The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow requires the agent to generate and run scripts locally to calculate feasibility metrics. \n- Evidence: Workflow step 3.5 (Gate G2) mandates the creation of a ≤ 30-line PoC script in methods/Qx/poc/ and its immediate execution. \n- Risk: Dynamically generating code from external inputs (problem_parse.json, related_paper_analysis.md) and executing it allows for potential system exploitation if the input data contains malicious instructions designed to bypass the restricted library list (numpy, scipy, pandas, sklearn). \n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection from preceding pipeline artifacts. \n- Ingestion points: The skill reads workspace/problem/problem-parser/problem_parse.json and workspace/papers/related_paper_analysis.md, which contain data derived from external sources. \n- Boundary markers: There are no explicit delimiters or protective instructions provided to prevent the agent from interpreting data within these files as command overrides. \n- Capability inventory: The agent possesses the capability to write and execute scripts on the local filesystem. \n- Sanitization: The skill lacks validation or sanitization logic to ensure the "Math idea" or other fields used to generate PoC scripts do not contain harmful code or instructions.

method-selector