The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of reading and analyzing external project files.\n
Ingestion points: The skill uses Grep, Glob, and Read tools to ingest content from the local codebase into the agent's context, as documented in references/investigation-checklist.md.\n
Boundary markers: There are no explicit instructions or delimiters provided to the agent to distinguish between its own system instructions and the potentially untrusted data being read from codebase files.\n
Capability inventory: The skill has access to powerful tools, including the ability to execute shell commands (git, npm), modify local files, and interact with remote repositories through the GitHub CLI (gh) in references/thorough-track.md.\n
Sanitization: While the skill provides high-level advice to redact secrets/PII in references/investigation-checklist.md, it lacks automated sanitization or filtering of codebase content that could contain adversarial instructions.\n- [COMMAND_EXECUTION]: The skill makes extensive use of local command-line utilities to perform its tasks.\n
Evidence: The workflow includes git operations (worktrees, commits, pushes), npm commands (npm install, npm run test), and GitHub CLI (gh pr) for pull request management in references/thorough-track.md. These operations are standard for the skill's intended purpose of software development and debugging.

debug