Semantic dependency tracking for AI-native codebases — powered by the agent's own brain. When an AI agent changes one file, Reponoesis tells you — with cryptographic certainty — what else needs reviewing, and preserves the WHY for every architectural decision.
AI agents edit individual files brilliantly. They fail at cross-file semantic consistency.
Agent changes billing from $50 -> $10 in pricing.tsx
-> Doesn't update terms/page.tsx (still says "$50/month")
-> Doesn't update STORE_LISTING.md (still says "flat $50 fee")
-> Doesn't update HomeClient.tsx (still shows the old price in the hero)
-> Next agent opens the project -> has zero context on WHY $10 was chosen
This is not a bug. It's a structural gap — no tool tracks semantic dependencies across file types, and no tool preserves the WHY of architectural decisions for future agents.
1. Agent-Brain Extraction — The AI agent (Cursor/Claude/Gemini) reads code and writes its semantic understanding back into a Merkle-anchored graph via MCP tools. No static algorithms. No regex keyword lists. The agent's intelligence IS the extractor.
2. Concept Merkle Chain — Every concept (billing_model, auth_strategy, gdpr_consent) gets SHA3-256 fingerprinted. When a file changes, the chain link invalidates deterministically — no fuzzy "maybe."
3. Dynamic Agent Explanations — AI agents write plain-English descriptions of drift directly to the database. These override cryptographic warnings, explaining the business logic changes verbatim in the developer interface.
4. Offline Contradiction Engine — Agents log conflicts between files (e.g. mismatched constants or version specs) into a local SQLite database. Pre-commit hooks gate Git commits if unresolved contradictions exist.
5. Repo Remembrance — ADRs (Architecture Decision Records) with full WHY rationale are stored and surfaced to new agent sessions via rpn_get_context. No more agent amnesia on "why did we choose X?"
View the integrity dashboard, active context desynchronization alerts, code diff mutations, and governing ADR rationale details.
Every file and ADR is rendered as a package on a dark PCB board grid, displaying semantic concept connections and active drift violations as glowing trace lines.
Browse the list of active Architectural Decision Records (ADRs), check their status, and audit governed files with agent-verified drift insights.
packages/
core/ - SQLite graph, SHA3 Merkle chain, concept indexer, agent write-back API
cli/ - rpn init | scan | check | decide | bind | why | pack | status | query | ui
mcp/ - MCP server for Cursor / Claude Code / Trae / Gemini (9 rpn_* tools)
ui/ - React visualizer circuit board HUD and blast-radius simulation sandbox
Prerequisites: Node.js 18+ and git
# 1. Clone the repo
git clone https://github.com/MathewAddala/reponoesis.git
cd reponoesis
# 2. Install dependencies & build
npm install
npm run build
# 3. Link CLI globally (so rpn command works anywhere)
cd packages/cli
npm link
cd ../..
# 4. Link MCP server globally
cd packages/mcp
npm link
cd ../..
# 5. Verify
rpn --version
rpn-mcp --help# 1. Go into your project
cd /path/to/your/project
# 2. Initialize Reponoesis
rpn init
# 3. Index your project
rpn scan
# 4. Check chain health
rpn statusAdd to your Cursor / Claude Code / Trae / Gemini mcp config:
{
"rpn": {
"command": "rpn-mcp",
"args": ["--project", "/absolute/path/to/your/project"]
}
}Your AI agent now has 9 tools:
| Tool | When | What |
|---|---|---|
rpn_get_context |
Session start | Load full project context: decisions, WHY rationale, concept map, contradictions |
rpn_impact_map |
Before editing | What will break if I change these files? |
rpn_validate |
After editing | Are all chains intact? What contradictions are active? |
rpn_record_concept |
Agent identified meaning | Write semantic understanding into the graph |
rpn_record_decision |
Agent knows the WHY | Create ADR + auto-bind files in one call |
rpn_record_violation |
Agent detects logic conflict | Write a logic contradiction and proposed fix back to SQLite |
rpn_record_drift_explanation |
Agent explains broken chain | Store custom plain-English drift explanation in the database |
rpn_query |
Search | Where does concept X live across the codebase? |
rpn_acknowledge |
Intentional change | Mark drift as deliberate |
rpn init # Initialize in current project
rpn scan # Full index
rpn check # Run before commit (also auto-runs via pre-commit hook)
rpn status # Chain health dashboard
rpn status --broken # Lists details of all broken chains and active contradictions
rpn query "billing_model" # Find all locations of a concept
rpn ui # Start the local visualizer HUD on localhost:3000
# Architecture Decisions (ADRs)
rpn decide billing_v2 \
--title "Change flat fee from $50 to $10" \
--status ACCEPTED \
--body "Chose $10 to match competitor pricing after A/B test..." \
--files src/pricing.tsx,src/terms/page.tsx # <- auto-binds, no separate rpn bind needed
rpn why src/pricing.tsx # Explain why a file is the way it is
rpn pack # Export handover JSON for next agent sessionDeveloper changes billing in pricing.tsx
|
Cursor AI / Claude sees the diff (already reading the file)
|
Agent calls: rpn_validate({ changed_files: ["abs/path/pricing.tsx"] })
-> Reponoesis re-indexes, returns: "billing_model concept affected, 3 files share it"
|
Agent uses ITS OWN intelligence to understand:
"This is a pricing change that affects terms and HomeClient"
|
Agent calls: rpn_record_decision({
label: "billing_v2",
title: "Change fee from $50 to $10",
body: "..agent's own WHY reasoning..",
files: ["pricing.tsx", "terms/page.tsx", "HomeClient.tsx"] <- auto-binds all
})
|
Reponoesis: ADR created, all 3 files Merkle-anchored -> status: CLEAN
|
New agent opens project -> calls rpn_get_context() -> immediately knows why $10
| Package | npm name | Description |
|---|---|---|
@engine/core |
(internal) | SQLite graph, SHA3 fingerprinting, Merkle chain, agent write-back API |
@engine/cli |
(publish as rpn) |
CLI tool — rpn command |
@engine/mcp |
(publish as rpn-mcp) |
MCP server — 9 rpn_* tools for AI agents |
@engine/ui |
(internal) | React-based visual circuit board visualizer |