gstack-guard

GStack Guard Adapter

This is a Codex role-system adapter for Garry Tan's gstack gstack-guard method.

When To Use

Use this when:

• a role prompt needs stronger boundaries;
• the agent risks touching forbidden files or actions;
• architecture wants guardrails before delegation;
• the user explicitly invokes $gstack-guard.

Workflow

1. Keep the active role boundary. Do not expand scope just because this gstack method is useful.
2. Read the relevant repo/docs/evidence first when the task depends on current state.
3. Read ../gstack/references/methodology.md if you need the shared method map, then use the section named Engineering Execution Methods.
4. Produce: guardrails, forbidden scope, stop conditions, validation gates, and revised prompt wording.
5. Return the result in the active role's normal format, including boundaries, validation, and unresolved decisions when applicable.

Boundaries

• Treat upstream gstack as external methodology, not local-owned project state.
• Do not run upstream gstack runtime, telemetry, browser-cookie import, upgrade checks, or host routing injection automatically.
• Do not create or edit CLAUDE.md, .claude/, .agents/, or upstream routing files unless the user explicitly asks for upstream gstack installation work.
• Do not write files, commit, push, deploy, restart, migrate, clean, delete, or change production unless the active role prompt explicitly allows it.
• Preserve this repository's QA versus 测试 split: formal test cases/reports belong to 测试 and $test-case-report-builder.