The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell script, find-polluter.sh, designed to iterate through test files and execute npm test to isolate tests that cause state pollution. It also provides examples for running diagnostic commands such as git init and codesign to verify environment behavior.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze external data like error messages, stack traces, and system logs to identify root causes.
Ingestion points: Error messages, stack traces, and component logs processed during Phase 1: Root Cause Investigation.
Boundary markers: The skill enforces a strict 4-phase logical workflow that requires independent evidence gathering before any code changes are proposed, which serves as a process-level safeguard against following adversarial instructions embedded in logs.
Capability inventory: Shell execution capabilities for testing (npm test) and diagnostic tools (git, security, codesign).
Sanitization: There is no explicit sanitization of ingested log content, but the methodology focuses on structural analysis rather than direct execution of data.
[PRIVILEGE_ESCALATION]: The skill's documentation and examples describe the use of system-level utilities such as codesign and macOS security (keychain/identity management) to diagnose and verify build integrity and environment state. These are utilized within the scope of their intended purpose for debugging technical failures.

systematic-debugging