baku-coding-discipline

Pass

Audited by Gen Agent Trust Hub on Jun 15, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions and specific commands to download and install additional engineering skills (e.g., 'tdd', 'diagnose') from public GitHub repositories, including 'github.com/mattpocock/skills' and 'github.com/multica-ai/andrej-karpathy-skills'. These are recognized as well-known sources for developer tools.
  • [COMMAND_EXECUTION]: The skill contains instructions for the agent to execute shell commands for installation ('npx skills add'), validation ('python3 scripts/validate.py'), and testing. It defines a 'Read-only' mode that restricts execution and modifications until a plan is approved by the user.
  • [PROMPT_INJECTION]: Indirect prompt injection surface identified due to the skill's primary function of processing external code and requirements.
  • Ingestion points: The agent reads and processes user requirements, source code, project specifications, and AI-generated code patches ('SKILL.md', 'references/mode-routing.md').
  • Boundary markers: Absent; the skill does not use specific delimiters or 'ignore' instructions for external data within its prompts.
  • Capability inventory: The skill has the capability to execute shell commands, modify files, and install external packages at runtime.
  • Sanitization: No explicit sanitization or validation of the input content is implemented. However, the risk is addressed via mandatory 'Read-only' modes for initial research tasks.
  • [SAFE]: The skill includes a proactive defensive script ('scripts/validate.py') that scans for potential secrets such as API keys, tokens, and AWS credentials using regular expression patterns, serving as a best-practice security safeguard.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 15, 2026, 07:48 AM
Security Audit — agent-trust-hub — baku-coding-discipline