release-doc
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: Several library scripts interact with system tools to gather git and GitHub information.\n
- Evidence:
parse-commits.cjs,bump-version.cjs, andextract-pr-metadata.cjsusechild_process.execSyncto rungitandghcommands.\n- [EXTERNAL_DOWNLOADS]: The skill uses an external API to perform text transformations.\n - Evidence:
lib/transform-llm.cjsmakes network requests to Anthropic's official API (api.anthropic.com).\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from git logs and pull request summaries which serves as an attack surface.\n - Ingestion points: Reads git logs (
git-log-*.txt), diffs, and pull request metadata from external repositories (specified inSKILL.mdStep 0b and R1.1/R1.2).\n - Boundary markers: The workflow implements a 'Human Review Gate' and uses
escapeMarkdowninrender-template.cjs, though explicit delimiters for raw diff content are not strictly defined in instructions.\n - Capability inventory: The skill can write files (reports), execute commands (git/gh), and access the network (Anthropic API).\n
- Sanitization: Employs
lib/utils.cjs::escapeMarkdownto sanitize commit descriptions before rendering into markdown.\n- [REMOTE_CODE_EXECUTION]: A utility script contains a potential command injection vulnerability if user-supplied arguments are not handled carefully.\n - Evidence:
lib/bump-version.cjsbuilds a git command string using the--serviceargument (tagPattern) without sanitization before passing it toexecSync.
Audit Metadata