The Agent Skills Directory

COMMAND_EXECUTION (LOW): The skill provides the find-polluter.sh bash script which executes npm test on files matching a user-defined pattern. Additionally, SKILL.md suggests using system commands like security list-keychains and codesign to troubleshoot build and signing issues. These are legitimate tools for technical debugging but constitute direct command execution capabilities.
PROMPT_INJECTION (LOW): The skill uses strong directives such as 'The Iron Law' and 'Violating the letter of this process is violating the spirit of debugging' to override the agent's typical behavior and enforce a strict methodology. This is a task-specific behavioral constraint rather than a malicious safety bypass.
Indirect Prompt Injection (LOW): The skill is designed to ingest and act upon data from external sources such as error logs, stack traces, and git diffs.
Ingestion points: Error messages and stack traces parsed in SKILL.md Phase 1.
Boundary markers: Absent; there are no specific delimiters or instructions to ignore embedded commands in logs.
Capability inventory: npm test execution via find-polluter.sh and various shell diagnostics suggested in SKILL.md.
Sanitization: Absent; the skill encourages the direct analysis of raw system and test output.

systematic-debugging