review
Pass
Audited by Gen Agent Trust Hub on May 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a shell script located at
${ARC_ROOT}/scripts/cleanup-orphaned-agents.shto manage subagent processes. It also performs standard Git operations (git fetch,git diff,git commit) to analyze code changes and update plan documentation. - [PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface (Category 8). It ingests untrusted data from project plans, conversation history, and git diffs, interpolating this content directly into prompts for subagents (
sonnetmodel). - Ingestion points: Plan files (
docs/arc/plans/*.md), conversation context, and git diffs. - Boundary markers: None detected; external content is placed directly into subagent tasks.
- Capability inventory: The agent can write to the filesystem, perform Git commits, and execute shell scripts.
- Sanitization: No sanitization or escaping of the ingested plan content is described before it is passed to subagents.
Audit Metadata