speckit-plan

Warn

Audited by Gen Agent Trust Hub on Aug 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local shell script at .specify/scripts/bash/setup-plan.sh from the repository root during initialization.
  • [COMMAND_EXECUTION]: The skill implements a dynamic hook system that reads command names from a configuration file (.specify/extensions.yml) and instructs the agent to execute them using the EXECUTE_COMMAND directive. This facilitates arbitrary command execution based on project-local content.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting data from configuration files and script outputs to drive agent behavior and command execution. Ingestion points: Reads from .specify/extensions.yml and the JSON output of the setup script. Boundary markers: None. The instructions do not include delimiters or warnings to treat this external data as untrusted. Capability inventory: The skill can perform file system writes (e.g., to CLAUDE.md, research.md), execute shell scripts, and trigger arbitrary agent commands. Sanitization: The skill applies shell argument escaping for user arguments but lacks validation or sanitization for commands and descriptions loaded from the extension configuration.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 7, 2026, 06:22 AM
Security Audit — agent-trust-hub — speckit-plan