Interactive walkthrough

How Govinuity works

Walk through the governed continuity loop, from surfacing a candidate decision to observing its injection in a future agent session.

Getting started

Clone and run locally

Govinuity runs on your machine. No accounts, no telemetry, no hosted service required for the current release.

$ git clone https://github.com/Taprooted/Govinuity.git

$ cd Govinuity

$ npm install

added packages

$ cp .env.example .env.local

$ npm run dev

Ready on http://localhost:3000

Step 1 — Surface

A candidate decision is surfaced

An agent, script, or human submits a candidate decision via POST /api/decisions. It enters Govinuity as a proposal — not an instruction. The included harvest script (scripts/harvest_proposals.py) automates this — reading Claude Code sessions by default, or any agent tool via --input — and extracting candidates on a schedule.

POST /api/decisions

// Candidate context, not authority yet

{

"body": "All DB migrations must be human-reviewed before production.",

"status": "proposed",

"proposal_class": "durable_constraint",

"summary_for_human": "Prevents unreviewed production migrations.",

"why_surfaced": "Repeated correction during an agent session",

"rationale": "Bad migrations are hard to reverse."

}

Step 2 — Review

A human reviews the proposal

The Review page surfaces proposed continuity objects with rationale, trigger, reversibility, scope, and conflict signals.

localhost:3000/review

Proposed — awaiting review

All DB migrations must be human-reviewed before running in production.

Submitted by agent session · durable_constraint

Trigger: Repeated correction during an agent session
Reversibility: Low
Scope: global
Conflicts: None detected

Step 3 — Ratify

Approve, defer, or reject

You decide what deserves authority. Only approved decisions become eligible for injection into future sessions. Every active decision carries a named ratifier — decisions can be superseded or archived later, but the ratification chain is always visible.

localhost:3000/review

Proposed — awaiting review

All DB migrations must be human-reviewed before running in production.

Step 4 — Inject

Ratified decisions become future context

Generate a GOVERNED_CONTINUITY.md file for Claude Code, or use the memory API to inject governed context into another agent workflow.

localhost:3000/decisions

---

generated_at: 2026-04-14T10:00:00Z

decisions_injected: 1

decisions_excluded: 0

---

### DB migration review gate

ratified_by: govinuity-review · scope: always active

All DB migrations must be human-reviewed before running in production.

# In CLAUDE.md:

@.claude/GOVERNED_CONTINUITY.md

Step 5 — Measure

Every injection is logged

Each continuity run records what was injected, what was excluded, and why. The harvest script detects outcome signals automatically from session transcripts — whether a decision was followed, whether correction was required, whether stale context leaked through — and posts them as run annotations. Continuity becomes measurable, not just observable.

localhost:3000/runs

run-1744…k3f9a Claude Code session — project-api 1 decision injected Injected

run-1744…m7c2x Claude Code session — project-web 1 injected, 1 excluded Injected

run-1744…p9b1z API call — agent/deploy-bot 0 injected, 1 excluded Scoped out

1 of 6

Use ← → arrow keys to navigate