code-yeongyu/oh-my-openagent

[Feature]: Persist Momus review verdicts to disk

Open

#3002 opened on Apr 1, 2026

View on GitHub
 (3 comments) (2 reactions) (1 assignee)TypeScript (57,825 stars) (4,692 forks)batch import
agentenhancementhelp wantedtriage:feature-request

Description

Prerequisites

  • I will write this issue in English (see our Language Policy)
  • I have searched existing issues and discussions to avoid duplicates
  • This feature request is specific to oh-my-opencode (not OpenCode core)
  • I have read the documentation or asked an AI coding agent with this project's GitHub URL loaded and couldn't find the answer

Problem Description

Momus reviews are expensive (1-2 min on gpt-5.4 xhigh, 32k thinking budget) but ephemeral. Once the session ends, the verdict is gone — no record of what was checked, what was flagged, or how many cycles a plan went through before approval.

This means no audit trail when approved plans fail during execution, no revision history for debugging multi-cycle review loops, and wasted cost when sessions compact and Momus re-reviews from scratch.

Proposed Solution

Write verdicts alongside plans. Momus stays read-only — the orchestrator writes the artifact after receiving the result.

.sisyphus/plans/
    implement-auth.md
    implement-auth.reviews/
      001-reject.md
      002-okay.md

Each file captures verdict, timestamp, model, plan hash (to detect changes between reviews), duration, and a summary of references checked.

Integration point: After task(subagent_type="momus") returns, either in the Sisyphus prompt path or as a lightweight hook. Momus needs zero changes — no new tools, no prompt modifications.

Reusable infra: Boulder state (src/features/boulder-state/storage.ts) already reads/writes .sisyphus/ to disk. Plans already live at .sisyphus/plans/*.md, so .reviews/ is a natural sibling.

Alternatives Considered

boulder.json — Compact but not human-readable. Developers want to glance at review history in their editor, not parse JSON. Markdown is git-friendly and mirrors the plan format.

Additional Context

Complementary to #1013 (parallel reviews), #1152 (multi-model quorum), and #2290 (compound learning) — faster or more accurate reviews are still ephemeral without persistence.

Feature Type

New Hook

Contribution

  • I'm willing to submit a PR for this feature
  • I can help with testing
  • I can help with documentation

Contributor guide

Tech stack
typescriptnodejs
Domain
backendtoolingdeveloper experience
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
blocked
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
TypeScriptNode.js fsKnowledge of Boulder state moduleUnderstanding of Momus review flow
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
50
Research direction
Review the Boulder state module in `src/features/boulder state/storage.ts` to understand how it reads/writes `.sisyphus/` to disk. Then examine the Sisyphus agent orchestration to locate where `task(subagent type="momus")` returns. Implement a hook after Momus returns that writes the verdict to `.sisyphus/plans/<plan name>.reviews/` following the proposed format. Ensure the implementation does not modify Momus itself.