The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and execute code from the local workspace (e.g., derivation.ipynb, pytest). If the repository being reviewed is malicious, it could contain instructions or scripts designed to exploit the agent's execution environment.
Ingestion points: Mathematical source code, derivation notebooks (derivation.ipynb), and test files within the target project.
Boundary markers: Absent. The skill instructions proceed to execute content without explicit sanitization or isolation markers.
Capability inventory: The skill can execute shell commands (pytest, jupyter nbconvert), read filesystem metadata via git, and process complex file structures.
Sanitization: No evidence of sandboxing or input validation for the scripts being executed.
[DYNAMIC_EXECUTION]: The workflow involves executing external code files found in the environment to verify mathematical correctness.
Evidence: jupyter nbconvert --execute derivation.ipynb in SKILL.md (Step 5: Proof of Work).
Evidence: pytest tests/math/ --benchmark in SKILL.md (Step 5: Proof of Work).
[COMMAND_EXECUTION]: The skill uses shell commands to synchronize context and assess the state of the repository.
Evidence: pwd && git status -sb && git diff --stat origin/main..HEAD in SKILL.md (Step 1: Context Sync).

math-review