Architecture diagram: version, agent count (66→63), MCP tool count (213→0) and AgentDB version wrong; Jujutsu VCS and Consensus Protocols not in the package
Nobody has claimed this yet.
- Dominant language
- TypeScript
- Stars
- 812
- Forks
- 175
- Avg merge
- 2m
- Merged PRs (30d)
- 3
Description
Summary
The architecture diagram states several countable facts that do not match the shipped package. This issue covers only the checkable numbers and the two components that aren't in the package — the functional defects in the three middle layers are already filed separately and are listed at the bottom as cross-references, not re-reported here.
Measured against agentic-flow 2.1.2 (npm latest, global install), Node 22, macOS 15.6.
1. Version numbers
| diagram | actual |
|---|---|
Agentic-Flow v2.0.0 |
npm latest = 2.1.2; repo main package.json = 2.0.2-alpha |
AgentDB@alpha v2.0.0-alpha.2.11 |
installed and declared in package.json as 3.0.0-alpha.20 |
No current artifact is 2.0.0. The AgentDB line is the more confusing one: it names a 2.0.0-alpha.2.11 that the package does not depend on, while the real dependency is a different major (3.0.0-alpha.20).
2. Agent count: 66 → 63
$ agentic-flow --list | grep -cE "^\s+[a-z]"
63
Separately worth noting, since it produces a third number: the worker-integration capability table is only 6 entries (AGENT_CAPABILITIES in dist/workers/worker-agent-integration.js:14), which is what workers integration reports as "Total Agents" (#210). So the package presents 6, 63 and 66 in three places.
3. MCP Tools: 213 → 0 by default
$ agentic-flow mcp list
No MCP servers configured
💡 Add a server with: npx agentic-flow mcp add NAME --npm PACKAGE
A fresh install exposes no MCP tools and no bundled servers. If 213 is the total across servers a user might optionally add, the diagram should say so — as drawn it sits inside the product box as a shipped capability.
4. Two "Supporting Systems" are not in the package
Jujutsu VCS / Quantum-Safe — nothing matching *jujutsu* or jj-agent exists anywhere in the installed 2.1.2 tree, and the declared bins are only:
{"agentic-flow":"dist/cli-proxy.js","agentdb":"dist/cli/agentdb-proxy.js","agentic-flow-repair":"dist/repair/cli.js"}
It was shipped once — agentic-flow@2.0.0-alpha declares jj-agent and agentic-jujutsu bins and carries packages/agentic-jujutsu. So this was removed somewhere between 2.0.0-alpha and 2.1.2 while remaining on the diagram.
Consensus Protocols — no path matching *consensus* under dist/.
For contrast, the other two supporting entries are present and I'm not disputing them: agent-booster (7 files) and quic (12 files) both ship, though I haven't exercised either.
5. Unfalsifiable as drawn (not a defect, a suggestion)
Flash Attention (2.49x-7.47x), HNSW Indexing (150x-12,500x), Agent Booster (352x faster) and GNN Query Refinement (+12.4%) are ratios with no stated baseline, hardware, or dataset. A reader cannot confirm or refute them. bench:attention does report flash at ~0.07ms against a <5ms target, which is a real measurement — citing that (with its baseline) would be stronger than a bare multiplier.
The +12.4% figure is a special case: GNN cannot initialize at all in 2.1.2 (GraphNeuralNetwork not found in @ruvector/gnn — #189/#215), so whatever produced that number is not reproducible with the shipped package.
Cross-references — already filed, not re-reported here
The diagram's three middle layers are the components audited in these issues:
- Coordination Layer → AttentionCoordinator — all four methods fail; two return all-
NaNwithout erroring (#216) - Coordination Layer → Expert Routing (MoE) — returns input order with
routingScores: [null,null,null](#190, #216) - EnhancedAgentDBWrapper → 5 Attention Mechanisms — all five return K and ignore V (#191, #215)
- EnhancedAgentDBWrapper → Vector Storage —
controller.store/.retrieve/.deleteall missing (#213, #215) - EnhancedAgentDBWrapper → GraphRoPE Position Embeddings — method absent from the service (#193, #215, #216)
- EnhancedAgentDBWrapper → GNN Query Refinement — service never initializes (#189, #215)
- Supporting Systems → ReasoningBank Memory System — segfaults the host process, exit 139 (#194)
Suggested fix
Update the four numbers, drop or mark-as-planned the two absent systems, and either cite baselines for the speedup ratios or move them to a benchmark page where the methodology can live alongside them.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the architecture diagram and compare its labels against package.json, the installed package, and the commands agentic-flow --list and agentic-flow mcp list. Check dist/workers/worker-agent-integration.js:14 and the declared bins when verifying the agent and supporting-system claims. Done means the four numbers and absent systems are corrected or clearly marked, with benchmark ratios given baselines or moved to a documented benchmark page.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- nodejs, typescript
- Domain
- documentation
- Issue type
- Documentation
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 55/100