The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes find-polluter.sh, a bash script that executes npm test on files identified by a user-provided pattern. This allows for arbitrary code execution within the context of the testing framework.
[COMMAND_EXECUTION]: The SKILL.md instructions explicitly guide the agent to execute various shell commands for debugging, including env, security, and codesign.
[DATA_EXFILTRATION]: The SKILL.md file provides examples for debugging multi-component systems that involve accessing sensitive environment variables (env | grep IDENTITY) and macOS keychain information (security list-keychains, security find-identity). While intended for troubleshooting build or signing failures, these commands expose credentials and cryptographic identities to the agent context.
[PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface by mandating that the agent 'Read Error Messages Carefully' and 'Gather Evidence' from logs and component boundaries. This processes untrusted data (error strings, log output) that could contain malicious instructions designed to influence agent behavior during the debugging process.
Ingestion points: SKILL.md (Phase 1: Read Error Messages, Gather Evidence in Multi-Component Systems).
Boundary markers: Absent; there are no instructions to ignore or delimit instructions within the logs or errors.
Capability inventory: Subprocess execution via find-polluter.sh and various shell commands suggested in SKILL.md and root-cause-tracing.md.
Sanitization: Absent; the agent is encouraged to read and analyze raw traces and logs directly.

systematic-debugging