delegate-review
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, potentially untrusted data such as
git diffoutputs and source code files. This content is then analyzed by subagents (e.g., Claude Opus, GPT models), creating a vulnerability where malicious instructions embedded in the reviewed code could influence the subagent's actions. - Ingestion points: Untrusted data enters the skill via
git diff,git log, and direct file reading based on the user-defined review scope. - Boundary markers: The skill employs a structured Markdown request template (Objective, Scope, Context, Constraints) to help the agent distinguish between meta-instructions and the data being processed.
- Capability inventory: The skill uses
Bashto run orchestration logic and delegates tasks to powerful subagents which may have their own tool-use capabilities (e.g., file access or code execution). - Sanitization: The provided orchestration scripts do not demonstrate explicit sanitization or filtering of the ingested diff content before it is delivered to the subagent.
- [COMMAND_EXECUTION]: The skill contains multiple shell scripts in the
scripts/directory (e.g.,run.sh,dispatch.sh,prepare.sh) that are used to manage the lifecycle of the delegation. These scripts execute local system commands includingnode,git, and other internal bash utilities. - [DYNAMIC_EXECUTION]: The shell scripts act as shims to execute a bundled Node.js script (
delegate-cli.mjs) found within the skill directory. This pattern is used to run the core logic of the delegation tool in the local environment.
Audit Metadata