The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with local project files to automate the release sequence, creating a potential surface for indirect injection if those files were compromised.
Ingestion points: Reads version information and changelog updates from package.json and CHANGELOG.md.
Boundary markers: Absent; the skill executes shell commands directly using the data read from files.
Capability inventory: Includes file system reads/writes via npm, version control operations via git, and network communication via npx and npm.
Sanitization: Relies on the structured input requirements of the npm and git CLI tools.
[CREDENTIALS_UNSAFE]: The skill references a NPM_TOKEN for authentication with the npm registry.
Evidence: The instructions state that staging uses a "configured NPM_TOKEN credential." This follows standard industry practices for automated CI/CD workflows and does not involve hardcoded secrets or unsafe storage instructions.

releasing