The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to manage Git repositories, perform branch checkouts, and interact with the GitHub API via the gh command-line interface.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from GitHub PR bodies and comments, creating a surface for indirect prompt injection.
Ingestion points: Data enters the agent context through gh pr view commands in the setup stage (SKILL.md, 01-setup/CONTEXT.md).
Boundary markers: The skill uses structured handoff files (e.g., handoff.md, pr-body.md) and explicit headers to delimit external content.
Capability inventory: The agent has shell access via the Bash tool, allowing it to perform Git operations, GitHub interactions, and dependency installation (SKILL.md, 03-fix/CONTEXT.md).
Sanitization: Mitigation is implemented through a mandatory "Claims vs Diff" reconciliation gate and by isolating the review stage within a clean-context subagent (02-review/CONTEXT.md).
[EXTERNAL_DOWNLOADS]: To perform its tasks, the skill clones repositories from GitHub and runs dependency installation commands such as npm ci within the target project's environment.
[DYNAMIC_EXECUTION]: The skill includes a test suite (tests/test_installed_resources.py) that uses subprocess.run() to execute bash scripts for validating its shared receipt and head-binding helpers. This execution is confined to the skill's development and testing context.

double-check