deep-research
Pass
Audited by Gen Agent Trust Hub on Mar 4, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of reading and summarizing untrusted external content and codebase files.
- Ingestion points: Data enters the agent context through WebFetch (external websites), Read/Glob/Grep (local codebase), and ListMcpResourcesTool (MCP resources) during Phase 2 and Phase 4.
- Boundary markers: Prompts utilize XML-style delimiters such as
<user-task>and<hypotheses>and include explicit instructions for sub-agents to treat the enclosed content as data rather than executable instructions. - Capability inventory: Specialist agents have access to Bash (for environment queries), Write/Edit (limited to .speak-memory updates), and the Agent tool for spawning specialist sub-agents. The orchestrator has access to web search and file system tools.
- Sanitization: The skill relies on the underlying model's adherence to safety guidelines and explicit 'ignore instructions' warnings; it does not implement programmatic sanitization or filtering of retrieved content.
- [COMMAND_EXECUTION]: The skill uses the Bash tool to perform environment checks and dependency audits during Phase 1 and Phase 2.
- System Discovery: Executes
git rev-parse --show-toplevelto identify the repository root. - Dependency Auditing: Runs package manager queries including
npm ls,pip list, andcargo treeto identify the project's technology stack. - [EXTERNAL_DOWNLOADS]: Automated web research is conducted using specialized search and fetch tools to gather documentation and best practices.
- Tools: Employs WebSearch and WebFetch to retrieve content from the public internet.
- Context: External content is used to confirm or disconfirm hypotheses during the research and adversarial challenge phases.
- [REMOTE_CODE_EXECUTION]: The skill utilizes the npx package runner to search for available skills in the npm marketplace.
- Mechanism: Executes
npx skills findwith task-relevant keywords to identify additional capabilities. - Source: This operation targets the official npm registry, which is a well-known and trusted technology service.
Audit Metadata