agentic-flow 3.2.0 → 3.3.0

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

@@ -1,6 +1,6 @@
 ---
 name: agentic-flow-ecosystem
-description: Use agentic-flow's optional ruvnet integrations - aidefence prompt-injection screening, @ruvector/mincut-wasm task partitioning, sublinear-time-solver PageRank and linear solves, agent-booster code edits, ONNX WASM embeddings and the native HNSW router. Use when splitting work between agent teams, ranking files or agents in a dependency graph, screening untrusted text for prompt injection, solving a linear system, or checking which integrations are installed.
+description: Use agentic-flow's optional ruvnet integrations - aidefence prompt-injection screening, @ruvector/mincut-wasm task partitioning, @ruvnet/sublinear-core PageRank, personalized PageRank and linear solves, agent-booster code edits, ONNX WASM embeddings and the native HNSW router. Use when splitting work between agent teams, ranking files or agents in a dependency graph, screening untrusted text for prompt injection, solving a linear system, or checking which integrations are installed.
 ---
 
 # agentic-flow ecosystem integrations (Codex)
@@ -15,8 +15,8 @@ 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) |
-| Rank agents, files or tasks by influence in a graph | `sublinear_pagerank` | `sublinear-time-solver` | Gauss-Seidel |
-| Solve a diagonally dominant linear system | `sublinear_solve` | `sublinear-time-solver` | Gauss-Seidel |
+| 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` | — | — |
 
@@ -35,8 +35,9 @@ All five computation tools are side-effect free and available under the default
 
 ## 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.
-- `sublinear_solve({ matrix, vector })`: dense rows or `{ size, entries: [{ row, col, value }] }`. Every answer is checked against its residual; if sublinear-time-solver's answer fails the check, Gauss-Seidel answers instead and `libraryRejected` says why. `converged: false` means no engine met the tolerance (usually: the matrix is not diagonally dominant).
+- `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`).
+- Personalized: add `seeds: ["file-a"]` (or `{ "file-a": 2, "file-b": 1 }`) to rank what matters to those nodes, e.g. the files most relevant to the one being edited. On a large graph add `approximate: true, top: 10`: local forward push, whose work depends on `epsilon` (default 1e-6), not on the graph size; `residualMass` bounds the total L1 error and only the nodes it reached are returned (`top` caps them). `approximate` without `seeds` is rejected.
+- `sublinear_solve({ matrix, vector, method })`: dense rows or `{ size, entries: [{ row, col, value }] }`; `method` defaults to `auto` (conjugate gradient for symmetric matrices with a positive diagonal, else Gauss-Seidel; also `jacobi`, `neumann`). Every answer is checked against its residual `max|Mx - b|`; if the core package's answer fails the check, Gauss-Seidel answers instead and `libraryRejected` says why. `converged: false` means no engine met the tolerance (usually: the matrix is not diagonally dominant).
 - The bundled `sublinear/*` agents (matrix-optimizer, pagerank-analyzer, consensus-coordinator, performance-optimizer, trading-predictor) can use these two tools instead of a separate sublinear-time-solver MCP server.
 
 ## Embeddings and routing (library, no tool)

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

@@ -1,6 +1,6 @@
 ---
 name: agentic-flow-ecosystem
-description: Use agentic-flow's optional ruvnet integrations - aidefence prompt-injection screening, @ruvector/mincut-wasm task partitioning, sublinear-time-solver PageRank and linear solves, agent-booster code edits, ONNX WASM embeddings and the native HNSW router. Use when splitting work between agent teams, ranking files or agents in a dependency graph, screening untrusted text for prompt injection, solving a linear system, or checking which integrations are installed.
+description: Use agentic-flow's optional ruvnet integrations - aidefence prompt-injection screening, @ruvector/mincut-wasm task partitioning, @ruvnet/sublinear-core PageRank, personalized PageRank and linear solves, agent-booster code edits, ONNX WASM embeddings and the native HNSW router. Use when splitting work between agent teams, ranking files or agents in a dependency graph, screening untrusted text for prompt injection, solving a linear system, or checking which integrations are installed.
 version: 1.0.0
 invocable: true
 author: agentic-flow
@@ -21,8 +21,8 @@ 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) |
-| Rank agents, files or tasks by influence in a graph | `sublinear_pagerank` | `sublinear-time-solver` | Gauss-Seidel |
-| Solve a diagonally dominant linear system | `sublinear_solve` | `sublinear-time-solver` | Gauss-Seidel |
+| 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` | — | — |
 
@@ -41,8 +41,9 @@ All five computation tools are side-effect free and available under the default
 
 ## 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.
-- `sublinear_solve({ matrix, vector })`: dense rows or `{ size, entries: [{ row, col, value }] }`. Every answer is checked against its residual; if sublinear-time-solver's answer fails the check, Gauss-Seidel answers instead and `libraryRejected` says why. `converged: false` means no engine met the tolerance (usually: the matrix is not diagonally dominant).
+- `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`).
+- Personalized: add `seeds: ["file-a"]` (or `{ "file-a": 2, "file-b": 1 }`) to rank what matters to those nodes, e.g. the files most relevant to the one being edited. On a large graph add `approximate: true, top: 10`: local forward push, whose work depends on `epsilon` (default 1e-6), not on the graph size; `residualMass` bounds the total L1 error and only the nodes it reached are returned (`top` caps them). `approximate` without `seeds` is rejected.
+- `sublinear_solve({ matrix, vector, method })`: dense rows or `{ size, entries: [{ row, col, value }] }`; `method` defaults to `auto` (conjugate gradient for symmetric matrices with a positive diagonal, else Gauss-Seidel; also `jacobi`, `neumann`). Every answer is checked against its residual `max|Mx - b|`; if the core package's answer fails the check, Gauss-Seidel answers instead and `libraryRejected` says why. `converged: false` means no engine met the tolerance (usually: the matrix is not diagonally dominant).
 - The bundled `sublinear/*` agents (matrix-optimizer, pagerank-analyzer, consensus-coordinator, performance-optimizer, trading-predictor) can use these two tools instead of a separate sublinear-time-solver MCP server.
 
 ## Embeddings and routing (library, no tool)

package/CHANGELOG.md

@@ -2,6 +2,31 @@
 
 All notable changes to this project will be documented in this file.
 
+## [3.3.0] - 2026-11-05
+
+Replaces the optional `sublinear-time-solver` with its zero-dependency core, `@ruvnet/sublinear-core@^1.0.0`, and drops the workarounds 3.2.0 needed for sublinear-time-solver 1.7.2.
+
+### Added
+- **`sublinear_pagerank`: personalized PageRank and forward-push top-k.** `seeds` (node ids, or `{ node: weight }`) ranks what matters to those nodes. `approximate: true` (needs `seeds`; `epsilon`, default 1e-6) runs local forward push, whose work depends on `epsilon` rather than the graph size; it returns only the nodes it reached (`top` caps them) and `residualMass`, an upper bound on the total L1 error. Results now carry `method` (`power-iteration`, `forward-push`, `gauss-seidel`) and `personalized`. `damping` is the probability of following an edge (restart probability r maps to `1 - r`).
+- **`sublinear_solve`: `method`** (`auto` default — conjugate gradient for symmetric matrices with a positive diagonal, else Gauss-Seidel — or `gauss-seidel`, `conjugate-gradient`, `jacobi`, `neumann`). Results carry `method` and the real `iterations` (3.2.0 reported 0 for library answers).
+- `agentic-flow/integrations` also exports `loadSublinearCore` and the `PageRankOptions`, `PageRankEdge`, `SolveMethod`, `SublinearEngine` types.
+
+### Changed
+- `engine` in `sublinear_solve` / `sublinear_pagerank` results is `sublinear-core` (was `sublinear-time-solver`); `builtin` is unchanged. `ecosystem_status` lists `@ruvnet/sublinear-core`.
+- Every library answer is still verified before it is returned: solves by `max|Mx - b|` recomputed here, PageRank by convergence and scores summing to 1 (forward push: estimates plus residual mass summing to 1). A failed check falls back to the built-in Gauss-Seidel and sets `libraryRejected`, as before.
+
+### Security
+- **`@ruvector/attention` `^0.1.4` → `^2.2.3`** ([GHSA-p62c-q7fc-qmg9](https://github.com/ruvnet/ruvector/security/advisories/GHSA-p62c-q7fc-qmg9)). The old range resolved 0.1.32, where a key/query length mismatch panicked in Rust and aborted Node (SIGABRT); 2.2.3 throws a catchable error. The runtime API is unchanged (same 39 exports and methods); outputs match 0.1.32 except float rounding (hyperbolic ≤ 1.4e-3, MoE ≤ 1e-6). `agentdb` and `ruvector` still declare `^0.1.x` and keep their own nested copy until they release the same bump.
+- **`@ruvector/sona` `^0.1.4` → `^0.1.10`.** The old range resolved 0.1.8 in the lockfile, where JS input could abort Node — e.g. `applyMicroLora` / `applyBaseLora` with a vector that is not `hiddenDim` long; 0.1.9+ throws. SONA engines now come only from `sona-loader` (`RuVectorIntelligence` included), which also checks those lengths and throws a `RangeError`, so the error is the same even if an older nested copy is resolved.
+
+### Removed
+- The 1.7.2 workarounds: forward-push-only method selection, the solver-instance cache and WASM-probe settling, and the console-to-stderr redirection around the import (`withStdoutToStderr`, `importOptional`'s `quietStdout` option). The core prints nothing.
+
+### Install size
+Measured with `npm install agentic-flow@<version>` in an empty directory (Linux x64, Node 22):
+- 3.2.0: 1,796,535,838 bytes (1.8 GB), 600 packages. 3.3.0: 1,795,364,813 bytes (1.8 GB), 597 packages — 1.2 MB and 3 packages (`sublinear-time-solver`, its `commander`, `temporal-lead-solver`) fewer.
+- On its own, `sublinear-time-solver` installs 122 MB in 170 packages; `@ruvnet/sublinear-core` installs 220 kB in 1. In agentic-flow's tree nearly all of sublinear-time-solver's dependencies are shared with agentic-flow's own, so the "~115 MB" in the 3.2.0 notes below was its standalone cost, not what it added to an agentic-flow install.
+
 ## [3.2.0] - 2026-11-05
 
 Wires in six optional ruvnet packages that were missing, undeclared or listed but unused. Each is an `optionalDependency` (installed by default), loaded lazily, with a built-in fallback, and each result reports which engine answered.

package/README.md

@@ -362,7 +362,7 @@ additionally needs the official `swebench` Docker harness — see
 - **Performance**: `benchmark_run`, `bottleneck_analyze`, `token_usage`
 - **And 200+ more tools!**
 
-### 🔌 Ecosystem integrations (3.2.0)
+### 🔌 Ecosystem integrations (3.2.0, solver core in 3.3.0)
 
 Optional ruvnet packages, installed by default and loaded lazily, each with a built-in fallback (`npm install --omit=optional` skips them):
 
@@ -370,7 +370,7 @@ 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` |
-| [`sublinear-time-solver`](https://www.npmjs.com/package/sublinear-time-solver) | verified linear solves and PageRank | `sublinear_solve`, `sublinear_pagerank` |
+| [`@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` | — |