ship
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPERSISTENCE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a wide array of git, CLI, and repository-specific test/build commands to automate the release process. It dynamically determines base branches and manages complex merge conflicts via user interaction.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including repository source code, diffs, plan files, and PR descriptions. It mitigates injection risks by using the Agent tool for isolation, employing strict delimiters, and providing explicit instructions to subagents to treat repository content as data rather than instructions.
- [DYNAMIC_EXECUTION]: Dispatches multiple specialist subagents (e.g., Security, Performance, Data Migration) to perform code reviews. These subagents are given dynamic prompts and context based on the current branch's changes, operating within fresh-context boundaries to prevent bias or cross-contamination.
- [EXTERNAL_DOWNLOADS]: Installs and updates well-known development utilities such as 'fastlane' via Homebrew and '@openai/codex' via NPM when missing. It also interfaces with 'Aside AI browser' for manual web-based release steps.
- [CREDENTIALS_UNSAFE]: Handles sensitive release credentials, including Apple App Store Connect API keys and session tokens. It implements a strict security protocol for these secrets, writing them to files with owner-only permissions (0600) and explicitly preventing their inclusion in logs, chat history, or shell argv.
- [PERSISTENCE]: Provides functionality to install a managed git pre-push hook. This hook acts as a security persistence mechanism designed to prevent the accidental pushing of credentials (API keys, private keys) to remote repositories.
Audit Metadata