The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the gh (GitHub CLI), git, and jq tools to manage CI/CD workflows and parse repository data. These actions are consistent with the skill's purpose as a maintainer tool and target the official repository NVIDIA/NemoClaw.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from external sources.
Ingestion points: The agent reads job names, run titles, PR metadata, and API responses via gh run list, gh run view, and gh api (e.g., in references/release-context.md and references/manual-pr.md).
Boundary markers: The skill explicitly instructs the agent to avoid making release decisions ("Do not decide whether a release can proceed") and provides specific routing for different types of requests to maintain context isolation.
Capability inventory: The agent can dispatch workflows (gh workflow run), interact with repository APIs, and manage local test files.
Sanitization: The instructions mandate the use of jq for processing JSON output, ensuring that external data is handled as structured content rather than raw text where possible.
[CREDENTIALS_UNSAFE]: The skill documents the handling of sensitive environment variables and secrets (such as NVIDIA_API_KEY, BREV_API_KEY, and GH_TOKEN) within the CI environment. It does not hardcode credentials but rather provides maintainer instructions for secret rotation, resource cleanup, and monitoring for potential credential exposure during E2E runs.

nemoclaw-maintainer-e2e