omc-setup
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple bash and node scripts bundled with the plugin to handle the configuration lifecycle. Specifically:
- Invokes
setup-claude-md.shandsetup-progress.shfrom the plugin root to manage CLAUDE.md and state tracking. - Uses
nodeto probe configuration states and repair plugin cache references. - Uses
jqextensively to safely read and mutate JSON configuration files. - [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the 'oh-my-claude-sisyphus' global CLI tool and checks for updates using well-known services:
- Performs
npm install -gfor the project's own CLI tool after prompting the user. - Uses
npm viewto check for the latest versions on the official registry. - Interacts with the GitHub API via the
ghCLI to star the project repository, contingent on user authentication and consent. - [PERSISTENCE]: The skill establishes persistence for its environment by:
- Modifying global and project-local
CLAUDE.mdfiles to inject managed imports and OMC markers. - Updating
~/.claude/settings.jsonto enable experimental features (Agent Teams) and configure display modes. - Modifying
.git/info/excludeto manage project-specific file ignore rules. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its interaction with external data:
- Ingestion points: Reads and parses
settings.json,.omc-config.json, and.omc/state/setup-state.jsonduring the setup and resume phases. - Boundary markers: Does not utilize explicit boundary markers or 'ignore' instructions for the data read from these files.
- Capability inventory: Possesses capabilities for shell execution (
bash), package management (npm), and local file modification across all phases. - Sanitization: Employs
jqfor JSON manipulation, which provides structural validation and prevents syntax-level corruption, though it does not sanitize the semantic content of the instructions.
Audit Metadata