The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local shell commands to perform its tasks. These are standard operations for a development-focused tool.
Evidence: Uses ls to check for file existence in SKILL.md Step 0.
Evidence: Uses git tag, git log, and git diff to analyze repository history and changes across various steps.
Evidence: Uses gh pr list and gh api to retrieve pull request and contributor metadata from GitHub in Step 2 and Step 5.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests data from external contributors.
Ingestion points: Pull request bodies (via gh pr list), commit messages (via git log), and changeset files (.changeset/*.md).
Boundary markers: Not explicitly defined in the instructions to separate data from commands.
Capability inventory: Capability to read/write local markdown files and execute git/GitHub CLI commands.
Sanitization: The skill includes logic to filter bot accounts and emails, and Step 6 enforces a mandatory human review of the generated draft before it is used, which significantly mitigates the risk of executing or propagating malicious instructions hidden in metadata.

release-changelog