The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the gh (GitHub) CLI, git, and bash to manage repository state, trigger workflows, and interact with GitHub APIs. These operations are essential for the skill's purpose (CI/CD management) and are protected by confirmation gates for sensitive actions.
[DYNAMIC_EXECUTION]: During the setup and update phases, the skill generates a new AI agent file (.claude/agents/deploy-admin.md) by populating a template with repository configuration and workflow data. It also generates and writes GitHub Action workflow YAML files to the repository.
[REMOTE_CODE_EXECUTION]: The release process (P4) involves probing for and executing scripts found within the user's repository (e.g., version bump and post-release scripts). While these are local to the project, they represent a vector for executing arbitrary code if the repository itself contains malicious content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository, including package.json scripts, RELEASE-NOTES.md content, and git commit history to automate changelog generation and versioning.
Ingestion points: Project configuration files (package.json, CLAUDE.local.md) and documentation (RELEASE-NOTES.md) are read from the repository.
Boundary markers: The skill uses AskUserQuestion to confirm actions before proceeding with mutations based on extracted data.
Capability inventory: The skill can execute shell commands via bash, trigger GitHub Actions via gh, and modify files via Write and Edit tools.
Sanitization: The skill uses structured tools like jq and awk for parsing, but does not explicitly sanitize the extracted text before incorporating it into the generated agent template or changelog.
[DATA_EXFILTRATION]: The skill performs health checks using curl against URLs specified in the project's local configuration (CLAUDE.local.md). While intended for monitoring deployment status, this involves making network requests to user-defined endpoints.

deploy