chromatic-turbosnap-debug
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data provided by the user, such as build logs, CLI output, and configuration snippets. This represents an indirect prompt injection attack surface. However, the risk is mitigated because the skill's instructions focus on generating a structured diagnosis card and recommending commands for the user to run, rather than the agent performing high-privilege actions or executing instructions found within the data.
- Ingestion points: SKILL.md and template.md define inputs for CLI output, build logs, and configuration files.
- Boundary markers: The skill uses a structured output contract (reference/output-contract.md) but does not explicitly define input delimiters.
- Capability inventory: The skill recommends diagnostic commands (npx chromatic) but does not perform file writes, network operations, or dynamic code execution itself.
- Sanitization: No explicit sanitization of user-provided logs is mentioned.
- [COMMAND_EXECUTION]: The skill includes a command catalog (reference/command-catalog.md) that recommends diagnostic shell commands for the user to execute locally. These commands (e.g., npx chromatic trace, npx @chromatic-com/turbosnap-helper) are legitimate vendor-specific tools for Chromatic and are recommended based on the diagnostic context.
- [DATA_EXFILTRATION]: The skill describes workflows involving human-operated artifacts, such as sharing hosted metadata URLs with support. The instructions explicitly state that the agent should not fetch these URLs directly, which prevents automated exfiltration of data from those URLs by the agent.
Audit Metadata