Release v2.4.0 — Initial Public Release · leocelis/ivd

IVD (Intent-Verified Development) v2.4 — First Public Release

A framework for the AI Agents era — making software understanding executable, verifiable, and durable.

Intent is the primary artifact. Code and docs derive from it.

8 core principles — from Intent is Primary to Innovation through Inversion
13 recipes — proven patterns for agents, workflows, data mapping, logging, and more
4 intent templates with a level guide (system, workflow, module, task)
MCP server with 15 tools — AI agent integration via stdio (local) or SSE (remote)
219 tests (unit + e2e), with CI running on every push and PR
Deployment support for DigitalOcean App Platform

These FDRs represent the intellectual core of v2.4:

FDR-005: Post-Implementation Verification Protocol — AI re-reads intent from disk before verifying, preventing drift between what was agreed and what was built
FDR-008: Cognitive Architecture of LLMs — the theoretical foundation explaining why IVD works: structured intent saturates the contextual knowledge channel, preventing parametric gap-filling (hallucination)
FDR-013: 2-Attempt Enrichment Trigger — same error twice = mandatory external enrichment (GitHub search, changelogs, docs) before retrying. Breaks the retry-same-approach loop
FDR-014: Constraint-Segmented Implementation — for 3+ constraints: GROUP → IMPLEMENT → RE-READ from disk → VERIFY per segment → cross-cut sweep. Addresses the lost-in-the-middle degradation pattern

See DECISIONS.md for the full decision log.

git clone https://github.com/leocelis/ivd.git
cd ivd
./mcp_server/devops/setup.sh

Add to your MCP client config:

{
  "mcpServers": {
    "ivd": {
      "command": "path/to/ivd/.venv/bin/python",
      "args": ["-m", "mcp_server"]
    }
  }
}

MIT