The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The generated GitHub Actions workflows depend on the anthropics/claude-code-action@v1 and actions/checkout@v4 actions, which are hosted on trusted and well-known repositories.
[COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh) and git commands via the Bash tool to inspect repository state, check for secrets, and validate workflow files during the setup phase.
[DATA_EXFILTRATION]: The automated workflow templates use gh run view --log-failed to retrieve failure logs. These logs are processed locally by the AI action but contain sensitive project operational details.
[PROMPT_INJECTION]: The skill creates an automated process that ingests untrusted CI logs and has the capability to modify code via Pull Requests.
Ingestion points: CI logs and job summaries are read into the model context in the generated workflow steps defined in REFERENCE.md.
Capability inventory: The workflow templates request contents: write, pull-requests: write, and issues: write permissions, enabling the agent to commit changes and create PRs.
Boundary markers: The instructions in REFERENCE.md structure the task but do not include explicit separators or warnings to ignore commands embedded within the log data.
Sanitization: There is no evidence of log sanitization before processing, relying on the model's internal safety filters to ignore potential malicious content in logs.

ci-autofix-reusable