sdlc-design-arch

Warn

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes shell commands and a Python script (.wiki-llm/tools/wiki_llm.py) found within the target project's repository. This grants the agent the capability to run code provided by the repository it is currently processing.
  • [DYNAMIC_EXECUTION]: The script path for the tool T is determined dynamically based on the project's root folder ($WIKI) and executed at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from project documentation (Requirements, Features, ChangeRequests) to generate architecture artifacts. * Ingestion points: SKILL.md specifies reading requirements and PRDs from the wiki repository. * Boundary markers: The protocol includes three human approval gates (Intent, Design, Change-set). * Capability inventory: The agent can read/write files and execute shell/Python commands. * Sanitization: No explicit programmatic sanitization is described for the content of the markdown files before processing.
  • [EXTERNAL_DOWNLOADS]: The skill performs git operations to synchronize with a remote repository. * Evidence: git fetch and git push commands are used to manage the wiki's default and draft branches.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 13, 2026, 09:20 AM