ai-coding-agents

Pass

Audited by Gen Agent Trust Hub on Oct 17, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references numerous official documentation sites and public repositories from trusted organizations and well-known technology services, including Anthropic, OpenAI, Google, GitHub, and Cloudflare. These references are used for architectural guidance and tool integration without any unverifiable or dangerous third-party downloads.
  • [COMMAND_EXECUTION]: The skill includes a diagnostic script, scripts/smoke_test.sh, which is intended to validate the local environment before an agent session. This script performs benign operations such as checking for installed CLIs, verifying the presence of project configuration directories, and probing the reachability of Anthropic and OpenAI APIs using secure practices for API key handling.
  • [DYNAMIC_EXECUTION]: Scaffolding templates for the Agent SDK (e.g., assets/templates/sdk-agent-py.py) implement standard tool execution patterns to facilitate development tasks such as running test suites (via pytest) or linters (via ruff). These operations are encapsulated within defined tool structures for use by the created agents.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides architectures for agents that ingest untrusted data in the form of repository source code. It mitigates associated risks by documenting mandatory evidence chains: ingestion occurs through file-reading tools; boundary markers are established via the 'Prompt Injection via Code' patterns in references/debugging-guide.md; capabilities are explicitly scoped in archetype templates; and sanitization is implemented through safety hooks that block access to sensitive files like .env, credentials, and .git.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 17, 2026, 08:26 AM