robzilla1738/supergoal

What it solves

Supergoal eliminates the "babysitting" required when using AI coding agents. Instead of a user having to prompt the agent for every single step of a plan, Supergoal automates the entire lifecycle from deep planning and codebase reconnaissance to autonomous execution and final auditing, requiring only two human interactions: one approval of the plan and one paste of the execution command.

How it works

Supergoal operates in two primary stages: planning and execution.

1. Planning (/supergoal): The tool performs a deep analysis of the task, which includes detecting the environment (greenfield vs. brownfield), scanning the codebase, identifying risks, and decomposing the work into an adaptive number of phases. It writes detailed specifications (ROADMAP, STATE, and phase-specific files) to a namespaced directory under .supergoal/ and performs a self-critique of the plan. Once validated, it provides the user with a single /goal command.
2. Execution (/goal): When the user pastes the command, the agent enters an autonomous loop. It reads the phase specs from disk, executes the work, and verifies the results. It includes a "3-strike" self-healing mechanism: it first auto-retries, then attempts a focused fix-spec, and finally hands off to the human if it remains blocked.
3. Final Audit: After all phases are complete, the tool re-verifies the entire project against the original roadmap and the initial baseline commit to ensure all deliverables are present and no regressions were introduced.

Who it’s for

Developers using Claude Code or Codex CLI who want to automate complex, multi-step feature builds or refactors without manual step-by-step prompting.

Highlights

• Autonomous Execution: Reduces human toil to a single approval and a single paste.
• Deep Planning: Includes codebase reconnaissance, risk assessment, and adaptive phase decomposition.
• Self-Healing: Built-in 3-strike failure recovery (auto-retry $\rightarrow$ fix-spec $\rightarrow$ handoff).
• Rigorous Auditing: Final verification against the original roadmap and a baseline reference of the working tree.
• Memory Integration: Automatically reads and writes project-specific learnings to a memory directory to improve future runs.