The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the shell to execute a sequence of git, gh (GitHub CLI), and npm commands to inspect changes, manage local/remote branches, and submit code to GitHub.
[PRIVILEGE_ESCALATION]: The instructions mandate that the agent requests required_permissions: ["all"] when calling shell tools for gh and git push. This is designed to allow the agent to access the host OS keyring for GitHub authentication, which is typically restricted in sandboxed environments.
[INDIRECT_PROMPT_INJECTION]: The procedure requires the agent to read and strictly adhere to instructions found in repository files such as CONTRIBUTING.md, .github/PULL_REQUEST_TEMPLATE.md, and packages/*/AGENTS.md. These files could potentially contain adversarial instructions that influence the agent's PR generation logic.
Ingestion points: Reads multiple markdown files from the local repository during the workflow.
Boundary markers: Absent; the skill explicitly instructs the agent to follow the PR template structure "exactly" without stripping external content.
Capability inventory: Shell execution (git, gh, npm), network access (pushing code), and file system modification (creating branches/commits).
Sanitization: Absent; external file content is interpolated into the final pull request body.
[EXTERNAL_DOWNLOADS]: The skill executes npm run ci as a pre-submission health check. This command typically triggers the installation or update of dependencies from the npm registry if the local environment is not fully synchronized.

submit-pr-from-current-changes