speckit-plan

Warn

Audited by Gen Agent Trust Hub on Oct 12, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill reads and executes command strings from a local YAML configuration file ('.specify/extensions.yml') via a hook system. This allows the repository content to define arbitrary shell commands that the agent will execute.\n- [COMMAND_EXECUTION]: The workflow requires executing a local bash script ('.specify/scripts/bash/setup-plan.sh') from the repository root, inheriting the risks associated with executing scripts provided within the workspace.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from several untrusted sources, including user arguments and project configuration files.\n
  • Ingestion points: User input ($ARGUMENTS in SKILL.md), '.specify/extensions.yml' (SKILL.md), and project documentation files like FEATURE_SPEC and constitution.md.\n
  • Boundary markers: The instructions do not define clear delimiters to separate untrusted data from the agent's core planning logic.\n
  • Capability inventory: The skill can execute shell commands through hooks and setup scripts, and it performs write operations to project files (e.g., 'research.md', 'data-model.md', 'CLAUDE.md').\n
  • Sanitization: The skill uses a specific sanitization rule for escaping single quotes in arguments passed to the setup script, but this does not mitigate potential injection through other ingested data sources.\n- [METADATA_POISONING]: The 'author' field in the metadata ('github-spec-kit') does not align with the identified vendor 'mixpanel' and its recognized resource patterns.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 12, 2026, 03:24 PM