The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The scripts/find-polluter.sh script executes npm test on local files found via a user-defined pattern. While this involves dynamic command execution, it is the primary and documented function of the skill to identify state pollution in a test suite.
[PROMPT_INJECTION] (SAFE): The markdown instructions (e.g., in references/verification.md and SKILL.md) establish rigid procedural boundaries for the agent (such as the 'Iron Law' of verification). These are functional constraints designed to improve reliability and do not attempt to bypass the LLM's safety filters.
[DATA_EXPOSURE] (SAFE): The documentation encourages the use of stack traces and logging (references/root-cause-tracing.md) to assist in debugging. The analysis confirms these practices are restricted to standard diagnostic output and do not target sensitive system files or credentials.
[INDIRECT_PROMPT_INJECTION] (SAFE): The skill processes local test files which represents a potential attack surface. However, this is inherent to the tool's purpose of debugging existing codebases.
Ingestion points: scripts/find-polluter.sh reads and iterates over files matching a specific pattern.
Boundary markers: Absent, as it relies on standard test runner execution.
Capability inventory: The script can execute npm test, find, and ls on the local filesystem.
Sanitization: The script uses double-quotes for the npm test argument to mitigate basic shell splitting, though it remains a developer-oriented tool with high local privileges.

Debugging