skill-research-process

Pass

Audited by Gen Agent Trust Hub on May 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill fetches untrusted data from the open web to create instructions and documentation for other agents.
  • Ingestion points: Research data retrieved from external URLs via MCP tools as specified in references/agent-prompts.md.
  • Boundary markers: The workflow does not currently implement automated boundary markers or instructions to ignore embedded prompts within the fetched content.
  • Capability inventory: The skill coordinates subagents using the Agent tool and executes local shell commands for project management.
  • Sanitization: The process relies on manual verification gates (e.g., Quality Gate 2 in SKILL.md) and anti-hallucination checkpoints rather than programmatic sanitization.
  • [COMMAND_EXECUTION]: Local Utility Invocation. The skill uses shell commands to run local Python scripts for initializing projects (init_skill.py) and validating the final skill package (package_skill.py) within the plugins/plugin-creator/ directory.
  • [EXTERNAL_DOWNLOADS]: Automated Information Retrieval. The research phase involves fetching content from third-party documentation sites and GitHub repositories using MCP tools (WebFetch, Ref, Exa) and fallbacks such as the GitHub CLI.
Audit Metadata
Risk Level
SAFE
Analyzed
May 10, 2026, 01:52 PM