The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses execSync to run local git and gh CLI commands to extract repository history and pull request details. It implements a sanitizeGitRef function in lib/parse-commits.cjs to mitigate the risk of shell injection through branch or tag names.
[DATA_EXFILTRATION]: The script lib/transform-llm.cjs performs network operations by sending release note content to the Anthropic API (api.anthropic.com) for AI-powered summarization. This is a documented feature using a well-known service and requires an user-provided ANTHROPIC_API_KEY via environment variables.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface where untrusted data from the repository is processed and passed to an LLM.
Ingestion points: Git commit subjects and bodies (read in lib/parse-commits.cjs) and pull request metadata (fetched in lib/extract-pr-metadata.cjs).
Boundary markers: The interpolation logic in lib/transform-llm.cjs does not use explicit delimiters or instructions to ignore potential commands embedded in the commit messages.
Capability inventory: Subprocess execution for git/gh CLI, network requests to Anthropic API, and file system writes for markdown reports and changelogs.
Sanitization: Use of escapeMarkdown in render-template.cjs prevents markdown formatting breakage but does not filter logic-based injection strings.
[SAFE]: The skill implements robust path validation via the validateOutputPath utility in lib/utils.cjs. This check ensures all generated files are written within the project's root directory, preventing path traversal attacks.

release-notes