The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh) to interact with GitHub Actions, specifically listing workflow runs and downloading feedback artifacts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub Actions artifacts to update its own instruction files and CLAUDE.md.
Ingestion points: Workflow artifacts named agent-feedback-* are downloaded and read in Step 3.
Boundary markers: The instructions explicitly direct the agent to rephrase insights in its own words and avoid verbatim copying of feedback text.
Capability inventory: The agent has the ability to execute gh commands and write to various skill files within the .claude/ directory and CLAUDE.md.
Sanitization: The skill performs file-level validation (checking for null bytes, size limits of 50KB, and file extensions) and content-level sanitization (stripping shell commands and code blocks).

analyze-feedback