setup-claude-md
Fail
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to "test at least the build and lint commands by running them" as part of the validation criteria in SKILL.md. These commands are dynamically identified by analyzing project configuration files such as package.json, pyproject.toml, or Cargo.toml.\n- [REMOTE_CODE_EXECUTION]: Because the skill automatically executes commands extracted from the project environment, it presents a significant risk of executing malicious code if an attacker has modified the project's build or lint scripts. This bypasses typical user review of shell commands.\n- [PROMPT_INJECTION]: The skill processes content from several untrusted sources, including docs/project-plan-refined.md and any existing CLAUDE.md. These files provide an attack surface for indirect prompt injection, where malicious instructions embedded in the project documentation could influence the agent's behavior during the generation or validation phase.\n
- Ingestion points: package.json, pyproject.toml, Cargo.toml, docs/project-plan-refined.md, and CLAUDE.md.\n
- Boundary markers: Absent. The instructions do not specify any delimiters or warnings to ignore instructions found within these project files.\n
- Capability inventory: Bash (ls, find, cat), Write, and the permission to run discovered shell commands.\n
- Sanitization: None. The agent is encouraged to use the commands found in the codebase as ground truth for verification.
Recommendations
- AI detected serious security threats
Audit Metadata