The Agent Skills Directory

[Indirect Prompt Injection]: The skill processes external data that could influence its final decision-making. It is designed to analyze changesets, git diffs, and GitHub pull request content to recommend whether to ship or block a release.
Ingestion points: The agent reads content from .changeset/*.md files, package metadata, git commit history, and GitHub PR descriptions/diffs.
Boundary markers: The skill relies on natural language instructions to separate the audit logic from the data being audited, though it includes specific validation rules for changeset frontmatter.
Capability inventory: The skill uses git commands and local helper scripts to perform its analysis.
Sanitization: While the skill validates the structure of changesets, the natural language content within those files or PR descriptions is not sanitized, creating a surface for potential manipulation of the audit reasoning.
[Command Execution]: The skill invokes a local bash script (find_latest_release_tag.sh) to synchronize remote tags and identify the most recent release. The script uses standard git fetch and git tag operations to provide the necessary context for the release comparison.

final-release-review