The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands (e.g., gh repo clone, git switch -c, scorecard --repo) that interpolate user-supplied strings such as repository_full_name and branch_name. This presents a risk of command injection if the inputs are not properly sanitized.
[PROMPT_INJECTION]: The skill processes external data from a JSON proposal file (proposal_json_path) to determine which changes to apply. This represents an indirect prompt injection surface where a malicious proposal could trick the agent into committing backdoors or unintended changes to a repository.
Ingestion points: proposal_json_path file (SKILL.md).
Boundary markers: None specified for delimiting the JSON content from instructions.
Capability inventory: git commit, git push, gh pr create, and language-specific test runners (go test, npm test, cargo test) (SKILL.md).
Sanitization: Includes a leak check using grep to detect secrets in the diff before committing or pushing.
[EXTERNAL_DOWNLOADS]: Fetches source code and repository metadata from GitHub using the gh CLI and git. These operations target a well-known service for legitimate repository management tasks.

github-supply-chain-hardening-remediation