crack-setup
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local JavaScript scripts using the
nodecommand (e.g.,../crack/scripts/setup.mjs). These operations are used to scan for available models, plan configuration changes, and apply settings to the 'crack.toml' file. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external configuration files and model catalogs to generate setup recommendations, creating a potential surface for indirect instructions.
- Ingestion points: Reads user-provided configuration files (
crack.toml,config.toml) and model catalog metadata as described inSKILL.md. - Boundary markers: The instructions define a multi-step process where the agent must present a 'plan' and show 'concrete model/effort and file changes' for user approval before execution.
- Capability inventory: The skill is capable of reading and writing local configuration files and executing specific setup scripts through the
noderuntime. - Sanitization: The process incorporates a mandatory human-in-the-loop verification step where the agent must obtain a decision after producing a preview of intended changes.
Audit Metadata