The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to interact with version control and package managers.
Evidence: Utilizes git, npm, pnpm, yarn, and bun for operations such as git describe, git log, npm pack, and npm publish.
Context: All sensitive operations (publishing, tagging, and pushing) are explicitly gated behind user confirmation prompts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository's git history and file content which could theoretically contain malicious instructions.
Ingestion points: Reads output from git log and git diff in SKILL.md to determine changes and generate changelogs.
Boundary markers: The skill instructs the agent not to invent changelog entries and to base them strictly on actual commits.
Capability inventory: The skill can write files (CHANGELOG.md, MIGRATIONS.md, .changeset/*.md) and execute shell commands (npm publish, git push) through the agent's environment.
Sanitization: While no explicit technical sanitization is mentioned, the workflow requires a structured confirmation step where the user reviews proposed bumps and authors the release before publishing.
[DYNAMIC_EXECUTION]: The skill executes the project's own build scripts during the pre-publish verification phase.
Evidence: Step 4 in SKILL.md specifies running the package's build script to ensure compilation success before packaging.
[EXTERNAL_DOWNLOADS]: The skill may prompt the user to install an external package to support the workflow.
Evidence: Offers to install @changesets/cli using the project's package manager if changesets are not already initialized.

release