The Agent Skills Directory

[DYNAMIC_EXECUTION]: The gate-runner.sh script dynamically discovers and executes shell scripts from the skill's bundled directory and a user-defined state directory (CONTRIBUTE_STATE_DIR/user-gates/). Additionally, transition.sh executes dynamically generated Python code to manage YAML frontmatter in candidate files using yaml.safe_load.
[REMOTE_CODE_EXECUTION]: The test-runner.md agent is designed to execute an external repository's native test suite and install dependencies. While this involves running third-party code, the risk is a core component of the skill's purpose and is mitigated by the b16-local-check-allowlist.sh gate, which validates commands against a list of safe runners and blocks shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, such as GitHub issues and CONTRIBUTING.md files, to generate contribution drafts. It mitigates the risk of following malicious instructions embedded in this data through specific safety gates (d02-no-ai-bug-reports.sh and d03-no-ai-pr-reviews.sh) and instructions to use structured templates.

contribute-prepare