The Agent Skills Directory

[DATA_EXPOSURE]: The skill implements a defensive scanning phase (Phase 1.3) designed to prevent data exposure. It specifically checks for and blocks the commitment of user directory paths, local credentials (auth.json, cookies, session tokens), and private API keys (sk-, ghp_). This is a security-enhancing feature.
[COMMAND_EXECUTION]: The skill executes standard development commands including npx tsc, pnpm build, and pnpm test. These are necessary for the stated purpose of code quality verification and building the extension. It also utilizes git for version control and gh (GitHub CLI) for pull request creation.
[PROMPT_INJECTION]: The skill analyzes git diff output to generate commit messages and documentation. This presents a surface for indirect prompt injection if the code changes contain adversarial instructions. However, the skill's narrow scope and use of Conventional Commit templates act as a natural constraint.
Ingestion points: git diff, git diff --staged, and file contents during architecture documentation updates.
Boundary markers: None explicitly used to wrap the diff content before analysis.
Capability inventory: Git commit/push, GitHub CLI PR creation, and build script execution.
Sanitization: The skill performs pattern matching to block secrets but does not specifically sanitize the diff for injection attacks.
[SAFE]: The execution of git push and gh pr create are the primary intended functions of a 'ship' skill and are triggered by explicit user commands (/ship or /create-pr).

ship