The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git and gh CLI tools to manage branches, view changes, and create draft pull requests. To ensure safety, all state-altering commands such as git commit and git push are presented to the user as suggestions to be run manually, rather than being executed by the agent directly.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from existing workflow files. Potential risks associated with ingesting untrusted file content are mitigated by several guardrails:
Ingestion points: Reads and edits files within the .github/ directory using Glob, Grep, and Read tools.
Boundary markers: The skill explicitly requires showing a git diff and obtaining user confirmation before proceeding.
Capability inventory: Uses Edit for file modifications and Bash for branch creation and PR generation.
Sanitization: The multi-step process requires the user to manually execute the commit and push commands, providing a final layer of review for all generated content.
[EXTERNAL_DOWNLOADS]: The skill references the bitwarden/workflow-linter repository to verify that replacement actions are on an approved list. This is a vendor-internal resource used to enforce security policies and does not represent an untrusted external download.

action-remediate