The Agent Skills Directory

[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests and processes untrusted data from the repository.
Ingestion points: Reads git log, git diff output, package.json metadata, and existing CHANGELOG.md files.
Boundary markers: None detected; the agent is instructed to use these inputs directly to generate release notes and classify semver bumps.
Capability inventory: Includes arbitrary shell command execution (bash), file system writes (MIGRATIONS.md, CHANGELOG.md), and network operations (npm publish).
Sanitization: No explicit sanitization or instruction to ignore embedded directives in the ingested content was found.
[COMMAND_EXECUTION]: The workflow executes the build script defined in the user's package.json file as part of the verification process. It also performs smoke tests by importing built entry points in a temporary directory, which executes the project's compiled code.
[EXTERNAL_DOWNLOADS]: The skill provides an option to download and install the @changesets/cli package from the npm registry if it detects a workspace that has not yet been initialized with changesets.

release