The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill integrates untrusted external data into AI prompts, creating a surface for injection attacks.
Ingestion points: ai-review.yml processes git diffs; issue-triage.yml processes issue titles and bodies; deploy.yml processes commit messages from git log; mention-bot.yml processes user comments and PR diffs.
Boundary markers: Prompts lack explicit delimiters or instructions to ignore embedded commands within the untrusted content (e.g., in ai-review.yml, the diff is interpolated directly).
Capability inventory: The skill can create PR reviews, add labels, post comments, fail deployment jobs (core.setFailed), and perform git force-pushes.
Sanitization: No sanitization or filtering of the ingested content is present before interpolation into prompts.
[COMMAND_EXECUTION]: Automated workflows execute powerful git and deployment commands based on repository state or events.
Evidence: auto-rebase.yml performs git rebase and git push --force-with-lease triggered by issue comments. rollback.yml executes git checkout and npm run deploy. smart-tests.yml uses grep on file paths extracted from git diff.
[DYNAMIC_EXECUTION]: The skill heavily relies on the actions/github-script action to execute JavaScript at runtime for API interactions and logic.
Evidence: Found in ai-review.yml, issue-triage.yml, deploy.yml, auto-rebase.yml, branch-cleanup.yml, and mention-bot.yml.

github-workflow-automation