skills/connorads/dotfiles/inspo/Gen Agent Trust Hub

inspo

Pass

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches design metadata from a remote API endpoint and downloads website screenshots from a cloud storage service.
  • Evidence: scripts/inspo.py makes requests to https://inspomcp.dev/api/mcp for tool execution and retrieves images from Vercel's public blob storage (https://0nme3pk5am3urwa9.public.blob.vercel-storage.com/captures).
  • Purpose: These operations are the primary function of the skill, serving visual design inspiration to the agent.
  • [COMMAND_EXECUTION]: The test suite executes the skill's main script as a subprocess to verify CLI behavior.
  • Evidence: tests/test_inspo.py uses subprocess.run to call scripts/inspo.py with testing-specific flags like --offline and --cache.
  • Context: This is a standard practice for unit and integration testing of command-line tools.
  • [DYNAMIC_EXECUTION]: The test suite dynamically loads the main script as a module to perform internal unit tests.
  • Evidence: tests/test_inspo.py utilizes importlib.util to load and execute scripts/inspo.py within the test process.
  • Context: This is a legitimate development pattern used for testing Python scripts that are not structured as standard packages.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes natural language descriptions and analysis ('autopsies') of websites from an external archive, which could theoretically contain instructions.
  • Ingestion points: Data is ingested from the external API in scripts/inspo.py and presented to the agent.
  • Boundary markers: SKILL.md contains explicit instructions warning the agent not to treat the text descriptions as a direct code brief, emphasizing the priority of visual inspection over prose.
  • Capability inventory: The skill can perform network reads and write files to a local cache directory (~/.cache/inspo).
  • Sanitization: Data is parsed as structured JSON; the script implements size limits (--max-bytes) to prevent resource exhaustion from large remote payloads.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 29, 2026, 08:57 PM