUTION]: The skill involves the runtime execution of test harnesses and verification scripts (arms.py, records.py, native_proof.py) to observe behavior changes across different code commits (SKILL.md).

prove-patch