The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches reference workflow templates from the github/gh-aw repository using the GitHub CLI. These downloads target a trusted organization and are used to seed the adoption process.
[COMMAND_EXECUTION]: The adoption prompts and coordination scripts utilize standard CLI tools such as gh, git, jq, and curl to automate branch creation, PR management, and workflow compilation. These operations are within the expected scope of a workflow adoption tool.
[INDIRECT_PROMPT_INJECTION]: The included repo-guardian.md workflow analyzes content from pull requests, which is an untrusted ingestion point.
Ingestion points: Reads the content of changed files in pull requests via the gh-aw framework.
Boundary markers: The workflow uses natural language rules to define what to reject, though explicit "ignore embedded instructions" markers for the analyzed content are not specified.
Capability inventory: The workflow is restricted by safe-outputs to a maximum of one comment per run and has read-only access to repository contents.
Sanitization: Findings are validated by a separate standard GitHub Action (repo-guardian-gate.yml) which requires a manual override comment for violations, effectively providing a human-in-the-loop safety mechanism.
[SAFE]: The skill provides extensive documentation on security best practices, including the use of least-privilege permissions, network firewall allowlisting, and safe-output limits to prevent runaway automation.

gh-aw-adoption