The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Node.js stack implementation in assets/stack_node.py includes a 'precise' discovery path that executes the typescript library directly from the target repository's node_modules folder. This is performed using node -e with a script that requires the local library, potentially allowing a malicious repository to execute arbitrary code during the analysis phase. While this is documented as an optional feature, it represents a significant attack surface for untrusted codebases.\n- [COMMAND_EXECUTION]: The skill relies on executing external system commands using subprocess.run in assets/rank_risk.py and assets/stack_node.py. Specifically, it invokes git for churn analysis and node for source code discovery. This behavior is inherent to the tool's purpose but involves spawning child processes using paths derived from the analyzed repository.\n- [INDIRECT_PROMPT_INJECTION]: The skill acts as a code generation engine that processes untrusted source code to create test suites. This creates a surface for indirect prompt injection where malicious patterns in the target codebase could influence the generated test code. The skill mandates a 'literal-emission rule' using repr() to mitigate this risk, but the ingestion of untrusted data into an automated code-writing workflow remains a low-level risk factor.

test-safety-net