The Agent Skills Directory

[Command Execution] (SAFE): The skill runs a local script scripts/release.sh to manage the release workflow. The actions (git commit, npm version, gh release) are standard for the skill's purpose and occur within the local repository environment.
[Indirect Prompt Injection] (LOW): The skill processes CHANGELOG.md which is an external data source. Evidence Chain: 1. Ingestion points: Step 1 reads CHANGELOG.md to summarize changes. 2. Boundary markers: Absent; the agent is asked to summarize the content directly. 3. Capability inventory: The skill can perform git push, npm version, and gh release create via scripts/release.sh. 4. Sanitization: No explicit sanitization of the changelog content or the version argument in the shell script. However, the workflow requires the user to review the summary and confirm the release, providing a human-in-the-loop safeguard.

release