Decompose DAG: an epistemic scaffold for vibe coding. Claims graph, MCP server for agents, live dashboard for humans.
View on GitHub ↗An epistemic scaffold for vibe coding. DDAG turns a build target into a graph of claims, makes a coding agent verify them bottom-up against evidence, and keeps an honest, replayable record of every decision, including the ones that turned out wrong. Agents use it through an MCP server; humans watch it through a live dashboard.
Coding with an agent has two failure modes that a todo list cannot catch:
1. Sloppy product. The agent reports "done" on things it never checked, and nothing forces it to say what evidence it looked at. 2. Audits that never end. Every audit pass finds new issues, because the audit is a report, not a plan: nobody can say which properties have been examined by which method, and which have not.
DDAG answers both with one structure. The target is the root claim. Claims are decomposed into parts (arcs point part → whole). A claim can only be judged when all of its parts are solid, so the agent has to work bottom-up. Every judgment carries evidence and is pinned to the git commit and the hashes of the files it cites; when those files change, the judgment is marked stale on the dashboard. Findings are recorded as issues on the claim they refute, with full detail, and a working version is recorded after the commit that concluded it. The root is solid only when every claim under it has been verified. The whole history is an append-only chain of events, replayed deterministically, so the process itself can be read back.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/DreamC0der-AI--ddag.html)
ChatGPT thinks. Codex works. Use ChatGPT as the planning brain while keeping the Codex harness.
shinthink/blitzstrike⚡ Blitz Strike — a universal MCP penetration-testing toolbelt. Structured methodology: reconnaissance & attack-surface mapping, source-to-sink analysis, and live validation. 57 escalation chains, 130-
0xwilliamortiz/openclaude-improvedruns anywhere. uses anything
halofyai/halofyHalofy is the open access and governance layer for AI agents across your organization. Identity, policy, provenance, audit, and signed erasure.
CaptureGrubEnchant/SolidWorksSolidWorks MCP Server connects an AI assistant to a running SolidWorks instance. Sketch, extrude, fillet, export STEP/STL, generate macros.
arielshad/3d-asset-server3d assets api and mcp
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.