agentic-flow 3.4.4 → 3.5.1

package/.agents/skills/agentic-flow-ecosystem/SKILL.md

@@ -15,12 +15,13 @@ agentic-flow 3.2 ships in-process integrations with six optional ruvnet packages
 |---|---|---|---|
 | Is a prompt or retrieved text trying to hijack the agent? | `prompt_injection_scan` | `aidefence` | built-in patterns |
 | Split a task-dependency graph between two agent teams | `swarm_partition_tasks` | `@ruvector/mincut-wasm` (Node >= 22) | Stoer-Wagner (exact, up to 400 tasks; always used on Node 20) |
+| Order a task DAG into parallel waves and find the bottleneck chain | `swarm_schedule_tasks` | `@ruvector/rudag` | Kahn sort + longest path |
 | Rank agents, files or tasks by influence in a graph | `sublinear_pagerank` | `@ruvnet/sublinear-core` | Gauss-Seidel |
 | Solve a diagonally dominant linear system | `sublinear_solve` | `@ruvnet/sublinear-core` | Gauss-Seidel |
 | Apply a code edit without an LLM round-trip | `agent_booster_edit_file`, `agent_booster_batch_edit` | `agent-booster` | `npx agent-booster@0.2.2` |
 | See which packages are installed | `ecosystem_status` | — | — |
 
-All five computation tools are side-effect free and available under the default `readonly` MCP profile. The `agent_booster_*` tools write files and need the `admin` profile.
+All six computation tools are side-effect free and available under the default `readonly` MCP profile. The `agent_booster_*` tools write files and need the `admin` profile.
 
 ## Prompt injection
 
@@ -33,6 +34,10 @@ All five computation tools are side-effect free and available under the default
 
 `swarm_partition_tasks({ tasks, dependencies: [{ from, to, weight }] })` returns `groups` (two task lists), `cutEdges` (the dependencies the teams must coordinate on), `cutWeight` and `isolated` (tasks with no dependencies — schedule them anywhere). Weight is coupling strength (shared files, API contracts), default 1. The cut is the weakest seam, not a balanced split: one loosely attached task against the rest is a valid answer.
 
+## Task scheduling
+
+`swarm_schedule_tasks({ tasks: [{ id, cost }], dependencies: [{ from, to }] })` (`from` must finish before `to` starts) returns `order`, `waves` (tasks with the same earliest wave can go to separate agents at once), `earliestStart` per task, `criticalPath` (`tasks` and `cost`: the chain that bounds the run; only shortening it shortens the run), `totalCost` and `parallelism` (`totalCost / criticalPath.cost`, the speed-up ceiling from more agents). `cost` is any additive unit (minutes, tokens), default 1. A cycle is rejected with the cycle named. Use it after `swarm_partition_tasks` to order work inside each team.
+
 ## PageRank and linear solves
 
 - `sublinear_pagerank({ edges: [{ from, to, weight }], damping: 0.85, top: 10 })`: scores sum to 1. Use it to find the most depended-on files or the agent whose output most others consume. `damping` is the probability of following an edge (a restart probability r maps to `1 - r`).

package/.claude/skills/agentic-flow-ecosystem/skill.md

@@ -7,6 +7,7 @@ author: agentic-flow
 capabilities:
   - prompt_injection_screening
   - task_partitioning
+  - task_scheduling
   - graph_ranking
   - linear_solves
 ---
@@ -21,12 +22,13 @@ agentic-flow 3.2 ships in-process integrations with six optional ruvnet packages
 |---|---|---|---|
 | Is a prompt or retrieved text trying to hijack the agent? | `prompt_injection_scan` | `aidefence` | built-in patterns |
 | Split a task-dependency graph between two agent teams | `swarm_partition_tasks` | `@ruvector/mincut-wasm` (Node >= 22) | Stoer-Wagner (exact, up to 400 tasks; always used on Node 20) |
+| Order a task DAG into parallel waves and find the bottleneck chain | `swarm_schedule_tasks` | `@ruvector/rudag` | Kahn sort + longest path |
 | Rank agents, files or tasks by influence in a graph | `sublinear_pagerank` | `@ruvnet/sublinear-core` | Gauss-Seidel |
 | Solve a diagonally dominant linear system | `sublinear_solve` | `@ruvnet/sublinear-core` | Gauss-Seidel |
 | Apply a code edit without an LLM round-trip | `agent_booster_edit_file`, `agent_booster_batch_edit` | `agent-booster` | `npx agent-booster@0.2.2` |
 | See which packages are installed | `ecosystem_status` | — | — |
 
-All five computation tools are side-effect free and available under the default `readonly` MCP profile. The `agent_booster_*` tools write files and need the `admin` profile.
+All six computation tools are side-effect free and available under the default `readonly` MCP profile. The `agent_booster_*` tools write files and need the `admin` profile.
 
 ## Prompt injection
 
@@ -39,6 +41,10 @@ All five computation tools are side-effect free and available under the default
 
 `swarm_partition_tasks({ tasks, dependencies: [{ from, to, weight }] })` returns `groups` (two task lists), `cutEdges` (the dependencies the teams must coordinate on), `cutWeight` and `isolated` (tasks with no dependencies — schedule them anywhere). Weight is coupling strength (shared files, API contracts), default 1. The cut is the weakest seam, not a balanced split: one loosely attached task against the rest is a valid answer.
 
+## Task scheduling
+
+`swarm_schedule_tasks({ tasks: [{ id, cost }], dependencies: [{ from, to }] })` (`from` must finish before `to` starts) returns `order`, `waves` (tasks with the same earliest wave can go to separate agents at once), `earliestStart` per task, `criticalPath` (`tasks` and `cost`: the chain that bounds the run; only shortening it shortens the run), `totalCost` and `parallelism` (`totalCost / criticalPath.cost`, the speed-up ceiling from more agents). `cost` is any additive unit (minutes, tokens), default 1. A cycle is rejected with the cycle named. Use it after `swarm_partition_tasks` to order work inside each team.
+
 ## PageRank and linear solves
 
 - `sublinear_pagerank({ edges: [{ from, to, weight }], damping: 0.85, top: 10 })`: scores sum to 1. Use it to find the most depended-on files or the agent whose output most others consume. `damping` is the probability of following an edge (a restart probability r maps to `1 - r`).

package/CHANGELOG.md

@@ -2,6 +2,25 @@
 
 All notable changes to this project will be documented in this file.
 
+## [3.5.1] - 2026-11-09
+
+### Fixed
+- **MCP servers report the real version.** `serverInfo.version` of the stdio server that `agentic-flow mcp` starts was a hard-coded `1.0.8`; the in-process SDK server and the FastMCP stdio/HTTP servers said `1.0.0`/`1.1.12`. All now read the installed package version (`mcpServerVersion()` in `src/utils/package-version.ts`). `tests/mcp/server-version.test.ts` starts the stdio server and checks the initialize response (red on 3.5.0: `1.0.8`, `1.0.0`).
+- **`src/core/attention-native.ts` runs on @ruvector/attention 2.x.** Every class failed on 2.x: `FlashAttention(dim)`, `HyperbolicAttention(dim)` and `MoEAttention(dim)` did not construct (2.x needs a block size, a curvature and a `{ dim, numExperts, topK }` config), and `MultiHeadAttention`/`LinearAttention` passed flat arrays where 2.x takes key and value rows. The wrappers now use the 2.x shapes, accept one vector or a list of rows for keys and values, check dimensions, and honour the mask (MultiHead) and `topK` (MoE). `HyperbolicAttention.forward` returns the Poincare distance to the nearest key; `MoEAttention.forward` returns empty `expertWeights` (2.x does not expose them). `tests/integrations/attention-native-wrappers.test.ts` loads the real package and checks FlashAttention against exact softmax attention (7/8 fail on 3.5.0, 8/8 now). The live path (`RuVectorIntelligence`) was fixed in 3.3.1.
+
+### Removed
+- `src/core/agentdb-wrapper.ts` (and its README and two test files). It was not reachable through the package `exports` map and called a reflexion API agentdb never had (`controller.store` with raw vectors), so 8 of its unit tests and 15 of its 19 integration tests failed against agentdb 3.7. Use `AgentDBFast` or agentdb directly.
+- `src/agentdb/tests/frontier-features.test.ts`: it tested the vendored, unshipped copy of agentdb's `CausalMemoryGraph`/`ExplainableRecall` (excluded from the build); 7 of its tests failed on behaviour agentdb itself has since changed or fixed (for example `INSERT OR IGNORE` for provenance sources).
+
+## [3.5.0] - 2026-11-09
+
+### Added
+- **Task-graph scheduling: `swarm_schedule_tasks` MCP tool and `scheduleTaskGraph()` in `agentic-flow/integrations`.** Given tasks with costs and `{ from, to }` dependencies, returns a dependency-respecting `order`, parallel `waves` (tasks that can go to separate agents at once), `earliestStart` per task, the cost-weighted `criticalPath` (the chain that bounds the run), `totalCost` and `parallelism` (`totalCost / criticalPath.cost`). Dependency cycles are rejected with the cycle named. Complements `swarm_partition_tasks`: partition picks the seam between two teams, scheduling orders the work inside them. Admitted under the default `readonly` MCP profile (pure computation).
+- `@ruvector/rudag` `^0.1.2` as an optionalDependency (Rust/WASM topological sort and critical path, MIT OR Apache-2.0, adds `idb`; 2 packages, no install scripts). Loaded lazily with `storage: null`, so it never touches IndexedDB or the filesystem. Without it a built-in Kahn sort + longest-path DP answers with the same result (`engine: 'builtin'`); `tests/integrations/task-schedule.test.ts` checks both engines against the real package and that they agree on a 300-task random DAG.
+
+### Changed
+- Package metadata (ADR-356, folded in from #261): `homepage` is https://ruv.io, `repository.directory` is `agentic-flow`, 16 keywords, a shorter description. The README gains an animated flow diagram, a "Works with / Used by" section and a ruv.io footer.
+
 ## [3.4.4] - 2026-11-08
 
 ### Fixed

package/README.md

@@ -1,5 +1,9 @@
 # 🚀 Agentic-Flow v2 — the Agentic Meta-Harness
 
+<!-- weaver:seo:flow:start -->
+![agentic-flow: query → LLM routing → agent harness → verified result](https://raw.githubusercontent.com/ruvnet/agentic-flow/main/agentic-flow/assets/agentic-flow-flow.svg)
+<!-- weaver:seo:flow:end -->
+
 > **Freeze the model, evolve the harness.** Agentic-Flow is an open _agentic meta-harness_: a runtime whose product is the **harness around a model**, not the model. It routes each query to the cost-optimal model, evolves its own harness and autonomously repairs code, then orchestrates 66 self-learning agents, 213 MCP tools, and multi-agent swarms on top.
 
 [![npm version](https://badge.fury.io/js/agentic-flow.svg)](https://www.npmjs.com/package/agentic-flow)
@@ -370,12 +374,28 @@ Optional ruvnet packages, installed by default and loaded lazily, each with a bu
 |---|---|---|
 | [`aidefence`](https://www.npmjs.com/package/aidefence) | prompt-injection screening of every routed task (`AGENTIC_FLOW_PROMPT_INJECTION`) | `prompt_injection_scan` |
 | [`@ruvector/mincut-wasm`](https://www.npmjs.com/package/@ruvector/mincut-wasm) | split a task-dependency graph between two agent teams (Node >= 22; built-in fallback on 20) | `swarm_partition_tasks` |
+| [`@ruvector/rudag`](https://www.npmjs.com/package/@ruvector/rudag) | order a task DAG into parallel waves and find its critical path (3.5.0; Kahn + longest-path fallback) | `swarm_schedule_tasks` |
 | [`@ruvnet/sublinear-core`](https://www.npmjs.com/package/@ruvnet/sublinear-core) | verified linear solves, PageRank, personalized PageRank and forward-push top-k (zero dependencies) | `sublinear_solve`, `sublinear_pagerank` |
 | [`agent-booster`](https://www.npmjs.com/package/agent-booster) | in-process code edits | `agent_booster_edit_file` (admin) |
 | [`@ruvector/router`](https://www.npmjs.com/package/@ruvector/router) | native HNSW index for `SemanticRouter` | — |
 | [`ruvector-onnx-embeddings-wasm`](https://www.npmjs.com/package/ruvector-onnx-embeddings-wasm) | all-MiniLM-L6-v2 embeddings for `AgentDBFast` | — |
 
-`ecosystem_status` reports what is installed; `import { partitionTaskGraph, pageRank, scanPromptInjection } from 'agentic-flow/integrations'` gives the same functions in code. Skills: `.claude/skills/agentic-flow-ecosystem` (Claude Code) and `.agents/skills/agentic-flow-ecosystem` (Codex), both installed by `agentic-flow init`.
+`ecosystem_status` reports what is installed; `import { partitionTaskGraph, scheduleTaskGraph, pageRank, scanPromptInjection } from 'agentic-flow/integrations'` gives the same functions in code. Skills: `.claude/skills/agentic-flow-ecosystem` (Claude Code) and `.agents/skills/agentic-flow-ecosystem` (Codex), both installed by `agentic-flow init`.
+
+**Schedule a swarm's tasks (3.5.0).** `from` must finish before `to` starts; `cost` is any additive unit (minutes, tokens), default 1. Tasks in the same wave can go to separate agents at once, and only shortening the critical path shortens the run:
+
+```js
+import { scheduleTaskGraph } from 'agentic-flow/integrations';
+
+const s = await scheduleTaskGraph(
+  [{ id: 'design', cost: 5 }, { id: 'api', cost: 3 }, { id: 'impl', cost: 2 }, { id: 'review', cost: 1 }, { id: 'docs', cost: 1 }],
+  [{ from: 'design', to: 'impl' }, { from: 'api', to: 'impl' }, { from: 'impl', to: 'review' }, { from: 'api', to: 'docs' }],
+);
+console.log(s.engine, JSON.stringify(s.waves), JSON.stringify(s.criticalPath), s.parallelism);
+// rudag [["design","api"],["impl","docs"],["review"]] {"tasks":["design","impl","review"],"cost":8} 1.5
+```
+
+The MCP tool `swarm_schedule_tasks` takes the same `{ tasks, dependencies }` and also returns `order`, `earliestStart` and `totalCost`. A dependency cycle is rejected with the cycle named. Without `@ruvector/rudag` the result is identical and `engine` is `builtin`.
 
 ### 🧩 Advanced Capabilities
 
@@ -2251,3 +2271,18 @@ npm install agentic-flow@3.2.0
 **Grade: A+ (Perfect Integration)**
 **Status: Production Ready**
 **Last Updated: 2025-12-03**
+
+<!-- weaver:seo:links:start -->
+## Works with / Used by
+
+Part of the [ruv.io](https://ruv.io) open-source stack for AI agents, LLM routing, agent orchestration.
+
+- **Works with** (its runtime dependencies): [`@metaharness/router`](https://www.npmjs.com/package/@metaharness/router), [`@ruvector/edge-full`](https://www.npmjs.com/package/@ruvector/edge-full), [`@ruvector/router`](https://www.npmjs.com/package/@ruvector/router), [`@ruvector/tiny-dancer`](https://www.npmjs.com/package/@ruvector/tiny-dancer), [`agentdb`](https://www.npmjs.com/package/agentdb), [`ruvector`](https://www.npmjs.com/package/ruvector), [`ruvector-onnx-embeddings-wasm`](https://www.npmjs.com/package/ruvector-onnx-embeddings-wasm), [`@ruvector/attention`](https://www.npmjs.com/package/@ruvector/attention), [`@ruvector/mincut-wasm`](https://www.npmjs.com/package/@ruvector/mincut-wasm), [`@ruvector/rudag`](https://www.npmjs.com/package/@ruvector/rudag), [`@ruvector/sona`](https://www.npmjs.com/package/@ruvector/sona), [`@ruvnet/sublinear-core`](https://www.npmjs.com/package/@ruvnet/sublinear-core), [`agent-booster`](https://www.npmjs.com/package/agent-booster), [`aidefence`](https://www.npmjs.com/package/aidefence)
+- **Used by** (published packages that depend on it): [`@agentic-robotics/cli`](https://www.npmjs.com/package/@agentic-robotics/cli), [`@agentic-robotics/mcp`](https://www.npmjs.com/package/@agentic-robotics/mcp), [`@agentic-robotics/self-learning`](https://www.npmjs.com/package/@agentic-robotics/self-learning), [`@claude-flow/browser`](https://www.npmjs.com/package/@claude-flow/browser), [`@claude-flow/cli`](https://www.npmjs.com/package/@claude-flow/cli), [`@claude-flow/embeddings`](https://www.npmjs.com/package/@claude-flow/embeddings), [`@claude-flow/integration`](https://www.npmjs.com/package/@claude-flow/integration), [`@claude-flow/plugin-agent-federation`](https://www.npmjs.com/package/@claude-flow/plugin-agent-federation), [`@foxruv/e2b-runner`](https://www.npmjs.com/package/@foxruv/e2b-runner), [`@foxruv/iris`](https://www.npmjs.com/package/@foxruv/iris), [`@foxruv/iris-ultrathink`](https://www.npmjs.com/package/@foxruv/iris-ultrathink), [`@foxruv/nova-medicina`](https://www.npmjs.com/package/@foxruv/nova-medicina)
+<!-- weaver:seo:links:end -->
+
+<!-- weaver:seo:footer:start -->
+---
+
+Built by [rUv](https://ruv.io) · [ruv.io](https://ruv.io) · [Source](https://github.com/ruvnet/agentic-flow)
+<!-- weaver:seo:footer:end -->