The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze untrusted data sources including error messages, stack traces, and build logs (SKILL.md, Phase 1). This creates a vulnerability surface for indirect prompt injection, as malicious instructions could be embedded in these external data sources to influence the agent's behavior.
Ingestion points: Error messages, stack traces, and logs (SKILL.md, Phase 1).
Boundary markers: None identified; the skill does not define specific delimiters to separate untrusted data from instructions.
Capability inventory: The skill allows for command execution via npm test and diagnostic tools like security and env (SKILL.md and find-polluter.sh).
Sanitization: No sanitization or filtering logic is present for the processed logs.
[COMMAND_EXECUTION]: The skill includes find-polluter.sh, a bash script that executes project tests sequentially using npm test. This is a standard developer tool functionality but constitutes a direct command execution capability.
[DYNAMIC_EXECUTION]: The debugging methodology encourages the use of diagnostic instrumentation, which involves the dynamic execution of environment-probing commands like security list-keychains and codesign to verify configuration during multi-component system failures (SKILL.md, Phase 1, Step 4).
[PROMPT_INJECTION]: The skill contains strong behavioral directives and 'Iron Laws' (e.g., 'NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST') designed to override the agent's default problem-solving approach in favor of a specific debugging methodology. While not a safety bypass, it is a deliberate instruction to override standard operational patterns.

systematic-debugging