claudish-usage

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides numerous examples of shell commands involving the claudish CLI, including installation (npm install -g claudish) and execution (claudish --model <model> "task"). These commands enable interaction with external AI services.
  • [EXTERNAL_DOWNLOADS]: Recommends downloading and installing the claudish package from public registries (NPM or Bun).
  • [PRIVILEGE_ESCALATION]: Notes the availability of a --dangerous flag for the CLI, which is documented to disable internal sandboxing mechanisms.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines workflows that ingest external data and process it through third-party models, creating a potential injection surface.
  • Ingestion points: Data enters the environment via file reads and standard input passed to the claudish CLI in SKILL.md.
  • Boundary markers: The skill advocates for a 'File-Based Sub-Agent Pattern' using markdown templates to structure and separate data.
  • Capability inventory: The instruction set utilizes Bash, Write, Read, and Task tools.
  • Sanitization: No specific output sanitization or filtering is defined beyond requested markdown formatting.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 08:19 PM
Security Audit — agent-trust-hub — claudish-usage