anycap-blog-production

Pass

Audited by Gen Agent Trust Hub on Apr 24, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8). It ingests untrusted data such as bullets, benchmarks, and URLs as specified in SKILL.md and references/input-brief.md. No explicit boundary markers or sanitization logic are present to isolate this data. The skill possesses significant capabilities, including executing shell commands via the anycap CLI, mkdir, and pnpm, creating a risk if malicious instructions are embedded in the ingested content.
  • [COMMAND_EXECUTION]: The skill executes multiple shell commands to manage the production workflow. This includes creating directories via mkdir, optimizing images using the sips utility, and running code linting through pnpm exec eslint.
  • [COMMAND_EXECUTION]: It relies on a proprietary CLI tool anycap for core functions, including checking system status, exploring model schemas, generating assets (image, video, and music), and performing vision-based verification of generated content.
  • [DATA_EXFILTRATION]: The skill processes user-provided inputs and research data, which are then transmitted to vendor-controlled endpoints via the anycap CLI tool for processing and generation. While this is the intended functionality of the skill, it involves the transfer of potentially sensitive user data to external services.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 24, 2026, 08:44 AM