ralph
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the
omcCLI tool and a Node.js bridge script (bridge/cli.cjs) to perform verification tasks, record environment baselines, and execute session management commands. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external sources, including a task-specific
prd.jsonfile and a configurable MCP tool (companyContext.tool). - Ingestion points: Data enters the agent context through the
prd.jsonfile and the markdown output of thecompanyContext.toolcall. - Boundary markers: The skill uses structural markdown tags and explicitly instructs the agent to treat external tool output as "quoted advisory context only."
- Capability inventory: The agent has the ability to execute shell commands via the
omctool, perform file system writes to session state, and delegate sub-tasks to other agents. - Sanitization: The instructions include explicit directives to avoid treating external data as executable instructions and to rely on automated verification rather than manual judgment of external content.
- [PERSISTENCE]: The skill maintains task state across multiple iterations by writing to session-scoped files in the
.omc/directory and maintaining aprogress.txtlog, which is necessary for its function as a persistent loop.
Audit Metadata