The Agent Skills Directory

[COMMAND_EXECUTION]: The skill triggers the execution of shell commands for software development lifecycle tasks. Key commands include npm install for dependency management within worktrees, npm run test:run for validating bug fixes, git worktree for isolated development, and gh pr for pull request management. While these are legitimate developer actions, they represent a broad set of capabilities available to the agent.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted input sources, including user-provided symptom descriptions and reproduction steps. This data is incorporated into the agent's operational context to drive the debugging process. Maliciously crafted input could potentially attempt to hijack the agent's logic during the investigation or implementation phases.
Ingestion points: Bug reports, reproduction steps, and log content ingested through the triage and investigation modules defined in SKILL.md and references/investigation-checklist.md.
Boundary markers: There are no explicit security delimiters used when interpolating external input into prompts.
Capability inventory: The agent has access to file system reading/writing via Grep and Read tools, shell execution for testing and version control, and network access for PR creation.
Sanitization: The references/investigation-checklist.md file instructs users to redact secrets and PII, providing a manual check against data exposure, but does not provide programmatic protection against malicious prompt injection content.

debug