gsd-discuss-phase
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill defines a shell function
gsd_runthat executes a local Node.js script (gsd-tools.cjs) used to query workflow configurations. The script searches for this shim in project-local directories and the user's home directory. - [EXTERNAL_DOWNLOADS]: The skill provides an installation command for the
@opengsd/gsd-corepackage from the npm registry in the event that required tools are not found. This is a standard installation instruction for the framework's own dependencies. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data from project documentation (PROJECT.md, REQUIREMENTS.md, STATE.md) and user-supplied arguments to adapt its questioning strategy. While this ingests untrusted data, the skill uses these as context for a structured decision-making process rather than direct instructions.
- [DYNAMIC_EXECUTION]: Based on user arguments or configuration, the skill dynamically selects and instructs the agent to execute specific supplementary workflow files located in
~/.claude/gsd-core/workflows/. This is a standard design pattern for the open-gsd framework to maintain modularity.
Audit Metadata