The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Git and GitHub CLI (gh) to perform repository management, including creating commits, managing branches, tagging releases, and configuring repository secrets.
[EXTERNAL_DOWNLOADS]: Configures workflows that download dependencies via package managers (npm, pnpm, bun, yarn) and utilizes external GitHub Actions from well-known sources like Actions and Softprops.
[CREDENTIALS_UNSAFE]: While handling sensitive data like NPM_TOKEN and Apple certificates, the skill follows safe practices by using gh secret set and explicitly instructing the agent to avoid printing secret values to the UI or logs.
[PROMPT_INJECTION]: The instructions grant the agent high autonomy to automatically commit changes and infer version bumps based on workspace state, which could be exploited if the workspace is pre-seeded with malicious content before the skill is invoked.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository history and branch names to generate release notes and commit messages.
Ingestion points: Reads commit history via git log and version descriptions from CHANGELOG.md and branch names.
Boundary markers: Uses EOF heredoc delimiters in shell scripts to encapsulate dynamically generated content when writing to files or GitHub Actions outputs.
Capability inventory: Includes file system write access for configuration and repository management via git and gh CLI.
Sanitization: Employs text processing tools like awk and grep to filter specific segments of the changelog and commit messages, though it lacks strict structural validation.

lovstudio-release-via-cicd