@exellix/ai-tasks 9.1.0 → 10.0.0

package/.docs/possible-components/integration/platform.md

@@ -1,54 +0,0 @@
-# Platform integration: aifunctions-js, Activix, identity, Xynthesis
-
-**Single source of truth** for stack rules referenced from [`../pre-component/`](../pre-component/README.md) and [`../post-component/`](../post-component/README.md). Do not copy these sections into handler docs.
-
----
-
-## 1. Generative AI: aifunctions-js (not ai-gateway)
-
-- Use **[aifunctions-js](https://www.npmjs.com/package/aifunctions-js)** for text completions (`Client.ask`, presets, OpenRouter defaults as configured).
-- Do **not** use a legacy **ai-gateway** as the primary API. `{ invoke(instructions, workingMemory) }` shapes are **test/migration shims** only.
-- Parameters: system/user messages, optional `model`, `temperature`, `maxTokens`, `timeoutMs`, usage from response.
-
-In this repo, `src/internal/runLlmTextCall.ts` follows the same assumption.
-
----
-
-## 1b. Pre-task synthesis: @athenices/xynthesis
-
-- **Synthesized-context** is built on **@athenices/xynthesis**: templates, markdown vs structured modes, validation, `setSynthesisInvoker`.
-- **SynthesisInvoker** implementations perform HTTP/LLM via **aifunctions-js**, not ai-gateway.
-- Activix: record **pipeline_pre** / synthesis work with the same **`identity`** as the run (§3).
-
-Detail: [`../pre-component/handler-synthesized-context.md`](../pre-component/handler-synthesized-context.md).
-
----
-
-## 2. Activix
-
-- Record **LLM-backed** and meaningful sub-steps (scoping, synthesis, audit cycles, polish passes, System-2 plan), not only outer `runTask`.
-- Handlers receive **activix** + **correlationId** or use a shared wrapper (`withPhaseRecord`-style).
-
----
-
-## 3. Identity on Activix records
-
-Canonical: **`docs/activix-identity.md`** (repo root).
-
-| Rule | Detail |
-|------|--------|
-| Field | Top-level **`identity`** on records. |
-| Merge | `{ ...upstreamIdentity, taskId, skillId }`; runtime overrides duplicate keys. |
-| Propagation | Same `identity` on **every** phase: `narrix`, `pipeline_pre`, `direct`, `audit`, `polish`, `narrix_then_direct`, etc. |
-| Correlation | Prefer **`correlationId === taskId`** for timeline queries. |
-
----
-
-## 4. Summary
-
-| Concern | Choice |
-|---------|--------|
-| LLM transport | **aifunctions-js** |
-| Pre synthesis orchestration | **@athenices/xynthesis** + invoker → aifunctions-js |
-| Observability | **Activix** |
-| Task continuity | **`identity`** on all related records |

package/.docs/possible-components/integration/reintegrate-into-ai-tasks.md

@@ -1,26 +0,0 @@
-# Re-integrating Pre/Post packages into `ai-tasks`
-
-When pre/post exist as **separate packages** (or workspaces), this is how they plug back into **this** repo.
-
-## Orchestration anchor
-
-- **`src/core/task-sdk.ts`** — `WorexClientTasks.runTask`: narrix preprocessor, `executionPipeline` PRE/POST, `_executeDirect`, `_runSynthesizedContextPreStep`, `runAuditPostStep`, `runPolishPostStep`, Activix `withPhaseRecord`, `identity` / `effectiveIdentity`.
-
-## Integration steps (checklist)
-
-1. **Replace inline calls** with package entrypoints that accept the same **context** shapes as [`../pre-component/unified-protocol.md`](../pre-component/unified-protocol.md) and [`../post-component/unified-protocol.md`](../post-component/unified-protocol.md) (or thin adapters).  
-2. **Inject** Activix client + `correlationId` + merged **`identity`** into both engines per [`platform.md`](platform.md).  
-3. **Synthesis:** package must register **SynthesisInvoker** using **aifunctions-js**; keep `src/synthesis/synthesisGatewayCompat.ts` only if tests still need legacy `invoke`.  
-4. **Types:** re-export or depend on a shared types package; keep `RunTaskRequest`, `ExecutionStep`, `AuditConfig`, `PolishConfig`, `SynthesisConfig` aligned with `src/types/task-types.ts`.  
-5. **Templates (post):** ensure `templates/post-steps/**` are on disk or set `AUDIT_TEMPLATES_PATH` / `POLISH_TEMPLATES_PATH` — see [`../post-component/gaps-and-artifacts.md`](../post-component/gaps-and-artifacts.md).  
-6. **Tests:** point existing e2e tests at the integrated packages; keep gateway overrides working.
-
-## Identity & Activix
-
-- Follow **`docs/activix-identity.md`**; ensure every phase record from the imported engine includes **`identity`**.  
-- Phase labels to preserve for dashboards: `narrix`, `pipeline_pre`, `direct`, `audit`, `polish`, `narrix_then_direct`, etc.
-
-## Related repo docs
-
-- `docs/activix-identity.md`  
-- `documenations/identity-metadata-contract.md` (if present)

package/.docs/possible-components/integration/roadmap-and-checklists.md

@@ -1,54 +0,0 @@
-# Roadmap & checklists (cross-cutting)
-
-Phased work for **platform + pre engine + post engine + hardening**. Handler behavior lives only under [`../pre-component/`](../pre-component/README.md) and [`../post-component/`](../post-component/README.md).
-
----
-
-## Phase A — Shared foundation
-
-- [ ] **A1.** Activix wrapper: start/success/fail with mandatory **`identity`** (upstream + `taskId` / `skillId`).  
-- [ ] **A2.** **aifunctions-js** factory (model, timeout, usage).  
-- [ ] **A3.** **SynthesisInvoker** → **aifunctions-js** `ask` (see [`platform.md`](platform.md) §1b).
-
-## Phase B — Pre-Task engine
-
-- [ ] **B1.** Registry + merge `patches` / `prior` — [`../pre-component/unified-protocol.md`](../pre-component/unified-protocol.md).  
-- [ ] **B2.** `narrix-preprocessor` + web scope — [`../pre-component/handler-web-scope.md`](../pre-component/handler-web-scope.md).  
-- [ ] **B3.** `pipeline.synthesized-context` — [`../pre-component/handler-synthesized-context.md`](../pre-component/handler-synthesized-context.md).  
-- [ ] **B4.** `direct.ai-scoping` — [`../pre-component/handler-ai-scoping.md`](../pre-component/handler-ai-scoping.md).  
-- [ ] **B5.** Narrix System-2 — [`../pre-component/handler-narrix-system2.md`](../pre-component/handler-narrix-system2.md).
-
-## Phase C — Post-Task engine
-
-- [ ] **C1.** Runner + `currentOutput` chain — [`../post-component/unified-protocol.md`](../post-component/unified-protocol.md).  
-- [ ] **C2.** Audit — [`../post-component/handler-audit.md`](../post-component/handler-audit.md).  
-- [ ] **C3.** Polish — [`../post-component/handler-polish.md`](../post-component/handler-polish.md).  
-- [ ] **C4.** Audit `reRunMain` paths: same **`identity`** / Activix continuity.
-
-## Phase D — Hardening
-
-- [ ] **D1.** Env parity — see [`../pre-component/gaps-and-artifacts.md`](../pre-component/gaps-and-artifacts.md) and [`../post-component/gaps-and-artifacts.md`](../post-component/gaps-and-artifacts.md).  
-- [ ] **D2.** Test DI: mock clients, invoker, Activix.  
-- [ ] **D3.** Drift review vs this spec tree.
-
----
-
-## Pre-task layering (reference)
-
-```text
-RunTaskRequest + identity
-        │
-        ▼
-┌───────────────────┐
-│  Pre-Task Engine  │  Activix + identity per phase
-└─────────┬─────────┘
-          │
-    ┌─────┴─────┬─────────────────┬──────────────────┐
-    ▼           ▼                 ▼                  ▼
- narrix     synthesized-      ai-scoping      (extensible)
-            context
-            │
-            ├── resolveSourceMaterial (host)
-            ├── @athenices/xynthesis
-            └── SynthesisInvoker ──► aifunctions-js
-```

package/.docs/possible-components/post-component/README.md

@@ -1,18 +0,0 @@
-# Post-Task component — spec pack
-
-**Start here to implement:** [`builder-guide.md`](builder-guide.md) (complete `AuditConfig` / `PolishConfig`, LLM output shapes, host prerequisites, file manifest, acceptance).  
-Platform rules: [`../integration/platform.md`](../integration/platform.md).
-
-## Read order
-
-1. [`builder-guide.md`](builder-guide.md) — **full implementation entry**.  
-2. [`unified-protocol.md`](unified-protocol.md) — protocol, engine duties.  
-3. [`handler-audit.md`](handler-audit.md)  
-4. [`handler-polish.md`](handler-polish.md)  
-5. [`gaps-and-artifacts.md`](gaps-and-artifacts.md) — template paths, packaging, known gaps.
-
-## Outside this folder
-
-- **Platform:** [`../integration/platform.md`](../integration/platform.md)  
-- **Checklist (Post Phase C):** [`../integration/roadmap-and-checklists.md`](../integration/roadmap-and-checklists.md)  
-- **Merge back into repo:** [`../integration/reintegrate-into-ai-tasks.md`](../integration/reintegrate-into-ai-tasks.md)

package/.docs/possible-components/post-component/builder-guide.md

@@ -1,175 +0,0 @@
-# Post-Task component — **builder guide** (single entry for implementation)
-
-Everything required to implement the **pipeline POST** engine. **Full** Activix/identity rules: [`../integration/platform.md`](../integration/platform.md).
-
----
-
-## 1. What you are building
-
-An engine that runs **after MAIN** when `executionPipeline` contains `phase: "post"` steps. It:
-
-- Takes **MAIN** `RunTaskResponse`, **`contextMarkdown`** actually sent to MAIN, and **`reRunMain(overrideContext)`** for audit loops
-- Runs handlers in **pipeline array order** (e.g. audit then polish)
-- Produces final **`outputText`**, merges **`metadata.postSteps`**, appends **`intermediateSteps`**
-
-**Protocol:** [`unified-protocol.md`](unified-protocol.md).
-
----
-
-## 2. Stack (minimum)
-
-| Topic | Rule |
-|-------|------|
-| LLM | **aifunctions-js** via shared `runLlmTextCall` pattern; test gateways `setAuditGateway`, `setPolishGateway`. |
-| Xynthesis | **Not** used in default audit/polish. |
-| Observability | **Activix** on post phases `audit`, `polish`; same **`identity`** as parent run; MAIN re-runs from audit must keep identity. |
-
----
-
-## 3. Host prerequisites
-
-| Requirement | Why |
-|-------------|-----|
-| **Exactly one** `main` step in `executionPipeline` | Current SDK invariant. |
-| **Capture context** | Host must record the markdown context passed to MAIN (`captureContext` pattern) so audit/polish receive **`contextMarkdown`**. |
-| **`reRunMain`** | Callback re-executes MAIN with **only** context replaced by `contextMarkdown + feedback` (audit). |
-
----
-
-## 4. NPM dependencies (from `@woroces/ai-tasks`)
-
-| Package | Role |
-|---------|------|
-| `aifunctions-js` | Audit, polish, audit merge LLM. |
-| `handlebars` | Template render in `loadAuditTemplates` / `loadPolishTemplates`. |
-| `nx-cache` | Optional audit result cache (cycle 1, default guidelines). |
-
-Types may come from `@woroces/ai-skills` / shared types package (`ModelConfig`, `RunTaskRequest`, `RunTaskResponse`).
-
----
-
-## 5. Config types (complete)
-
-### `AuditConfig` (`src/types/task-types.ts`)
-
-| Field | Type | Default / notes |
-|-------|------|------------------|
-| `gateway.must` | `AuditCheck[]` | Required list; `{ check, weight }`. |
-| `gateway.should` | `AuditCheck[]` | Required list. |
-| `threshold` | number | Default **80** (weighted score 0–100). |
-| `minCycles` | number | Default **1**. |
-| `maxCycles` | number | Default **3**. |
-| `selectionStrategy` | `"best"` \| `"synthesis"` | Merge two candidates + re-audit when `synthesis`. |
-| `auditModelConfig` | `ModelConfig?` | Plus env `AUDIT_MODEL`, `POST_STEP_MODEL`. |
-| `synthesisModelConfig` | `ModelConfig?` | Merge step; env `AUDIT_SYNTHESIS_MODEL`, `POST_STEP_MODEL`, main model fallback. |
-| `customAuditGuidelines` | string? | Appended to audit system template. |
-| `customSynthesisGuidelines` | string? | Merge template. |
-| `fallbackToBest` | boolean | Default **true** if merge path loses. |
-| `auditTimeoutMs` | number | Env `AUDIT_TIMEOUT_MS`, `POST_STEP_TIMEOUT_MS` — default **30_000**. |
-
-### `PolishConfig`
-
-| Field | Type | Notes |
-|-------|------|--------|
-| `checklist` | `PolishChecklistItem[]` | `{ instruction, priority: high\|medium\|low }`. |
-| `maxPasses` | number | Default **1**. |
-| `modelConfig` | `ModelConfig?` | Env `POLISH_MODEL`, `POST_STEP_MODEL`. |
-| `customGuidelines` | string? | |
-| `includeOriginalContext` | boolean? | When true, pass `originalInput` + `promptContext` into user prompt. |
-| `timeoutMs` | number | Env `POLISH_TIMEOUT_MS`, `POST_STEP_TIMEOUT_MS` — default **30_000**. |
-
----
-
-## 6. LLM output contracts
-
-### Audit (`parseAuditOutputFromMarkdown`)
-
-Model should emit markdown with:
-
-- `### Checks` — blocks containing **Check / Kind (must|should) / Passed (yes|no) / Feedback** (see `src/post-steps/audit/parseAuditOutput.ts` regexes).  
-- `### Overall feedback` — free text.
-
-Parser fills `AuditLLMOutput`; host computes **weighted score** from gateway weights and **must** all-pass.
-
-### Polish (`parsePolishOutput`)
-
-Model should return **JSON**:
-
-```json
-{
-  "polishedOutput": "string",
-  "changeNotes": [
-    { "instruction": "string", "action": "changed" | "no-change", "note": "string" }
-  ]
-}
-```
-
-Parser may need fence stripping (implement as in repo).
-
----
-
-## 7. Environment variables
-
-| Variable | Purpose |
-|----------|---------|
-| `AUDIT_MODEL`, `AUDIT_SYNTHESIS_MODEL`, `POST_STEP_MODEL` | Model fallbacks |
-| `AUDIT_TIMEOUT_MS`, `POST_STEP_TIMEOUT_MS` | Audit |
-| `POLISH_MODEL`, `POLISH_TIMEOUT_MS` | Polish |
-| `AUDIT_TEMPLATES_PATH`, `POLISH_TEMPLATES_PATH` | Override cwd base for template discovery |
-
----
-
-## 8. Source file manifest
-
-| Path | Role |
-|------|------|
-| `src/post-steps/audit/runAudit.ts` | Orchestration, cycles, synthesis selection |
-| `src/post-steps/audit/runAuditCall.ts` | Single LLM call |
-| `src/post-steps/audit/loadAuditTemplates.ts` | Disk + Handlebars |
-| `src/post-steps/audit/parseAuditOutput.ts` | Markdown → checks |
-| `src/post-steps/polish/runPolish.ts` | Multi-pass |
-| `src/post-steps/polish/runPolishCall.ts` | Single LLM |
-| `src/post-steps/polish/loadPolishTemplates.ts` | Disk + Handlebars |
-| `src/post-steps/polish/parsePolishOutput.ts` | JSON |
-| `src/post-steps/resolvePostStepConfig.ts` | `resolveModel`, `resolveTimeout` |
-| `src/internal/runLlmTextCall.ts` | Shared LLM |
-
-Templates: see [`gaps-and-artifacts.md`](gaps-and-artifacts.md).
-
----
-
-## 9. Response shaping (host)
-
-After post pipeline:
-
-- Set `rawText` / `parsed` to final string (current SDK sets both to final text).  
-- `metadata.postSteps = { audit?, polish? }` with `AuditPostStepMetadata` / `PolishPostStepMetadata`.  
-- Renumber or append `intermediateSteps` (synthesis step from PRE may precede in full response).
-
----
-
-## 10. Definition of done
-
-- [ ] POST steps run in pipeline order; `currentOutput` chains correctly.  
-- [ ] Audit: cache behavior matches spec when `customAuditGuidelines` unset; throws if all cycles fail LLM; `reRunMain` receives extended context.  
-- [ ] `selectionStrategy: synthesis` merges top two, re-audits, respects `fallbackToBest`.  
-- [ ] Polish: multi-pass with previous notes; parse failure stops pass loop gracefully.  
-- [ ] Templates load from disk with env override; ship `templates/post-steps/**` in package.  
-- [ ] Activix + identity on audit/polish phases and on MAIN re-entry.
-
----
-
-## 11. Types (canonical source)
-
-`AuditConfig`, `AuditPostStepMetadata`, `AuditCheck`, `AuditResult`, `PolishConfig`, `PolishPostStepMetadata`, `PolishLLMOutput`, `RunTaskResponsePostStepsMetadata`, `ExecutionStep`, `IntermediateStep` — **`src/types/task-types.ts`**.
-
----
-
-## 12. Read next
-
-1. [`unified-protocol.md`](unified-protocol.md)  
-2. [`handler-audit.md`](handler-audit.md)  
-3. [`handler-polish.md`](handler-polish.md)  
-4. [`gaps-and-artifacts.md`](gaps-and-artifacts.md)  
-5. [`../integration/roadmap-and-checklists.md`](../integration/roadmap-and-checklists.md) Phase C  
-6. [`../integration/reintegrate-into-ai-tasks.md`](../integration/reintegrate-into-ai-tasks.md)

package/.docs/possible-components/post-component/gaps-and-artifacts.md

@@ -1,52 +0,0 @@
-# Post-Task component — gaps, code, templates
-
-What to **ship, copy, or improve** for a standalone post package. **Complete config, env, contracts, acceptance:** [`builder-guide.md`](builder-guide.md).
-
-## On-disk templates (repo — **must ship** with post component)
-
-Loaded from `templates/post-steps/` relative to `process.cwd()` unless overridden by env.
-
-### Audit (`AUDIT_TEMPLATES_PATH` or cwd)
-
-| File | Purpose |
-|------|---------|
-| `templates/post-steps/audit/system.md` | Auditor system prompt (Handlebars). |
-| `templates/post-steps/audit/user.txt` | Auditor user template (`.txt`, not `.md` — matches `loadAuditTemplates.ts`). |
-| `templates/post-steps/audit/feedback-injection.md` | Feedback block appended to context for MAIN re-run. |
-| `templates/post-steps/audit/synthesis.md` | Merge-two-candidates system prompt (`selectionStrategy: synthesis`). |
-
-**Fallback:** in-code defaults in `loadAuditTemplates.ts` if files missing (runtime still works; packaging should include files for consistency).
-
-### Polish (`POLISH_TEMPLATES_PATH` or cwd)
-
-| File | Purpose |
-|------|---------|
-| `templates/post-steps/polish/system.md` | Polish system prompt. |
-| `templates/post-steps/polish/user.txt` | Polish user template (**`.txt`**). |
-
-## Code modules (anchors)
-
-| Path | Role |
-|------|------|
-| `src/post-steps/audit/runAudit.ts` | Cycle orchestration, synthesis selection |
-| `src/post-steps/audit/runAuditCall.ts` | Single audit LLM |
-| `src/post-steps/audit/loadAuditTemplates.ts` | Disk load + Handlebars compile helpers |
-| `src/post-steps/audit/parseAuditOutput.ts` | Markdown → `AuditLLMOutput` |
-| `src/post-steps/polish/runPolish.ts` | Multi-pass loop |
-| `src/post-steps/polish/runPolishCall.ts` | Single polish LLM |
-| `src/post-steps/polish/loadPolishTemplates.ts` | Disk load |
-| `src/post-steps/polish/parsePolishOutput.ts` | JSON → `PolishLLMOutput` |
-| `src/post-steps/resolvePostStepConfig.ts` | `resolveModel`, `resolveTimeout` |
-| `src/internal/runLlmTextCall.ts` | Shared LLM primitive |
-
-## Possible gaps / improvements
-
-| Gap | Suggestion |
-|-----|------------|
-| **user.txt** naming | Docs sometimes say “user.md”; code reads **`user.txt`** — keep extension aligned or add fallback read for `.md`. |
-| **nx-cache** | Audit caching is in-memory package `nx-cache`; document TTL (300s) and cache key fields for distributed deployments (may need Redis). |
-| **Strict JSON polish** | No provider JSON mode in spec; consider optional `response_format` when extracting. |
-
-## Environment
-
-Full list: [`builder-guide.md`](builder-guide.md) §7.

package/.docs/possible-components/post-component/handler-audit.md

@@ -1,47 +0,0 @@
-# Handler: audit quality gate
-
-> **Platform:** [`../integration/platform.md`](../integration/platform.md).
-
-## Purpose
-
-LLM evaluates MAIN output against **must** / **should** checks; weighted score; optional MAIN **re-run** with feedback; optional **merge** of top two candidates + re-audit.
-
-## Trigger (current repo)
-
-- `executionPipeline` POST `{ type: "audit", config?: AuditConfig }`
-- Needs **captured** `contextMarkdown` from MAIN (`task-sdk` pipeline path).
-- `src/post-steps/audit/runAudit.ts`, `runAuditCall.ts`
-
-## Config (`AuditConfig`)
-
-Full field list and defaults: [`builder-guide.md`](builder-guide.md) §5.
-
-## Loop (detailed)
-
-1. **Candidate text** — `mainResult.rawText` or `JSON.stringify(mainResult.parsed ?? …)`.  
-2. **Cache (cycle 1 only)** — If `customAuditGuidelines` is unset, SHA-256 key over `originalInput`, `contextMarkdown`, `currentOutput`, serialized must/should lists; **nx-cache** TTL **300_000 ms**. Skip cache when custom guidelines set.  
-3. **`runAuditCall`** — `runLlmTextCall` + `setAuditGateway` test override; timeout `auditTimeoutMs` / env.  
-4. **Parse** — `parseAuditOutputFromMarkdown`; on hard parse failure use empty checks + placeholder feedback (do not throw).  
-5. **Score** — `weightedScore = round((passedWeight / totalWeight) * 100)`; **must** = all `gateway.must` checks have matching parsed row with `passed: true`.  
-6. **Exit** when `mustOk && weightedScore >= threshold && cycle >= minCycles`, or `cycle >= maxCycles`.  
-7. **Continue** — `renderFeedbackInjection` → append to `contextMarkdown` → `reRunMain(nextContext)` → refresh `currentOutput` from new MAIN result.  
-8. **`runAuditCall` throws** — increment failure counter; push failed `intermediateStep`; if failures **≥ maxCycles**, throw `Audit: all cycles failed` with `phase: post:audit`.
-
-## Selection after loop
-
-- **`best`:** Pick candidate with highest `weightedScore` (implementation may prefer later cycle on tie).  
-- **`synthesis`:** If `candidates.length >= 2`, sort by score, take top **A** and **B**; `runLlmTextCall` with **system prompt only** from `renderSynthesis` (`userPrompt: ""`); on error use **A**’s output; **re-audit** merged text; keep merged only if merged score **strictly** beats best single and `fallbackToBest` allows.
-
-## LLM markdown shape (for prompt authors)
-
-Templates must encourage output compatible with `parseAuditOutputFromMarkdown`: sections `### Checks` and `### Overall feedback`, and per-check lines **Check / Kind / Passed / Feedback** (see `src/post-steps/audit/parseAuditOutput.ts`).
-
-## Outputs
-
-- **`outputText`:** Final string.  
-- **`AuditPostStepMetadata`:** `totalCycles`, `candidateScores`, `selectedCycleIndex`, `allMustPassed`, `synthesisUsed`, `auditResults`, `durationMs`.  
-- **`intermediateSteps`:** Per cycle + failures.
-
-## Types
-
-`AuditCheck`, `AuditResult`, `AuditConfig`, … — `src/types/task-types.ts`.

package/.docs/possible-components/post-component/handler-polish.md

@@ -1,41 +0,0 @@
-# Handler: polish / checklist refinement
-
-> **Platform:** [`../integration/platform.md`](../integration/platform.md).
-
-## Purpose
-
-One or more LLM passes over **currentOutput** with a prioritized checklist; expect JSON `{ polishedOutput, changeNotes }`.
-
-## Trigger (current repo)
-
-- POST `{ type: "polish", config?: PolishConfig }`
-- Runs **after** audit when both exist (`src/post-steps/polish/runPolish.ts`, `runPolishCall.ts`).
-
-## Config (`PolishConfig`)
-
-Checklist items with `high` | `medium` | `low`; `maxPasses`; `modelConfig`; `timeoutMs`; `customGuidelines`; `includeOriginalContext` gates `originalInput` + `promptContext` in prompt.
-
-## Pass flow
-
-`loadPolishTemplates` → `renderPolishSystem` / `renderPolishUser` (Handlebars; system gets `customGuidelines`, `passNumber`, optional `previousPassNotes`) → `runPolishCall` → `parsePolishOutput`.  
-**Success:** update working output, accumulate `changeNotes`, push ok `intermediateStep`.  
-**Parse or LLM failure:** push failing step, **break** loop; return last good `polishedOutput` (or initial `currentOutput`).
-
-## Expected JSON (model)
-
-Align prompts with:
-
-```json
-{
-  "polishedOutput": "<full revised text>",
-  "changeNotes": [
-    { "instruction": "<checklist line>", "action": "changed", "note": "…" }
-  ]
-}
-```
-
-`action` is `"changed"` or `"no-change"`. System template in repo defaults asks for this shape explicitly.
-
-## Types
-
-`PolishConfig`, `PolishLLMOutput`, `PolishPostStepMetadata` — `src/types/task-types.ts`. Full config table: [`builder-guide.md`](builder-guide.md) §5.

package/.docs/possible-components/post-component/unified-protocol.md

@@ -1,59 +0,0 @@
-# Post-Task component — unified protocol
-
-> **Scope:** Target architecture only.  
-> **Shared platform:** [`../integration/platform.md`](../integration/platform.md).
-
-## Goal
-
-One engine that runs **after MAIN**: consumes MAIN result, **context markdown** used for MAIN, and `reRunMain` for audit loops; outputs final text + `postSteps` metadata + intermediate steps.
-
-## Protocol (conceptual)
-
-```ts
-type PostHandlerId = "audit" | "polish" | string;
-
-interface PostHandlerContext {
-  request: RunTaskRequest;
-  identity?: Record<string, unknown>;
-  mainResult: RunTaskResponse;
-  contextMarkdown: string;
-  currentOutput: string;
-  config: unknown;
-  reRunMain: (overrideContext: string) => Promise<RunTaskResponse>;
-  jobId?: string;
-  agentId?: string;
-  correlationId?: string;
-  activix?: unknown;
-}
-
-interface PostHandlerResult {
-  ok: boolean;
-  outputText: string;
-  metadata?: Record<string, unknown>;
-  intermediateSteps?: IntermediateStep[];
-  error?: { code: string; message: string };
-}
-
-interface PostTaskHandler {
-  id: PostHandlerId;
-  order: number;
-  shouldRun(ctx: PostHandlerContext): boolean;
-  run(ctx: PostHandlerContext): Promise<PostHandlerResult>;
-}
-```
-
-**Chaining:** `currentOutput` flows audit → polish. Only **audit** uses `reRunMain`. **@athenices/xynthesis** is **not** in the default post design (audit/polish call **aifunctions-js** directly).
-
-## Built-in handlers
-
-| id | Pipeline `type` | Spec |
-|----|-----------------|------|
-| `audit` | `audit` | [`handler-audit.md`](handler-audit.md) |
-| `polish` | `polish` | [`handler-polish.md`](handler-polish.md) |
-
-## Engine duties
-
-- Walk `executionPipeline` POST steps in order.  
-- Seed `currentOutput` from MAIN (`rawText` or JSON `parsed`).  
-- Merge each handler’s `intermediateSteps` and `metadata` into `response.metadata.postSteps`.  
-- Final `rawText` / `parsed` = last `outputText`.

package/.docs/possible-components/pre-component/README.md

@@ -1,22 +0,0 @@
-# Pre-Task component — spec pack
-
-**Start here to implement:** [`builder-guide.md`](builder-guide.md) (dependencies, host contract, env, file manifest, acceptance criteria).  
-Shared platform (Activix / identity / aifunctions / Xynthesis) is defined **once** in [`../integration/platform.md`](../integration/platform.md); the builder guide summarizes only what you must comply with.
-
-## Read order
-
-1. [`builder-guide.md`](builder-guide.md) — **full implementation entry**.  
-2. [`unified-protocol.md`](unified-protocol.md) — protocol types, handler index, execution order.  
-3. Handler specs:  
-   - [`handler-narrix-preprocessor.md`](handler-narrix-preprocessor.md)  
-   - [`handler-web-scope.md`](handler-web-scope.md)  
-   - [`handler-narrix-system2.md`](handler-narrix-system2.md)  
-   - [`handler-synthesized-context.md`](handler-synthesized-context.md) (**@athenices/xynthesis**)  
-   - [`handler-ai-scoping.md`](handler-ai-scoping.md)  
-4. [`gaps-and-artifacts.md`](gaps-and-artifacts.md) — optional externalizations, open decisions.
-
-## Outside this folder
-
-- **LLM, Activix, identity, Xynthesis vs aifunctions-js:** [`../integration/platform.md`](../integration/platform.md)  
-- **Phased checklist (includes Pre Phase B):** [`../integration/roadmap-and-checklists.md`](../integration/roadmap-and-checklists.md)  
-- **Merging back into `ai-tasks`:** [`../integration/reintegrate-into-ai-tasks.md`](../integration/reintegrate-into-ai-tasks.md)