@exellix/ai-tasks 10.0.13 → 10.1.1

package/CHANGELOG.md

@@ -1,458 +1,479 @@
-# Changelog
-
-All notable changes to `@exellix/ai-tasks` are documented here.
-
-The published **npm version** always matches **`version`** in [`package.json`](package.json). If this file skips a semver, treat [`BREAKING-CHANGES.md`](BREAKING-CHANGES.md) and [`RUNTASK_REQUEST.md`](RUNTASK_REQUEST.md) as the source of truth for cross-version upgrades.
-
-## [10.0.13] - 2026-06-20
-
-### Changed
-
-- **`@exellix/ai-skills`** **^6.12.5** (graphenix 2.7.3 + catalox 5.9.8 train).
-- **`@x12i/catalox`** / **`@x12i/catalox-contracts`** **^5.9.8**.
-
-## [10.0.12] - 2026-06-20
-
-### Changed
-
-- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.7.3** (full format suite train).
-
-## [10.0.11] - 2026-06-20
-
-### Fixed
-
-- **`@exellix/fact-guard-*` dependencies:** publish semver **`^1.0.1`** in `dependencies` (10.0.10 accidentally shipped **`file:../fact-guard/...`** in `dependencies`, which breaks installs outside this monorepo).
-- **`prepublishOnly`:** fail fast if `file:` / `link:` / `workspace:` specs remain in `package.json` at publish time.
-
-### Changed
-
-- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.7.2** (compiler emits `webQueryTemplate` on `webScope` units; legacy `webScoping` rejected on plans).
-- **`@exellix/ai-skills`** **^6.12.4** (graphenix 2.7.2 train).
-- **Web scope validation:** enabled `webScope` PRE units require non-empty **`unitParams.webQueryTemplate`** (or `webQueryTemplates[]`); rejects legacy **`webScoping`** / NARRIX question fallbacks on unit params (`LEGACY_WEB_SCOPE_UNIT_PARAMS`, `WEB_QUERY_TEMPLATE_REQUIRED`, `WEB_QUERY_TEMPLATE_UNRESOLVED`).
-
-## [10.0.10] - 2026-06-20
-
-### Changed
-
-- **Web scoping:** migrate from **`@exellix/narrix-web-scoper`** to published **`@x12i/web-scoper@^1.1.1`**. New adapter module **`src/web-scope/`** compiles graph/task input to **`{ question, record?, options? }`** or **`searchMany`** packs; **`executionMemory.webContext`** uses **`{ ok, context?, error? }`** (replaces legacy **`available`**).
-- **Standalone PRE `webScope` unit:** playground and Graphenix plans invoke the same adapter via **`unitParams.enableWebScope`** + **`unitParams.options`**.
-- **Docs:** README, RUNTASK_REQUEST, and **`documenations/web-scoping-in-ai-tasks.md`** updated for the new contract.
-
-### Removed
-
-- Dependency on **`@exellix/narrix-web-scoper`**. Legacy **`narrix.webScoping`** / template fields are still mapped at invoke time where supported.
-
-## [10.0.9] - 2026-06-18
-
-### Changed
-
-- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.7.0** (fact-guard schema/validation/compiler train; optional `unitParams.factGuard` fully validated upstream).
-- **`@exellix/ai-skills`** **^6.12.3** (graphenix 2.7.0 train).
-
-## [10.0.8] - 2026-06-18
-
-### Changed
-
-- **`@exellix/fact-guard-*`** **^1.0.0** — re-scoped from `@x12i` to **`@exellix`** (npm publish train).
-- **`@exellix/xynthesis`** **^4.8.4**.
-
-## [10.0.7] - 2026-06-18
-
-### Changed
-
-- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.6.0** (optional `unitParams.factGuard` on sub-action units; validates against graphenix `FactGuardStepConfig`).
-- **`@exellix/ai-skills`** **^6.12.2** (graphenix 2.6.0 train).
-- **`@exellix/fact-guard-*`** **^1.0.0** (npm semver; was monorepo `file:` link).
-
-### Added
-
-- Node-plan validation tests for graphenix 2.6.0 fact-guard golden fixtures (`node-synthesis-fact-guard-*`, `node-polish-fact-guard-enabled`, `node-audit-fact-guard-nested`).
-
-## [10.0.6] - 2026-06-17
-
-### Added
-
-- **Fact-guard sub-actions (optional):** optional atom integrity checks on PRE synthesis, POST audit/polish/scoping, and execution-strategy planner/optimizer hops via **`@exellix/fact-guard-*` ^1.0.0**. Disabled by default (`factGuard.enabled: false` or absent = unchanged behavior). Kill-switch: **`FACT_GUARD_DISABLED=1`**. MAIN **`runSkill`** path unchanged.
-- **`FactGuardStepConfig`** on `SynthesisConfig`, `AuditConfig`, `PolishConfig`, `ExecutionStrategyInvocation`, and `aiScopingOptions`.
-- **`npm run test:fact-guard`** — integration tests (mocked LLM).
-
-### Changed
-
-- **`@exellix/xynthesis`** **^4.8.3** (fact-guard passthrough on `executeXynthesisAction`).
-
-## [10.0.5] - 2026-06-17
-
-### Changed
-
-- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.5.0** (CRS-FRS-003 train; jobMemory.context authoring preserved upstream).
-- **`@exellix/ai-skills`** **^6.12.1** — consumes graphenix 2.5.0 train.
-
-## [10.0.4] - 2026-06-15
-
-### Changed
-
-- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.4.0** (suite train bump from 2.3.0).
-- **`@exellix/ai-skills`** **^6.12.0** — consumes graphenix 2.4.0 train.
-
-## [10.0.2] - 2026-06-14
-
-### Added
-
-- **Graphenix 2.1.0 alignment:** `@x12i/graphenix-plan-format` and `@x12i/graphenix-executable-contracts` **^2.1.0** — rejects plans with `unitParams.synthesisInputStrategy` (`LEGACY_SYNTHESIS_INPUT_STRATEGY_ON_PLAN`).
-- **`resolveSynthesisContextStrategyKey`** — PRE synthesis template key from skill catalog (`@exellix/ai-skills` **^6.11.0**); path-driven source material via `invokeContract.reads` / `memoryPaths`.
-- Validation test for legacy synthesis field on compiled plans.
-
-### Changed
-
-- **`synthesisConfigFromUnit`** strips `unitParams.synthesisInputStrategy` from plan wire; no longer defaults plan enum to `policy`.
-- Wire fixture `graph-engine-pre-synthesis.json` — path-only synthesis unit (no `synthesisInputStrategy`).
-- **Docs:** README / RUNTASK_REQUEST — skill-catalog-only template key; graphenix **2.1.0** deps.
-
-## [10.0.1] - 2026-06-13
-
-### Added
-
-- **Graphenix 2.0.3 alignment:** `@x12i/graphenix-executable-contracts` and `@x12i/graphenix-plan-format` **^2.0.3** — orchestration strategy registry (`plan`, `split`, `merge`, `pick-best`, `webScope`), path-driven synthesis units (`invokeContract.reads`).
-- Golden tests for graphenix plan fixtures: `node-synthesis-execution-memory-only`, `node-split-merge`, `node-web-scope-standalone`.
-- **Synthesis path bridge:** `synthesisConfigFromUnit` copies `invokeContract.reads` → `memoryPaths` when omitted; defaults `synthesisInputStrategy` to `policy` for path-driven plans.
-
-### Changed
-
-- Graph authoring fixtures: removed deprecated `preInputStrategy` from `aiTaskStrategies`.
-- Wire fixture `graph-engine-pre-synthesis.json` aligned with graphenix 2.0.3 compile output (`reads`: `input`, `executionMemory`).
-- **Docs:** README Graphenix 10.x section; downstream [graph-model-change-2026-06-13.md](../temp/downstream/graph-model-change-2026-06-13.md).
-
-## [10.0.0] - 2026-06-13
-
-### Breaking
-
-- **Removed** `taskConfiguration` and `executionPipeline` from `RunTaskRequest` — use frozen `nodePlan` only.
-- **Removed** compile pipeline (`src/compile/*`, `taskConfigurationCompileError`).
-
-### Added
-
-- **PRE synthesis wiring:** `resolveRawTemplate(role: synthesisContext, strategyKey)` via `@exellix/ai-skills` **^6.10.0**.
-- **PRE/POST orchestration runtime:** `externalPreUtility` (`plan`, `split`, `webScope`), `externalPostUtility` (`merge`, `pick-best`), multi-shard MAIN + FuncX fan-in.
-- **Web scope:** Narrix preprocessor `enableWebScope` + standalone `webScope` unit.
-- **Strategy catalogs:** synthesis-input, pre/post orchestration (publish via `npm run publish:task-strategies`).
-- **nodePlan audit parity:** real `reRunMain` + synthesized `contextMarkdown`.
-
-### Changed
-
-- **`@exellix/ai-skills`** **^6.10.0** (support templates + single `professional-answer` Catalox seed).
-
-## [9.1.1] - 2026-06-13
-
-### Changed
-
-- **Run-task handoff closure:** Stale docs updated — node template scope is `nodePlan.invokeContract.taskVariable`, not `executionMemory.taskVariables`.
-- **`@exellix/ai-skills`** **^6.9.1** (WM source-of-truth alignment).
-
-### Added
-
-- Golden parity tests for all **7** node plan fixtures (`node-post-audit`, `node-input-synthesis-profile` added to `golden-fixtures.test.ts`).
-
-## [9.1.0] - 2026-06-09
-
-### Changed
-
-- **Graphenix 2.1.0 alignment:** Accept parent plan format `"executable-plan"` (v2) and trace format `"execution-trace"` (v2).
-- **Snapshot isolation:** `overlayUnitModelOnRequest` ignores concrete wire ids in `resolvedInvocationSnapshot.modelId`, preferring authored `modelSelection` (CR-1.3).
-
-## [9.0.5] - 2026-06-08
-
-### Changed
-
-- **`@x12i/funcx`** **^4.9.5** (direct dependency + npm override) — aligns PRE/POST xynthesis with funcx catalog gate via **`@x12i/ai-profiles`** (requires funcx ≥ 4.9.5; fixes subnet-style PRE `Unknown model "google/gemini-2.5-flash-lite"`).
-- **`@x12i/ai-profiles`** **^3.4.0**, **`@exellix/xynthesis`** **^4.6.5** (publish **4.6.6** for funcx bump).
-
-## [9.0.1] - 2026-06-07
-
-### Changed
-
-- Graphenix dependency alignment with latest CP release train: **`graphenix-plan-format`** and **`graphenix-trace-format`** @ **^1.1.1**; dev **`graphenix-authoring-format`** @ **^1.2.1**, **`graphenix-plan-compiler`** @ **^1.1.1**. **`graphenix-executable-contracts`** remains **^1.1.0**.
-- **`npm` overrides:** Pin transitive **`@x12i/funcx`** to **^4.6.0** (aligns with **`@exellix/xynthesis` ≥ 4.6.2** and **`@x12i/logxer` ^4.6.0**).
-
-## [9.0.0] - 2026-06-07
-
-### Changed (breaking)
-
-- **Graphenix node execution v2.1:** `runTask()` requires **`nodePlan`** (`NodeExecutionPlan`); production path is **`executeNodeFromPlan`** (unit loop + per-unit model selection). Legacy **`executionPipeline`** / **`taskConfiguration`** compile is **off the hot path** (still exported for migration tooling).
-- **Forbidden wire fields** when `nodePlan` is set: root **`modelConfig`**, **`executionPipeline`**, **`taskConfiguration`** — model slots come from `nodePlan.executionUnits[].modelSelection`.
-- **`executionTrace`** (optional): graphenix v2 trace append (`unit.*`, `model.invocation.*`).
-
-### Added
-
-- **`src/node-execution/`** — unit dispatch, Narrix units, PRE/MAIN/POST handlers, trace bridge.
-- Dependencies: **`@x12i/graphenix-executable-contracts`**, **`graphenix-plan-format`**, **`graphenix-trace-format`** @ **^1.1.0**; dev **`graphenix-plan-compiler`** @ **^1.1.0**.
-- Docs: [`documenations/node-execution-graphenix.md`](documenations/node-execution-graphenix.md), [`reports/graphenix-graph-engine-handoff.md`](../archived/ai-tasks/reports/graphenix-graph-engine-handoff.md); updated [`RUNTASK_REQUEST.md`](RUNTASK_REQUEST.md) and JSON schema.
-
-### Fixed
-
-- **`xynthesizedOutput`:** PRE synthesis restores `request.xynthesized` updates and **`response.xynthesizedPatch`** for graph-engine merge (`applySynthesisPreStepToRequest`).
-- **Polish POST:** wire **`structuredOutput`** (JSON mode) on polish LLM calls; node-execution POST handler merges **`parsed`** and **`intermediateSteps`**.
-
-## [8.8.0] - 2026-06-05
-
-### Fixed
-
-- **MAIN gateway wire shape:** `prepareMainSkillModelConfigForInvoke` applies ai-skills **`normalizeModelConfigForGatewayInvoke`** after profile resolution so MAIN invoke sends `{ provider, model }` tuple (never composite `openrouter/…` slug with `provider: unspecified`).
-- Consumes **`skillModel`** slot only for MAIN routing — no graph `modelConfig` merge at the execution boundary.
-
-### Changed
-
-- Bumped **`@exellix/ai-skills`** to **^6.5.0**, **`@exellix/xynthesis`** to **^4.6.0**, **`@x12i/ai-profiles`** to **^3.2.0** (`normalizeInvokeModel` contract).
-
-## [8.7.0] - 2026-06-05
-
-### Changed
-
-- Bumped **`@exellix/ai-skills`** to **^6.4.0** and **`@exellix/xynthesis`** to **^4.5.0** (gateway **≥ 10.1.0**, **`@x12i/ai-profiles` ^3.0.0**, **`@x12i/activix` ^8.6.0**).
-- Added direct **`@x12i/ai-profiles` ^3.0.0** dependency so orchestrator model resolution aligns with ai-skills / xynthesis (avoids hoisted **2.1.x** from legacy narrix transitive paths).
-- README stack alignment table and gateway 10.x notes updated for the new minimums.
-
-## [8.6.3] - 2026-06-05
-
-### Fixed (CR-13)
-
-- **`prepareMainSkillModelConfigForInvoke`:** Resolves `skillModel` / `llmCall.model` profile/choice aliases to concrete gateway wire ids once before ai-skills invoke (via `resolveModelReference`).
-- **`MAIN_SKILL_MODEL_RESOLUTION_FAILED`:** New `RunTaskModelResolutionError` code when MAIN alias resolution fails; Activix classification event `skill_model_resolution_failed`.
-- Bumped **`@exellix/ai-skills`** to **^6.3.0** for Optimixer `modelProfile` split + `SKILL_MODEL_RESOLUTION_FAILED`.
-
-## [8.5.0] - 2026-06-03
-
-### Changed (breaking)
-
-- **Execution strategies:** planner/optimizer invoke **`@exellix/xynthesis` ≥ 4.4.0** (`execution-plan`, `execution-evaluate-result`) with `modelConfig.preActionModel` / `postActionModel`. Removed direct `@x12i/funcx` `run()` from ai-tasks.
-- **`runPlanWebScopeQuestions`:** xynthesis `plan-web-scope-questions` only; removed `backend: "funcx"`.
-- **Catalog:** `runtimeKind: "xynthesis-action"`, `defaultSidekickAction` on planner/optimizer rows.
-
-### Added
-
-- `runPlannerViaXynthesis`, `runOptimizerViaXynthesis`, `setExecutionStrategyActionInvoker`, `setPlanWebScopeQuestionsActionInvoker`, `resolveSafeExecutionStrategyCatalogDefaultSidekickAction`.
-
-### Removed
-
-- `setFuncxExecutionStrategyInvoker`, `setResearchPlanQuestionsFuncxInvoker`, direct `@x12i/funcx` dependency.
-
-## [8.6.0] - 2026-06-04
-
-### Removed (breaking)
-
-- **Studio skill analysis** removed from the runtime package: **`analyzeRunTaskRequest`**, re-exports of **`buildSkillRequestAnalysisPacket`**, **`formatSkillRequestAnalysisMarkdown`**, **`FUNCX_ANALYZE_GATEWAY_INVOKE_REQUEST_ID`**, and related analysis types. Invoke preflight belongs in **studio / skills-manager** — see **[`documenations/studio-skill-invoke-preflight.md`](documenations/studio-skill-invoke-preflight.md)** (suggested design, not mandatory). Orchestrator static checks remain: **`validateRunTaskConfig`**, **`analyzeExpectedRunTaskInput`**.
-
-## [8.5.3] - 2026-06-04
-
-### Changed
-
-- Bumped **`@exellix/xynthesis`** to **^4.4.5** and **`@exellix/ai-skills`** to **^6.0.2** for Optimixer **max-token retry** on PRE/POST (xynthesis) and MAIN (ai-skills): one automatic re-invoke when output hits the token limit and Optimixer returns **`retryPrediction`** (no host **`maxTokens`** required).
-
-### Added
-
-- Unit test **`runPostStepLlmCall.maxTokenRetry.test.ts`** (mock Optimixer + synthesis invoker; no live LLM).
-
-## [8.4.0] - 2026-06-01 — `preActionModel`, `skillModel`, `postActionModel`. **`xynthesisModel` removed** (validation rejects with `NON_CANONICAL_MODEL_CONFIG`). PRE uses `preActionModel`; POST uses `postActionModel`.
-- **`@x12i/ai-profiles` ^2.0.0:** Required `catalogLane` on profile resolution; ai-tasks defaults `"text"` ( `"image"` for `vision` ). Registry source defaults to `auto`.
-- **`@exellix/xynthesis` ^4.2.1:** Minimum for ai-profiles v2 alignment.
-
-### Added
-
-- **`defaultAiTasksResolveAiProfileOptions`**, slot helpers `resolvePreActionModel` / `resolvePostActionModel`, builder `withPreActionModel` / `withPostActionModel`.
-
-### Removed
-
-- Exports **`resolveXynthesisModelForPre`**, **`getXynthesisModelFromSlots`**, builder **`withXynthesisModel`**.
-
-## [Unreleased — prior notes]
-
-### Changed
-
-- **`@x12i/ai-profiles` ≥1.8.0:** Direct dependency bumped from **^1.7.2**; **`npm` override** pins transitive `@x12i/ai-profiles` to this package's direct version (required fix in **1.8.0** for profile alias resolution). Live coverage: `npm run test:live` → `test/e2e/ai-profiles-live.test.ts` (registry `auto` + real xynthesis alias call).
-- **`@exellix/xynthesis` ^4.1.8:** Fixes OpenRouter **400 invalid model ID** on PRE/POST hops (wire id was `openrouter/<slug>` instead of bare OR API slug). ai-tasks invocation routing aligned (skill gateway shape vs xynthesis Funcx slug). Requires **≥4.1.8** — do not use **4.1.7** for alias-based xynthesis calls with OpenRouter.
-- **`@x12i/logxer` ^4.4.2:** Package diagnostics use **`createLogxer`** with diagnostic catalog (`.metadata/log-diagnostics.json`), **`warnCode`** / **`DebugLogAbstract`**, **`fieldEvidence`**, **`runWithLogContext`** during **`runTask()`**, and in-process **`getAiTasksJobLogs`** for debug UIs. Override pins transitive logxer to **4.4.2** (no nested **4.4.0** from `@x12i/optimixer`).
-
-## [8.1.1] - 2026-05-28
-
-### Changed
-
-- **`npm` overrides:** Pin transitive **`@x12i/activix`**, **`@x12i/catalox`**, and **`@x12i/funcx`** to this package's direct versions so graph-engine and other consumers resolve a single copy (no nested **funcx 3.9.x** / **activix 7.x** / **catalox 4.x** from `@exellix/narrix-web-scoper` or `@x12i/funcx`).
-
-## [8.1.0] - 2026-05-28
-
-### Added
-
-- **Model profile aliases (`@x12i/ai-profiles`):** Resolve ai-profiles profile/shortcut names and legacy tiers (`weak`, `strong`, `default`) to concrete provider model ids for **ai-skills MAIN** (`skillModel` / `llmCall.model`). Keep **ai-profiles aliases** on **xynthesis** paths (`modelConfig.xynthesisModel`, POST/PRE/scoping `llmCall.model`). Exports: `resolveModelReference`, `resolveModelReferenceForXynthesis`, `normalizeXynthesisModelAlias`, `resolveRunTaskModelReferences`, `isResolvableModelAlias`, `formatResolvedProfileModelId`.
-- **Dual-root `RunTaskRequest.inputs`:** Optional top-level plural payload bucket (distinct from `executionMemory.inputs`); forwarded on MAIN like `input`.
-
-### Changed
-
-- **Smart-input validation context:** `inputs.*` resolves against top-level **`request.inputs`** only (not `executionMemory.inputs`). Use **`executionMemory.inputs.*`** for execution-memory plural paths.
-- **PRE `memoryPaths`:** Dual-root fallback — `executionMemory.input.<rest>` falls back to `executionMemory.inputs.<firstKey>.<rest>` when singular path is missing.
-- **`@exellix/xynthesis` ^4.0.6:** Xynthesis LLM calls use alias-only model resolution (`resolveModelReferenceForXynthesis`); default post-step model is **`balanced`** when omitted. **`skillModel`** resolution for **`@exellix/ai-skills` ^5.8.6** unchanged.
-- **Validation:** `modelConfig.xynthesisModel` rejects concrete provider ids; accepts ai-profiles aliases (replaces 8.0.0 `MODEL_CONFIG_NOT_RESOLVED` for legacy tiers on both slots).
-
-### Documentation
-
-- README — [Model profile aliases](#model-profile-aliases-x12iai-profiles); [`.docs/ai-tasks-model-profile-aliases-7x.md`](../archived/ai-tasks/.docs/ai-tasks-model-profile-aliases-7x.md).
-
-## [8.0.9] - 2026-05-28
-
-### Changed
-
-- **Dependencies (graph-engine alignment):** **`@x12i/funcx` ^4.2.0** (no 3.9.x path), **`@x12i/activix` ^8.0.5**, **`@x12i/catalox` ^5.1.1**; **`@x12i/funcx/functions`** re-exports (`FUNCX_EXECUTION_PLAN_FUNCTION_ID`, etc.).
-
-## [8.0.8] - 2026-05-27
-
-### Added
-
-- **`getModelCapabilities`** re-export from `@exellix/xynthesis` (replaces removed **`MODEL_CAPABILITIES`** constant).
-
-### Fixed
-
-- **Exports:** Removed broken **`MODEL_CAPABILITIES`** re-export from `@exellix/xynthesis` 4.x (symbol removed upstream). Use **`getModelCapabilities(modelId)`** from `@exellix/ai-tasks` or `@exellix/xynthesis` instead.
-
-### Deprecated
-
-- **`8.0.7`** — do not use; re-exported **`MODEL_CAPABILITIES`** against **`@exellix/xynthesis` ≥ 4.x** where the symbol no longer exists. Upgrade to **8.0.8+**.
-
-## [7.7.1] - 2026-05-25
-
-### Added
-
-- **`RunTaskRequest.taskConfiguration`:** Graph-engine may forward the node `taskConfiguration` blob; **`compileTaskConfigurationOnRunTaskRequest`** (also at **`runTask()`** entry) maps **`aiTaskStrategies.pre: "synthesis"`** and **`aiTaskProfile.inputSynthesis.enabled`** into PRE **`synthesized-context`** + **`includeContextInPrompt: true`**. Exports **`TaskConfigurationCompileError`**, **`AI_TASK_PRE_ENABLE_SYNTHESIS`**.
-
-### Documentation
-
-- **`RUNTASK_REQUEST.md`**, **README**, [`.docs/flow-io/xynthesis-pre.md`](../archived/ai-tasks/.docs/flow-io/xynthesis-pre.md), graph-engine task [`reports/graph-engine-task-pre-synthesis-compile.md`](../archived/ai-tasks/reports/graph-engine-task-pre-synthesis-compile.md).
-- Fixtures: [`test/fixtures/run-task/graph-engine-pre-synthesis.json`](test/fixtures/run-task/graph-engine-pre-synthesis.json), [`test/fixtures/graphs/graph-qcrbz6t-pre-synthesis.json`](test/fixtures/graphs/graph-qcrbz6t-pre-synthesis.json).
-
-## [7.6.4] - 2026-05-18
-
-### Changed
-
-- **`@exellix/ai-skills` ≥ 5.5.0:** Dependency **5.5.2**; `pickRunSkillRequestFields` forwards top-level **`executionMemory`** on `runSkill` (native `buildWorkingMemory` support). Removed the 7.6.3 `jobMemory.executionMemory` nest bridge.
-
-### Dependencies
-
-- **`@exellix/ai-skills`:** `5.5.2` (was `5.4.0`).
-
-## [7.6.3] - 2026-05-18
-
-### Fixed
-
-- **Gateway `executionMemory` paths (interim):** `DirectExecutionStrategy` nested `request.executionMemory` on `jobMemory.executionMemory` before `runSkill` for ai-skills **&lt; 5.5.0**. Superseded in **7.6.4** by native top-level forward. Strips `RunTaskRequest`-only fields from the `runSkill` payload.
-
-### Documentation
-
-- **Change request:** [`reports/change-request-tasknode-variable-scopes-7.6.2.md`](../archived/ai-tasks/reports/change-request-tasknode-variable-scopes-7.6.2.md) — upstream work for `@exellix/ai-skills`, `@x12i/rendrix`, and optional gateway path aliases.
-
-## [7.6.2] - 2026-05-18
-
-### Fixed
-
-- **`smartInput.strict`:** Accepted on `RunTaskRequest.smartInput` and forwarded through `normalizeSmartInputConfig` to the gateway path.
-
-### Changed
-
-- **MAIN template variables:** `runTask` forwards **`variables`** (job/graph bucket) **as-is** via `passthroughJobTemplateVariables` — no longer folds `jobContext`, `input`, `xynthesized`, or `smartInput` into `variables`. Node scope remains on **`executionMemory.taskVariables`**.
-
-### Documentation
-
-- **`RUNTASK_REQUEST.md`:** TaskNode authoring map (`taskConfiguration`, `inputsConfig`, `taskVariable`, two-bucket `jobVariables` / `taskVariables`), dual `outputValidation`, `inputSynthesis` compile table, forbidden metadata.
-- **README**, web-scoping doc, Narrix JSDoc: authoring path **`taskConfiguration.narrix`** (not `metadata.narrix`).
-
-### Schema
-
-- **`run-task-request.json`:** `executionStrategyCatalogItems`, `smartInputRenderOptions`, Rendrix path objects, `smartInput.strict`, variable-bucket descriptions.
-
-## [7.6.0] - 2026-05-18
-
-### Added
-
-- **`xynthesized.execution` scope:** `XynthesizedMemory.execution`, `XynthesizedDestinationScope` includes **`"execution"`**, PRE **`xynthesizedOutput.destination: "execution"`** writes and returns **`xynthesizedPatch.execution`** (same replace/merge semantics as job/task; no redirect to other buckets).
-- **Smart-input:** `xynthesized.*` paths must use scope **`job`**, **`task`**, or **`execution`**; **`resolveXynthesizedSmartInputPath`** helper for in-package resolution against request **`xynthesized`**.
-- **`TaskRequestBuilder.withXynthesizedExecution(key, value)`**.
-
-### Documentation
-
-- **`RUNTASK_REQUEST.md`**, **`documenations/synthesized-context-guide.md`**, and **`documenations/schemas/v1/run-task-request.json`** updated for the execution bucket.
-
-## [7.2.0] - 2026-05-06
-
-### Changed
-
-- **Breaking (exports):** Replaced **`AI_TASKS_FUNCX_*`** constants with **`FUNCX_EXECUTION_PLAN_FUNCTION_ID`** (`execution/plan`), **`FUNCX_EXECUTION_EVALUATE_RESULT_FUNCTION_ID`** (`execution/evaluate-result`), **`FUNCX_RESEARCH_PLAN_QUESTIONS_FUNCTION_ID`** (`research/plan-questions`).
-- **Execution strategies:** Planner and optimizer call **`run(functionId, payload)`** with the **generic FuncX envelope** only (`goal`, `input`, `context`, `result?`, `args`, `attribution`). Responses are adapted via **`adaptExecutionPlanResult`** / **`adaptEvaluateResultResponse`** ([`genericExecutionFuncxEnvelope.ts`](src/execution-strategies/genericExecutionFuncxEnvelope.ts)).
-- **`runPlanWebScopeQuestions`:** Optional **`backend: "funcx"`** for **`research/plan-questions`**; **`setResearchPlanQuestionsFuncxInvoker`** for tests.
-- **`ExecutionType`:** Re-exported from **`@exellix/ai-tasks`** via local [`executionType.ts`](src/types/executionType.ts) (`DIRECT`) because **`@exellix/ai-skills`** no longer exports this symbol on supported versions.
-- **`@x12i/funcx` ≥ 3.8.2:** **`unwrapFuncxRunValue`** delegates to **`getRunJsonResult`** (use **3.8.2+** for current normalization behavior). Re-exports: **`getRunJsonResult`**, **`isHttpRunSuccessBody`**, **`genericExecutionEnvelopeJsonSchema`**, **`buildAskAttribution`**, type **`GenericExecutionEnvelope`**.
-
-## [7.1.2] - 2026-05-06
-
-### Added
-
-- **`narrix.webScopeQuestions`:** optional explicit web-scope questions on `RunTaskRequest.narrix`; when `webScopeTemplates` is unset, the NARRIX preprocessor runs `@exellix/narrix-web-scoper` **`scopeQuestionPack`** and stores a normalized `WebScoperResult` in `executionMemory.webContext`.
-- **`runPlanWebScopeQuestions`** (and **`parseAndValidatePlanWebScopeQuestionsOutput`**, **`buildPlanWebScopeQuestionsUserPrompt`**) for AI-driven web-scope planning; FuncX path uses **`research/plan-questions`** (see **Unreleased** / **`FUNCX_RESEARCH_PLAN_QUESTIONS_FUNCTION_ID`**). FR tracker: **`documenations/funcx-scoping-integration-gaps.md`** §6.
-
-### Documentation
-
-- **`documenations/web-scoping-in-ai-tasks.md`**, **README**, **RUNTASK_REQUEST.md:** documented `webScopeQuestions` shape and precedence vs templates.
-
-## [7.1.1] - 2026-05-06
-
-### Added
-
-- **Package root type exports:** `XynthesizedMemory`, `SmartInputConfig`, `XynthesizedDestinationScope`, `XynthesizedOutputConfig`, `ExecutionStrategyInvocation`, `ExecutionStrategyPhase`, and `ExecutionStrategyWrapperKey` are re-exported from `@exellix/ai-tasks` alongside `RunTaskRequest`, so hosts can depend on stable names instead of `NonNullable<RunTaskRequest['xynthesized']>` or `RunTaskRequest['executionStrategies'][number]`.
-
-### Documentation
-
-- **CHANGELOG discipline:** Documented that `package.json` is authoritative for the shipped semver and pointed cross-version migration readers at `BREAKING-CHANGES.md` / `RUNTASK_REQUEST.md` when intermediate entries are missing.
-
-_Note: Releases **after 6.0.1** through **before 7.1.1** were not listed here chronologically; consult git history and `BREAKING-CHANGES.md` for that span._
-
-## [6.0.1] - 2026-05-05
-
-### Fixed
-
-- **Synthesis E2E gate:** `test/synthesis/e2e-synthesis-real.test.ts` no longer requires `LLM_MODEL_STRONG` / `LLM_MODEL_NORMAL`; it runs when **`RUN_SYNTHESIS_E2E=1`** and a recognized provider key is set (`OPEN_ROUTER_KEY`, `GROK_API_KEY`, etc.). Warns when a model env var is missing.
-
-### Documentation
-
-- **Xynthesis upstream work:** [`documenations/xynthesis-upstream-fixes-checklist.md`](documenations/xynthesis-upstream-fixes-checklist.md) (Activix `diagnostics.owner`, `jobId`/`taskId` hardening in `runSidekickGatewayCall`, tests to add in xynthesis).
-
-## [6.0.0] - 2026-05-04
-
-### Breaking
-
-- **`runTask` / `RunTaskRequest`:** `agentId`, `jobTypeId`, and `taskTypeId` are **required** again (aligned with `@exellix/ai-skills` `RunSkillRequest`). `runTask()` calls **`assertRequiredRunSkillCorrelation`** and throws if any value is missing or whitespace-only.
-- **`TaskRequestBuilder.build()`** throws unless those three fields were set (use **`withAiSkillsCorrelation`** or **`withAgentId` / `withJobTypeId` / `withTaskTypeId`**).
-- **`TaskConvenienceMethods.executeDirect`** now requires an **`options`** object including **`agentId`**, **`jobTypeId`**, and **`taskTypeId`** (third argument is no longer optional).
-- **`TaskConvenienceMethods.executeWithType`** now requires the **`options`** argument (same correlation fields as `RunTaskRequest`).
-
-### Added
-
-- Exported **`assertRequiredRunSkillCorrelation`** from the package root for hosts that validate payloads before calling `runTask`.
-
-## [5.7.1] - 2026-05-04
-
-### Fixed
-
-- **Types:** `RunTaskRequest` now **omits and re-declares** `agentId`, `jobTypeId`, and `taskTypeId` as **optional**, matching `runTask()` usage and latest `@exellix/ai-skills` `RunSkillRequest` where those fields are required. Callers can still pass them when the gateway requires them.
-
-## [5.7.0] - 2026-05-04
-
-### Added
-
-- **RunTask request contract documentation:** [`RUNTASK_REQUEST.md`](RUNTASK_REQUEST.md) — normative field list, template merge precedence, `executionMemory` vs `jobMemory.execution`, `executionPipeline` semantics, xynthesis visibility notes, and examples.
-- **JSON Schema fragment:** [`documenations/schemas/v1/run-task-request.json`](documenations/schemas/v1/run-task-request.json) and OpenAPI component `RunTaskRequestSubset` in [`documenations/schemas/openapi-3.1-components.yaml`](documenations/schemas/openapi-3.1-components.yaml).
-- **Request shape helpers** (exported from package root): `normalizeRunTaskRequest`, `collectRunTaskRequestWarnings`, `collectExecutionPipelineWarnings`, `isRunTaskRequestWarningsEnabled`.
-- **Env-gated diagnostics:** `AI_TASKS_REQUEST_WARNINGS=1` (also `true` / `yes`) logs structured JSON for deprecated duplicate fields, ignored pipeline steps, and observability hints at the start of each `runTask` call.
-
-### Changed
-
-- **Types:** `RunTaskRequest` documents deprecated mirrors: optional root `question`, `inputs`, and `jobInput` (typed for discoverability; removal deferred to a future major after hosts migrate).
-
-### Notes
-
-- **Additive / non-breaking:** Runtime merge behavior is unchanged; warnings are opt-in via env.
-- **Cross-package:** Minimum orchestrator (e.g. graph-engine) version is **not declared** in `@exellix/ai-tasks` — follow the consuming app’s pinned `@exellix/ai-tasks` range; see `RUNTASK_REQUEST.md` § Orchestrator integration.
+# Changelog
+
+All notable changes to `@exellix/ai-tasks` are documented here.
+
+The published **npm version** always matches **`version`** in [`package.json`](package.json). If this file skips a semver, treat [`BREAKING-CHANGES.md`](BREAKING-CHANGES.md) and [`RUNTASK_REQUEST.md`](RUNTASK_REQUEST.md) as the source of truth for cross-version upgrades.
+
+## [10.1.1] - 2026-06-24
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.11.0** (2.11 train).
+- `traceLifecycle.ts` adapter prefers GTF-FR-001 exports from `@x12i/graphenix-trace-format` when present; falls back to local impl until upstream exports `markUnitStarted` / `recordModelInvocation*`.
+
+## [10.1.0] - 2026-06-24
+
+### Added
+
+- **AT-CR-001:** Dual-write v2 `executionTrace` indexes during node-plan execution — `unitExecutions`, `modelInvocations`, and `nodeExecutions` stay in sync with `unit.*` / `model.invocation.*` events when `RunTaskRequest.executionTrace` is present.
+- Lifecycle helpers in [`src/node-execution/traceLifecycle.ts`](src/node-execution/traceLifecycle.ts) (GTF-FR-001-compatible API; delegates event append to `@x12i/graphenix-trace-format`).
+- Observation sink for PRE / MAIN / POST plan units → gateway `usage` / `cost` on `modelInvocations` when available.
+- Tests: [`test/trace/executionTraceDualWrite.test.ts`](test/trace/executionTraceDualWrite.test.ts) (T1–T6).
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.10.0** (bundled with AT-CR-001).
+- Requires `@x12i/graphenix-trace-format` **^2.10.0** (index dual-write; upstream GTF-FR-001 export planned).
+
+## [10.0.13] - 2026-06-20
+
+### Changed
+
+- **`@exellix/ai-skills`** **^6.12.5** (graphenix 2.7.3 + catalox 5.9.8 train).
+- **`@x12i/catalox`** / **`@x12i/catalox-contracts`** **^5.9.8**.
+
+## [10.0.12] - 2026-06-20
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.7.3** (full format suite train).
+
+## [10.0.11] - 2026-06-20
+
+### Fixed
+
+- **`@exellix/fact-guard-*` dependencies:** publish semver **`^1.0.1`** in `dependencies` (10.0.10 accidentally shipped **`file:../fact-guard/...`** in `dependencies`, which breaks installs outside this monorepo).
+- **`prepublishOnly`:** fail fast if `file:` / `link:` / `workspace:` specs remain in `package.json` at publish time.
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.7.2** (compiler emits `webQueryTemplate` on `webScope` units; legacy `webScoping` rejected on plans).
+- **`@exellix/ai-skills`** **^6.12.4** (graphenix 2.7.2 train).
+- **Web scope validation:** enabled `webScope` PRE units require non-empty **`unitParams.webQueryTemplate`** (or `webQueryTemplates[]`); rejects legacy **`webScoping`** / NARRIX question fallbacks on unit params (`LEGACY_WEB_SCOPE_UNIT_PARAMS`, `WEB_QUERY_TEMPLATE_REQUIRED`, `WEB_QUERY_TEMPLATE_UNRESOLVED`).
+
+## [10.0.10] - 2026-06-20
+
+### Changed
+
+- **Web scoping:** migrate from **`@exellix/narrix-web-scoper`** to published **`@x12i/web-scoper@^1.1.1`**. New adapter module **`src/web-scope/`** compiles graph/task input to **`{ question, record?, options? }`** or **`searchMany`** packs; **`executionMemory.webContext`** uses **`{ ok, context?, error? }`** (replaces legacy **`available`**).
+- **Standalone PRE `webScope` unit:** playground and Graphenix plans invoke the same adapter via **`unitParams.enableWebScope`** + **`unitParams.options`**.
+- **Docs:** README, RUNTASK_REQUEST, and **`documenations/web-scoping-in-ai-tasks.md`** updated for the new contract.
+
+### Removed
+
+- Dependency on **`@exellix/narrix-web-scoper`**. Legacy **`narrix.webScoping`** / template fields are still mapped at invoke time where supported.
+
+## [10.0.9] - 2026-06-18
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.7.0** (fact-guard schema/validation/compiler train; optional `unitParams.factGuard` fully validated upstream).
+- **`@exellix/ai-skills`** **^6.12.3** (graphenix 2.7.0 train).
+
+## [10.0.8] - 2026-06-18
+
+### Changed
+
+- **`@exellix/fact-guard-*`** **^1.0.0** — re-scoped from `@x12i` to **`@exellix`** (npm publish train).
+- **`@exellix/xynthesis`** **^4.8.4**.
+
+## [10.0.7] - 2026-06-18
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.6.0** (optional `unitParams.factGuard` on sub-action units; validates against graphenix `FactGuardStepConfig`).
+- **`@exellix/ai-skills`** **^6.12.2** (graphenix 2.6.0 train).
+- **`@exellix/fact-guard-*`** **^1.0.0** (npm semver; was monorepo `file:` link).
+
+### Added
+
+- Node-plan validation tests for graphenix 2.6.0 fact-guard golden fixtures (`node-synthesis-fact-guard-*`, `node-polish-fact-guard-enabled`, `node-audit-fact-guard-nested`).
+
+## [10.0.6] - 2026-06-17
+
+### Added
+
+- **Fact-guard sub-actions (optional):** optional atom integrity checks on PRE synthesis, POST audit/polish/scoping, and execution-strategy planner/optimizer hops via **`@exellix/fact-guard-*` ^1.0.0**. Disabled by default (`factGuard.enabled: false` or absent = unchanged behavior). Kill-switch: **`FACT_GUARD_DISABLED=1`**. MAIN **`runSkill`** path unchanged.
+- **`FactGuardStepConfig`** on `SynthesisConfig`, `AuditConfig`, `PolishConfig`, `ExecutionStrategyInvocation`, and `aiScopingOptions`.
+- **`npm run test:fact-guard`** — integration tests (mocked LLM).
+
+### Changed
+
+- **`@exellix/xynthesis`** **^4.8.3** (fact-guard passthrough on `executeXynthesisAction`).
+
+## [10.0.5] - 2026-06-17
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.5.0** (CRS-FRS-003 train; jobMemory.context authoring preserved upstream).
+- **`@exellix/ai-skills`** **^6.12.1** — consumes graphenix 2.5.0 train.
+
+## [10.0.4] - 2026-06-15
+
+### Changed
+
+- **Graphenix alignment:** all `@x12i/graphenix-*` packages **^2.4.0** (suite train bump from 2.3.0).
+- **`@exellix/ai-skills`** **^6.12.0** — consumes graphenix 2.4.0 train.
+
+## [10.0.2] - 2026-06-14
+
+### Added
+
+- **Graphenix 2.1.0 alignment:** `@x12i/graphenix-plan-format` and `@x12i/graphenix-executable-contracts` **^2.1.0** — rejects plans with `unitParams.synthesisInputStrategy` (`LEGACY_SYNTHESIS_INPUT_STRATEGY_ON_PLAN`).
+- **`resolveSynthesisContextStrategyKey`** — PRE synthesis template key from skill catalog (`@exellix/ai-skills` **^6.11.0**); path-driven source material via `invokeContract.reads` / `memoryPaths`.
+- Validation test for legacy synthesis field on compiled plans.
+
+### Changed
+
+- **`synthesisConfigFromUnit`** strips `unitParams.synthesisInputStrategy` from plan wire; no longer defaults plan enum to `policy`.
+- Wire fixture `graph-engine-pre-synthesis.json` — path-only synthesis unit (no `synthesisInputStrategy`).
+- **Docs:** README / RUNTASK_REQUEST — skill-catalog-only template key; graphenix **2.1.0** deps.
+
+## [10.0.1] - 2026-06-13
+
+### Added
+
+- **Graphenix 2.0.3 alignment:** `@x12i/graphenix-executable-contracts` and `@x12i/graphenix-plan-format` **^2.0.3** — orchestration strategy registry (`plan`, `split`, `merge`, `pick-best`, `webScope`), path-driven synthesis units (`invokeContract.reads`).
+- Golden tests for graphenix plan fixtures: `node-synthesis-execution-memory-only`, `node-split-merge`, `node-web-scope-standalone`.
+- **Synthesis path bridge:** `synthesisConfigFromUnit` copies `invokeContract.reads` → `memoryPaths` when omitted; defaults `synthesisInputStrategy` to `policy` for path-driven plans.
+
+### Changed
+
+- Graph authoring fixtures: removed deprecated `preInputStrategy` from `aiTaskStrategies`.
+- Wire fixture `graph-engine-pre-synthesis.json` aligned with graphenix 2.0.3 compile output (`reads`: `input`, `executionMemory`).
+- **Docs:** README Graphenix 10.x section; downstream [graph-model-change-2026-06-13.md](../temp/downstream/graph-model-change-2026-06-13.md).
+
+## [10.0.0] - 2026-06-13
+
+### Breaking
+
+- **Removed** `taskConfiguration` and `executionPipeline` from `RunTaskRequest` — use frozen `nodePlan` only.
+- **Removed** compile pipeline (`src/compile/*`, `taskConfigurationCompileError`).
+
+### Added
+
+- **PRE synthesis wiring:** `resolveRawTemplate(role: synthesisContext, strategyKey)` via `@exellix/ai-skills` **^6.10.0**.
+- **PRE/POST orchestration runtime:** `externalPreUtility` (`plan`, `split`, `webScope`), `externalPostUtility` (`merge`, `pick-best`), multi-shard MAIN + FuncX fan-in.
+- **Web scope:** Narrix preprocessor `enableWebScope` + standalone `webScope` unit.
+- **Strategy catalogs:** synthesis-input, pre/post orchestration (publish via `npm run publish:task-strategies`).
+- **nodePlan audit parity:** real `reRunMain` + synthesized `contextMarkdown`.
+
+### Changed
+
+- **`@exellix/ai-skills`** **^6.10.0** (support templates + single `professional-answer` Catalox seed).
+
+## [9.1.1] - 2026-06-13
+
+### Changed
+
+- **Run-task handoff closure:** Stale docs updated — node template scope is `nodePlan.invokeContract.taskVariable`, not `executionMemory.taskVariables`.
+- **`@exellix/ai-skills`** **^6.9.1** (WM source-of-truth alignment).
+
+### Added
+
+- Golden parity tests for all **7** node plan fixtures (`node-post-audit`, `node-input-synthesis-profile` added to `golden-fixtures.test.ts`).
+
+## [9.1.0] - 2026-06-09
+
+### Changed
+
+- **Graphenix 2.1.0 alignment:** Accept parent plan format `"executable-plan"` (v2) and trace format `"execution-trace"` (v2).
+- **Snapshot isolation:** `overlayUnitModelOnRequest` ignores concrete wire ids in `resolvedInvocationSnapshot.modelId`, preferring authored `modelSelection` (CR-1.3).
+
+## [9.0.5] - 2026-06-08
+
+### Changed
+
+- **`@x12i/funcx`** **^4.9.5** (direct dependency + npm override) — aligns PRE/POST xynthesis with funcx catalog gate via **`@x12i/ai-profiles`** (requires funcx ≥ 4.9.5; fixes subnet-style PRE `Unknown model "google/gemini-2.5-flash-lite"`).
+- **`@x12i/ai-profiles`** **^3.4.0**, **`@exellix/xynthesis`** **^4.6.5** (publish **4.6.6** for funcx bump).
+
+## [9.0.1] - 2026-06-07
+
+### Changed
+
+- Graphenix dependency alignment with latest CP release train: **`graphenix-plan-format`** and **`graphenix-trace-format`** @ **^1.1.1**; dev **`graphenix-authoring-format`** @ **^1.2.1**, **`graphenix-plan-compiler`** @ **^1.1.1**. **`graphenix-executable-contracts`** remains **^1.1.0**.
+- **`npm` overrides:** Pin transitive **`@x12i/funcx`** to **^4.6.0** (aligns with **`@exellix/xynthesis` ≥ 4.6.2** and **`@x12i/logxer` ^4.6.0**).
+
+## [9.0.0] - 2026-06-07
+
+### Changed (breaking)
+
+- **Graphenix node execution v2.1:** `runTask()` requires **`nodePlan`** (`NodeExecutionPlan`); production path is **`executeNodeFromPlan`** (unit loop + per-unit model selection). Legacy **`executionPipeline`** / **`taskConfiguration`** compile is **off the hot path** (still exported for migration tooling).
+- **Forbidden wire fields** when `nodePlan` is set: root **`modelConfig`**, **`executionPipeline`**, **`taskConfiguration`** — model slots come from `nodePlan.executionUnits[].modelSelection`.
+- **`executionTrace`** (optional): graphenix v2 trace append (`unit.*`, `model.invocation.*`).
+
+### Added
+
+- **`src/node-execution/`** — unit dispatch, Narrix units, PRE/MAIN/POST handlers, trace bridge.
+- Dependencies: **`@x12i/graphenix-executable-contracts`**, **`graphenix-plan-format`**, **`graphenix-trace-format`** @ **^1.1.0**; dev **`graphenix-plan-compiler`** @ **^1.1.0**.
+- Docs: [`documenations/node-execution-graphenix.md`](documenations/node-execution-graphenix.md), [`reports/graphenix-graph-engine-handoff.md`](../archived/ai-tasks/reports/graphenix-graph-engine-handoff.md); updated [`RUNTASK_REQUEST.md`](RUNTASK_REQUEST.md) and JSON schema.
+
+### Fixed
+
+- **`xynthesizedOutput`:** PRE synthesis restores `request.xynthesized` updates and **`response.xynthesizedPatch`** for graph-engine merge (`applySynthesisPreStepToRequest`).
+- **Polish POST:** wire **`structuredOutput`** (JSON mode) on polish LLM calls; node-execution POST handler merges **`parsed`** and **`intermediateSteps`**.
+
+## [8.8.0] - 2026-06-05
+
+### Fixed
+
+- **MAIN gateway wire shape:** `prepareMainSkillModelConfigForInvoke` applies ai-skills **`normalizeModelConfigForGatewayInvoke`** after profile resolution so MAIN invoke sends `{ provider, model }` tuple (never composite `openrouter/…` slug with `provider: unspecified`).
+- Consumes **`skillModel`** slot only for MAIN routing — no graph `modelConfig` merge at the execution boundary.
+
+### Changed
+
+- Bumped **`@exellix/ai-skills`** to **^6.5.0**, **`@exellix/xynthesis`** to **^4.6.0**, **`@x12i/ai-profiles`** to **^3.2.0** (`normalizeInvokeModel` contract).
+
+## [8.7.0] - 2026-06-05
+
+### Changed
+
+- Bumped **`@exellix/ai-skills`** to **^6.4.0** and **`@exellix/xynthesis`** to **^4.5.0** (gateway **≥ 10.1.0**, **`@x12i/ai-profiles` ^3.0.0**, **`@x12i/activix` ^8.6.0**).
+- Added direct **`@x12i/ai-profiles` ^3.0.0** dependency so orchestrator model resolution aligns with ai-skills / xynthesis (avoids hoisted **2.1.x** from legacy narrix transitive paths).
+- README stack alignment table and gateway 10.x notes updated for the new minimums.
+
+## [8.6.3] - 2026-06-05
+
+### Fixed (CR-13)
+
+- **`prepareMainSkillModelConfigForInvoke`:** Resolves `skillModel` / `llmCall.model` profile/choice aliases to concrete gateway wire ids once before ai-skills invoke (via `resolveModelReference`).
+- **`MAIN_SKILL_MODEL_RESOLUTION_FAILED`:** New `RunTaskModelResolutionError` code when MAIN alias resolution fails; Activix classification event `skill_model_resolution_failed`.
+- Bumped **`@exellix/ai-skills`** to **^6.3.0** for Optimixer `modelProfile` split + `SKILL_MODEL_RESOLUTION_FAILED`.
+
+## [8.5.0] - 2026-06-03
+
+### Changed (breaking)
+
+- **Execution strategies:** planner/optimizer invoke **`@exellix/xynthesis` ≥ 4.4.0** (`execution-plan`, `execution-evaluate-result`) with `modelConfig.preActionModel` / `postActionModel`. Removed direct `@x12i/funcx` `run()` from ai-tasks.
+- **`runPlanWebScopeQuestions`:** xynthesis `plan-web-scope-questions` only; removed `backend: "funcx"`.
+- **Catalog:** `runtimeKind: "xynthesis-action"`, `defaultSidekickAction` on planner/optimizer rows.
+
+### Added
+
+- `runPlannerViaXynthesis`, `runOptimizerViaXynthesis`, `setExecutionStrategyActionInvoker`, `setPlanWebScopeQuestionsActionInvoker`, `resolveSafeExecutionStrategyCatalogDefaultSidekickAction`.
+
+### Removed
+
+- `setFuncxExecutionStrategyInvoker`, `setResearchPlanQuestionsFuncxInvoker`, direct `@x12i/funcx` dependency.
+
+## [8.6.0] - 2026-06-04
+
+### Removed (breaking)
+
+- **Studio skill analysis** removed from the runtime package: **`analyzeRunTaskRequest`**, re-exports of **`buildSkillRequestAnalysisPacket`**, **`formatSkillRequestAnalysisMarkdown`**, **`FUNCX_ANALYZE_GATEWAY_INVOKE_REQUEST_ID`**, and related analysis types. Invoke preflight belongs in **studio / skills-manager** — see **[`documenations/studio-skill-invoke-preflight.md`](documenations/studio-skill-invoke-preflight.md)** (suggested design, not mandatory). Orchestrator static checks remain: **`validateRunTaskConfig`**, **`analyzeExpectedRunTaskInput`**.
+
+## [8.5.3] - 2026-06-04
+
+### Changed
+
+- Bumped **`@exellix/xynthesis`** to **^4.4.5** and **`@exellix/ai-skills`** to **^6.0.2** for Optimixer **max-token retry** on PRE/POST (xynthesis) and MAIN (ai-skills): one automatic re-invoke when output hits the token limit and Optimixer returns **`retryPrediction`** (no host **`maxTokens`** required).
+
+### Added
+
+- Unit test **`runPostStepLlmCall.maxTokenRetry.test.ts`** (mock Optimixer + synthesis invoker; no live LLM).
+
+## [8.4.0] - 2026-06-01 — `preActionModel`, `skillModel`, `postActionModel`. **`xynthesisModel` removed** (validation rejects with `NON_CANONICAL_MODEL_CONFIG`). PRE uses `preActionModel`; POST uses `postActionModel`.
+- **`@x12i/ai-profiles` ^2.0.0:** Required `catalogLane` on profile resolution; ai-tasks defaults `"text"` ( `"image"` for `vision` ). Registry source defaults to `auto`.
+- **`@exellix/xynthesis` ^4.2.1:** Minimum for ai-profiles v2 alignment.
+
+### Added
+
+- **`defaultAiTasksResolveAiProfileOptions`**, slot helpers `resolvePreActionModel` / `resolvePostActionModel`, builder `withPreActionModel` / `withPostActionModel`.
+
+### Removed
+
+- Exports **`resolveXynthesisModelForPre`**, **`getXynthesisModelFromSlots`**, builder **`withXynthesisModel`**.
+
+## [Unreleased — prior notes]
+
+### Changed
+
+- **`@x12i/ai-profiles` ≥1.8.0:** Direct dependency bumped from **^1.7.2**; **`npm` override** pins transitive `@x12i/ai-profiles` to this package's direct version (required fix in **1.8.0** for profile alias resolution). Live coverage: `npm run test:live` → `test/e2e/ai-profiles-live.test.ts` (registry `auto` + real xynthesis alias call).
+- **`@exellix/xynthesis` ^4.1.8:** Fixes OpenRouter **400 invalid model ID** on PRE/POST hops (wire id was `openrouter/<slug>` instead of bare OR API slug). ai-tasks invocation routing aligned (skill gateway shape vs xynthesis Funcx slug). Requires **≥4.1.8** — do not use **4.1.7** for alias-based xynthesis calls with OpenRouter.
+- **`@x12i/logxer` ^4.4.2:** Package diagnostics use **`createLogxer`** with diagnostic catalog (`.metadata/log-diagnostics.json`), **`warnCode`** / **`DebugLogAbstract`**, **`fieldEvidence`**, **`runWithLogContext`** during **`runTask()`**, and in-process **`getAiTasksJobLogs`** for debug UIs. Override pins transitive logxer to **4.4.2** (no nested **4.4.0** from `@x12i/optimixer`).
+
+## [8.1.1] - 2026-05-28
+
+### Changed
+
+- **`npm` overrides:** Pin transitive **`@x12i/activix`**, **`@x12i/catalox`**, and **`@x12i/funcx`** to this package's direct versions so graph-engine and other consumers resolve a single copy (no nested **funcx 3.9.x** / **activix 7.x** / **catalox 4.x** from `@exellix/narrix-web-scoper` or `@x12i/funcx`).
+
+## [8.1.0] - 2026-05-28
+
+### Added
+
+- **Model profile aliases (`@x12i/ai-profiles`):** Resolve ai-profiles profile/shortcut names and legacy tiers (`weak`, `strong`, `default`) to concrete provider model ids for **ai-skills MAIN** (`skillModel` / `llmCall.model`). Keep **ai-profiles aliases** on **xynthesis** paths (`modelConfig.xynthesisModel`, POST/PRE/scoping `llmCall.model`). Exports: `resolveModelReference`, `resolveModelReferenceForXynthesis`, `normalizeXynthesisModelAlias`, `resolveRunTaskModelReferences`, `isResolvableModelAlias`, `formatResolvedProfileModelId`.
+- **Dual-root `RunTaskRequest.inputs`:** Optional top-level plural payload bucket (distinct from `executionMemory.inputs`); forwarded on MAIN like `input`.
+
+### Changed
+
+- **Smart-input validation context:** `inputs.*` resolves against top-level **`request.inputs`** only (not `executionMemory.inputs`). Use **`executionMemory.inputs.*`** for execution-memory plural paths.
+- **PRE `memoryPaths`:** Dual-root fallback — `executionMemory.input.<rest>` falls back to `executionMemory.inputs.<firstKey>.<rest>` when singular path is missing.
+- **`@exellix/xynthesis` ^4.0.6:** Xynthesis LLM calls use alias-only model resolution (`resolveModelReferenceForXynthesis`); default post-step model is **`balanced`** when omitted. **`skillModel`** resolution for **`@exellix/ai-skills` ^5.8.6** unchanged.
+- **Validation:** `modelConfig.xynthesisModel` rejects concrete provider ids; accepts ai-profiles aliases (replaces 8.0.0 `MODEL_CONFIG_NOT_RESOLVED` for legacy tiers on both slots).
+
+### Documentation
+
+- README — [Model profile aliases](#model-profile-aliases-x12iai-profiles); [`.docs/ai-tasks-model-profile-aliases-7x.md`](../archived/ai-tasks/.docs/ai-tasks-model-profile-aliases-7x.md).
+
+## [8.0.9] - 2026-05-28
+
+### Changed
+
+- **Dependencies (graph-engine alignment):** **`@x12i/funcx` ^4.2.0** (no 3.9.x path), **`@x12i/activix` ^8.0.5**, **`@x12i/catalox` ^5.1.1**; **`@x12i/funcx/functions`** re-exports (`FUNCX_EXECUTION_PLAN_FUNCTION_ID`, etc.).
+
+## [8.0.8] - 2026-05-27
+
+### Added
+
+- **`getModelCapabilities`** re-export from `@exellix/xynthesis` (replaces removed **`MODEL_CAPABILITIES`** constant).
+
+### Fixed
+
+- **Exports:** Removed broken **`MODEL_CAPABILITIES`** re-export from `@exellix/xynthesis` 4.x (symbol removed upstream). Use **`getModelCapabilities(modelId)`** from `@exellix/ai-tasks` or `@exellix/xynthesis` instead.
+
+### Deprecated
+
+- **`8.0.7`** — do not use; re-exported **`MODEL_CAPABILITIES`** against **`@exellix/xynthesis` ≥ 4.x** where the symbol no longer exists. Upgrade to **8.0.8+**.
+
+## [7.7.1] - 2026-05-25
+
+### Added
+
+- **`RunTaskRequest.taskConfiguration`:** Graph-engine may forward the node `taskConfiguration` blob; **`compileTaskConfigurationOnRunTaskRequest`** (also at **`runTask()`** entry) maps **`aiTaskStrategies.pre: "synthesis"`** and **`aiTaskProfile.inputSynthesis.enabled`** into PRE **`synthesized-context`** + **`includeContextInPrompt: true`**. Exports **`TaskConfigurationCompileError`**, **`AI_TASK_PRE_ENABLE_SYNTHESIS`**.
+
+### Documentation
+
+- **`RUNTASK_REQUEST.md`**, **README**, [`.docs/flow-io/xynthesis-pre.md`](../archived/ai-tasks/.docs/flow-io/xynthesis-pre.md), graph-engine task [`reports/graph-engine-task-pre-synthesis-compile.md`](../archived/ai-tasks/reports/graph-engine-task-pre-synthesis-compile.md).
+- Fixtures: [`test/fixtures/run-task/graph-engine-pre-synthesis.json`](test/fixtures/run-task/graph-engine-pre-synthesis.json), [`test/fixtures/graphs/graph-qcrbz6t-pre-synthesis.json`](test/fixtures/graphs/graph-qcrbz6t-pre-synthesis.json).
+
+## [7.6.4] - 2026-05-18
+
+### Changed
+
+- **`@exellix/ai-skills` ≥ 5.5.0:** Dependency **5.5.2**; `pickRunSkillRequestFields` forwards top-level **`executionMemory`** on `runSkill` (native `buildWorkingMemory` support). Removed the 7.6.3 `jobMemory.executionMemory` nest bridge.
+
+### Dependencies
+
+- **`@exellix/ai-skills`:** `5.5.2` (was `5.4.0`).
+
+## [7.6.3] - 2026-05-18
+
+### Fixed
+
+- **Gateway `executionMemory` paths (interim):** `DirectExecutionStrategy` nested `request.executionMemory` on `jobMemory.executionMemory` before `runSkill` for ai-skills **&lt; 5.5.0**. Superseded in **7.6.4** by native top-level forward. Strips `RunTaskRequest`-only fields from the `runSkill` payload.
+
+### Documentation
+
+- **Change request:** [`reports/change-request-tasknode-variable-scopes-7.6.2.md`](../archived/ai-tasks/reports/change-request-tasknode-variable-scopes-7.6.2.md) — upstream work for `@exellix/ai-skills`, `@x12i/rendrix`, and optional gateway path aliases.
+
+## [7.6.2] - 2026-05-18
+
+### Fixed
+
+- **`smartInput.strict`:** Accepted on `RunTaskRequest.smartInput` and forwarded through `normalizeSmartInputConfig` to the gateway path.
+
+### Changed
+
+- **MAIN template variables:** `runTask` forwards **`variables`** (job/graph bucket) **as-is** via `passthroughJobTemplateVariables` — no longer folds `jobContext`, `input`, `xynthesized`, or `smartInput` into `variables`. Node scope remains on **`executionMemory.taskVariables`**.
+
+### Documentation
+
+- **`RUNTASK_REQUEST.md`:** TaskNode authoring map (`taskConfiguration`, `inputsConfig`, `taskVariable`, two-bucket `jobVariables` / `taskVariables`), dual `outputValidation`, `inputSynthesis` compile table, forbidden metadata.
+- **README**, web-scoping doc, Narrix JSDoc: authoring path **`taskConfiguration.narrix`** (not `metadata.narrix`).
+
+### Schema
+
+- **`run-task-request.json`:** `executionStrategyCatalogItems`, `smartInputRenderOptions`, Rendrix path objects, `smartInput.strict`, variable-bucket descriptions.
+
+## [7.6.0] - 2026-05-18
+
+### Added
+
+- **`xynthesized.execution` scope:** `XynthesizedMemory.execution`, `XynthesizedDestinationScope` includes **`"execution"`**, PRE **`xynthesizedOutput.destination: "execution"`** writes and returns **`xynthesizedPatch.execution`** (same replace/merge semantics as job/task; no redirect to other buckets).
+- **Smart-input:** `xynthesized.*` paths must use scope **`job`**, **`task`**, or **`execution`**; **`resolveXynthesizedSmartInputPath`** helper for in-package resolution against request **`xynthesized`**.
+- **`TaskRequestBuilder.withXynthesizedExecution(key, value)`**.
+
+### Documentation
+
+- **`RUNTASK_REQUEST.md`**, **`documenations/synthesized-context-guide.md`**, and **`documenations/schemas/v1/run-task-request.json`** updated for the execution bucket.
+
+## [7.2.0] - 2026-05-06
+
+### Changed
+
+- **Breaking (exports):** Replaced **`AI_TASKS_FUNCX_*`** constants with **`FUNCX_EXECUTION_PLAN_FUNCTION_ID`** (`execution/plan`), **`FUNCX_EXECUTION_EVALUATE_RESULT_FUNCTION_ID`** (`execution/evaluate-result`), **`FUNCX_RESEARCH_PLAN_QUESTIONS_FUNCTION_ID`** (`research/plan-questions`).
+- **Execution strategies:** Planner and optimizer call **`run(functionId, payload)`** with the **generic FuncX envelope** only (`goal`, `input`, `context`, `result?`, `args`, `attribution`). Responses are adapted via **`adaptExecutionPlanResult`** / **`adaptEvaluateResultResponse`** ([`genericExecutionFuncxEnvelope.ts`](src/execution-strategies/genericExecutionFuncxEnvelope.ts)).
+- **`runPlanWebScopeQuestions`:** Optional **`backend: "funcx"`** for **`research/plan-questions`**; **`setResearchPlanQuestionsFuncxInvoker`** for tests.
+- **`ExecutionType`:** Re-exported from **`@exellix/ai-tasks`** via local [`executionType.ts`](src/types/executionType.ts) (`DIRECT`) because **`@exellix/ai-skills`** no longer exports this symbol on supported versions.
+- **`@x12i/funcx` ≥ 3.8.2:** **`unwrapFuncxRunValue`** delegates to **`getRunJsonResult`** (use **3.8.2+** for current normalization behavior). Re-exports: **`getRunJsonResult`**, **`isHttpRunSuccessBody`**, **`genericExecutionEnvelopeJsonSchema`**, **`buildAskAttribution`**, type **`GenericExecutionEnvelope`**.
+
+## [7.1.2] - 2026-05-06
+
+### Added
+
+- **`narrix.webScopeQuestions`:** optional explicit web-scope questions on `RunTaskRequest.narrix`; when `webScopeTemplates` is unset, the NARRIX preprocessor runs `@exellix/narrix-web-scoper` **`scopeQuestionPack`** and stores a normalized `WebScoperResult` in `executionMemory.webContext`.
+- **`runPlanWebScopeQuestions`** (and **`parseAndValidatePlanWebScopeQuestionsOutput`**, **`buildPlanWebScopeQuestionsUserPrompt`**) for AI-driven web-scope planning; FuncX path uses **`research/plan-questions`** (see **Unreleased** / **`FUNCX_RESEARCH_PLAN_QUESTIONS_FUNCTION_ID`**). FR tracker: **`documenations/funcx-scoping-integration-gaps.md`** §6.
+
+### Documentation
+
+- **`documenations/web-scoping-in-ai-tasks.md`**, **README**, **RUNTASK_REQUEST.md:** documented `webScopeQuestions` shape and precedence vs templates.
+
+## [7.1.1] - 2026-05-06
+
+### Added
+
+- **Package root type exports:** `XynthesizedMemory`, `SmartInputConfig`, `XynthesizedDestinationScope`, `XynthesizedOutputConfig`, `ExecutionStrategyInvocation`, `ExecutionStrategyPhase`, and `ExecutionStrategyWrapperKey` are re-exported from `@exellix/ai-tasks` alongside `RunTaskRequest`, so hosts can depend on stable names instead of `NonNullable<RunTaskRequest['xynthesized']>` or `RunTaskRequest['executionStrategies'][number]`.
+
+### Documentation
+
+- **CHANGELOG discipline:** Documented that `package.json` is authoritative for the shipped semver and pointed cross-version migration readers at `BREAKING-CHANGES.md` / `RUNTASK_REQUEST.md` when intermediate entries are missing.
+
+_Note: Releases **after 6.0.1** through **before 7.1.1** were not listed here chronologically; consult git history and `BREAKING-CHANGES.md` for that span._
+
+## [6.0.1] - 2026-05-05
+
+### Fixed
+
+- **Synthesis E2E gate:** `test/synthesis/e2e-synthesis-real.test.ts` no longer requires `LLM_MODEL_STRONG` / `LLM_MODEL_NORMAL`; it runs when **`RUN_SYNTHESIS_E2E=1`** and a recognized provider key is set (`OPEN_ROUTER_KEY`, `GROK_API_KEY`, etc.). Warns when a model env var is missing.
+
+### Documentation
+
+- **Xynthesis upstream work:** [`documenations/xynthesis-upstream-fixes-checklist.md`](documenations/xynthesis-upstream-fixes-checklist.md) (Activix `diagnostics.owner`, `jobId`/`taskId` hardening in `runSidekickGatewayCall`, tests to add in xynthesis).
+
+## [6.0.0] - 2026-05-04
+
+### Breaking
+
+- **`runTask` / `RunTaskRequest`:** `agentId`, `jobTypeId`, and `taskTypeId` are **required** again (aligned with `@exellix/ai-skills` `RunSkillRequest`). `runTask()` calls **`assertRequiredRunSkillCorrelation`** and throws if any value is missing or whitespace-only.
+- **`TaskRequestBuilder.build()`** throws unless those three fields were set (use **`withAiSkillsCorrelation`** or **`withAgentId` / `withJobTypeId` / `withTaskTypeId`**).
+- **`TaskConvenienceMethods.executeDirect`** now requires an **`options`** object including **`agentId`**, **`jobTypeId`**, and **`taskTypeId`** (third argument is no longer optional).
+- **`TaskConvenienceMethods.executeWithType`** now requires the **`options`** argument (same correlation fields as `RunTaskRequest`).
+
+### Added
+
+- Exported **`assertRequiredRunSkillCorrelation`** from the package root for hosts that validate payloads before calling `runTask`.
+
+## [5.7.1] - 2026-05-04
+
+### Fixed
+
+- **Types:** `RunTaskRequest` now **omits and re-declares** `agentId`, `jobTypeId`, and `taskTypeId` as **optional**, matching `runTask()` usage and latest `@exellix/ai-skills` `RunSkillRequest` where those fields are required. Callers can still pass them when the gateway requires them.
+
+## [5.7.0] - 2026-05-04
+
+### Added
+
+- **RunTask request contract documentation:** [`RUNTASK_REQUEST.md`](RUNTASK_REQUEST.md) — normative field list, template merge precedence, `executionMemory` vs `jobMemory.execution`, `executionPipeline` semantics, xynthesis visibility notes, and examples.
+- **JSON Schema fragment:** [`documenations/schemas/v1/run-task-request.json`](documenations/schemas/v1/run-task-request.json) and OpenAPI component `RunTaskRequestSubset` in [`documenations/schemas/openapi-3.1-components.yaml`](documenations/schemas/openapi-3.1-components.yaml).
+- **Request shape helpers** (exported from package root): `normalizeRunTaskRequest`, `collectRunTaskRequestWarnings`, `collectExecutionPipelineWarnings`, `isRunTaskRequestWarningsEnabled`.
+- **Env-gated diagnostics:** `AI_TASKS_REQUEST_WARNINGS=1` (also `true` / `yes`) logs structured JSON for deprecated duplicate fields, ignored pipeline steps, and observability hints at the start of each `runTask` call.
+
+### Changed
+
+- **Types:** `RunTaskRequest` documents deprecated mirrors: optional root `question`, `inputs`, and `jobInput` (typed for discoverability; removal deferred to a future major after hosts migrate).
+
+### Notes
+
+- **Additive / non-breaking:** Runtime merge behavior is unchanged; warnings are opt-in via env.
+- **Cross-package:** Minimum orchestrator (e.g. graph-engine) version is **not declared** in `@exellix/ai-tasks` — follow the consuming app’s pinned `@exellix/ai-tasks` range; see `RUNTASK_REQUEST.md` § Orchestrator integration.