В API генерации Task Pack добавь булево поле, показывающее, был ли refinement...

[drag1web]

ContextForge Task Pack

Task Pack: #21 — В API генерации Task Pack добавь булево поле, показывающее, был ли refinement...
Project: contextforge
Task type: general
Target: codex
Generated: 2026-07-16T16:36:33.446Z
GitHub issue draft created: 2026-07-16T16:40:44.384Z

Original Task

В API генерации Task Pack добавь булево поле, показывающее, был ли refinement получен из кеша.

Generated Task Pack

Open generated prompt

AI Task Pack

Target Tool

Codex

Task Type

general

Task

В API генерации Task Pack добавь булево поле, показывающее, был ли refinement получен из кеша.

Project Context

• Project: contextforge
• Package manager: npm
• Detected stack: React, Svelte, TypeScript, Vite, Electron, Express, PostgreSQL, SQLite, Tailwind CSS, Framer Motion, Docker
• Readiness score: 86

Project metadata:

{
  "name": "contextforge",
  "projectRoot": "<local-project>",
  "packageManager": "npm",
  "detectedStack": [
    "React",
    "Svelte",
    "TypeScript",
    "Vite",
    "Electron",
    "Express",
    "PostgreSQL",
    "SQLite",
    "Tailwind CSS",
    "Framer Motion",
    "Docker"
  ],
  "scripts": {
    "apps/desktop/renderer:dev": "vite --host 127.0.0.1 --port 5173 --strictPort",
    "dev": "vite --host 127.0.0.1 --port 5173 --strictPort",
    "apps/desktop/renderer:build": "tsc -b && vite build",
    "build": "tsc -b && vite build",
    "apps/desktop/renderer:preview": "vite preview",
    "preview": "vite preview",
    "root:dev": "concurrently -k -n SERVER,RENDERER,ELECTRON -c blue,green,magenta \"npm run dev:server\" \"npm run dev:renderer\" \"npm run dev:electron\"",
    "dev:server": "npm run dev -w @contextforge/server",
    "dev:renderer": "npm run dev -w @contextforge/renderer",
    "dev:electron": "wait-on tcp:5173 tcp:4000 && cross-env ELECTRON_RENDERER_URL=http://localhost:5173 electron apps/desktop/electron/main.cjs",
    "root:build": "npm run build -w @contextforge/renderer && npm run build -w @contextforge/server",
    "test:selector": "npm run test:selector -w @contextforge/server",
    "test:selector:rollout": "npm run test:selector:rollout -w @contextforge/server",
    "test:selector:rollout-ui": "npm exec -w @contextforge/server -- tsx ../scripts/pipelineBadge.smoke.ts",
    "benchmark:selector": "npm run benchmark:selector -w @contextforge/server",
    "benchmark:selector:validation": "npm run benchmark:selector:validation -w @contextforge/server",
    "benchmark:selector:snapshot": "npm run benchmark:selector:snapshot -w @contextforge/server",
    "db:up": "docker compose up -d",
    "db:down": "docker compose down",
    "db:logs": "docker compose logs -f postgres",
    "server:dev": "tsx watch src/index.ts",
    "server:build": "tsc",
    "server:test:selector": "tsx src/ollama/taskFileSelector.smoke.ts",
    "server:test:selector:replay": "tsx src/ollama/taskFileSelector.replay.ts",
    "test:selector:replay": "tsx src/ollama/taskFileSelector.replay.ts",
    "server:test:selector:benchmark": "tsx src/selection/benchmark/benchmarkSmoke.ts",
    "test:selector:benchmark": "tsx src/selection/benchmark/benchmarkSmoke.ts",
    "server:test:selector:rollout": "tsx src/selection/pipelineRollout.smoke.ts",
    "server:benchmark:selector": "tsx src/selection/benchmark/benchmarkRunner.ts",
    "server:benchmark:selector:validation": "tsx src/selection/benchmark/benchmarkRunner.ts --split validation --external-only",
    "server:benchmark:selector:snapshot": "tsx src/selection/benchmark/benchmarkSnapshot.ts",
    "server:start": "node dist/index.js",
    "start": "node dist/index.js"
  },
  "readinessScore": 86
}

Relevant File Candidates

Inspect these files before modifying code:

• server/src/routes/taskPacks.ts
  • kind: source
  • usage: inspect-and-edit
  • evidence: code-graph support signal: 86%
  • size: 80 KB
  • reason: API boundary imports the existing producer and is the implementation owner for exposing cached as the public boolean field generationcached.
• server/src/ollama/taskPackGenerationReliability.ts
  • kind: source
  • usage: inspect-only
  • evidence: code-graph support signal: 84%
  • size: 72 KB
  • reason: Existing producer already exposes the related value (cached); reuse it as the source of truth instead of creating a duplicate field inside the generated payload schema.
• apps/desktop/renderer/src/types/index.ts
  • kind: source
  • usage: inspect-only
  • evidence: code-graph support signal: 80%
  • size: 27 KB
  • reason: Existing client/shared contract already contains the related public field (generationcached); retain it to verify the API response shape without adding UI state or unrelated display files.

---

Code Context Snippets

These snippets are partial context only. Inspect full files before editing.

server/src/routes/taskPacks.ts

import { createHash } from "node:crypto";
import fs from "node:fs/promises";
import path from "node:path";
import { Router } from "express";
import { z } from "zod";

import { storage } from "../storage/index.js";
import {
  appendSelectorDiagnostics,
  getAppSettings,
} from "../settings/settingsService.js";
import {
  buildTaskPackRulesTemplatePrompt,
  RulesServiceError,
} from "../rules/rulesService.js";
import {
  analyzeTaskIntent,
  type TaskIntentAnalysis,
} from "../ollama/taskIntentAnalyzer.js";
import {
  generateReliableTaskPack,
  type TaskPackGenerationDiagnostics,
} from "../ollama/taskPackGenerationReliability.js";
import {
  type TaskFileSelection,
  type SelectedTaskFileUsage,
} from "../ollama/taskFileSelector.js";
import {
  scanProjectInventory,
  type ProjectInventory,
  type ProjectInventoryFile,
  type ProjectInventoryFileKind,
} from "../scanner/projectInventoryScanner.js";
import {
  evaluateContextSelectionQuality,
  type ContextSelectionQuality,
} from "../selection/contextQuality.js";
import { isSecretLikePath } from "../selection/safetyPolicy.js";
import type { FileSelectionEvidence } from "../selection/repositorySemanticIndex.js";
import { buildExportSafeProjectMetadata } from "../taskPacks/taskPackPrivacy.js";
import { resolveTaskUnderstandingInteraction } from "../taskPacks/taskUnderstandingInteraction.js";
import {
  applySelectionEvidenceGate,
  buildTaskExecutionContractFromIntent,
  type TaskExecutionContract,
} from "../taskPacks/taskExecutionContract.js";
import {
  applyTaskClarifications


<!-- Snippet truncated. Inspect the full file before editing. -->

server/src/ollama/taskPackGenerationReliability.ts

import { z } from "zod";

import type { TaskExecutionContract } from "../taskPacks/taskExecutionContract.js";
import type { AiProviderId } from "../ai/providerService.js";
import type { FileSelectionEvidence } from "../selection/repositorySemanticIndex.js";
import { recordPerformanceCacheEvent } from "../performance/performanceTrace.js";
import { generateWithConfiguredAi } from "../ai/providerService.js";
import {
  getAppSettings,
  type AppSettings,
} from "../settings/settingsService.js";
import {
  buildGenerationCacheKey,
  getCachedGeneration,
  setCachedGeneration,
} from "./generationCache.js";
import {
  extractExplicitReplacementValue,
  hasMissingReplacementValue,
} from "./taskValueGrounding.js";

export {
  extractExplicitReplacementValue,
  type ExplicitReplacementValue,
} from "./taskValueGrounding.js";

const MAX_PROMPT_CHARS = 24_000;
const MAX_REPAIR_RESPONSE_CHARS = 8_000;
const MAX_RETRY_PROMPT_CHARS = 28_000;
const MAX_ITEM_CHARS = 600;

const REFINEMENT_ARRAY_LIMITS = {
  implementationGuidance: 10,
  constraints: 8,
  acceptanceCriteria: 10,
  verificationSteps: 10,
  finalResponseRequirements: 8,
} as const;

const POLICY_REFINEMENT_ARRAY_LIMITS = {
  implementationGuidance: 7,
  constraints: 5,
  acceptanceCriteria: 6,
  verificationSteps: 5,
  finalResponseRequirements: 4,
} as const;

const boundedLineSchema = z
  .string()
  .trim()
  .min(3)
  .max(MAX_ITEM_CHARS)
  .transform((value) => normalizeGuidanceLine(value));

export const taskPackRefinementSchema = z.object({
  implementationGuidance: z
    .array(boundedLineSchema)
    .min(1)
    .ma


<!-- Snippet truncated. Inspect the full file before editing. -->

apps/desktop/renderer/src/types/index.ts

export interface ReadinessCheck {
  key: string;
  label: string;
  passed: boolean;
  points: number;
  message: string;
}

export interface ScannerPackageSummary {
  path: string;
  name: string | null;
  scripts: Record<string, string>;
}

export interface ScannerSignals {
  packageFiles: string[];
  docs: string[];
  envExamples: string[];
  testFiles: string[];
  testConfigs: string[];
  ciFiles: string[];
  lockFiles: string[];
  configs: string[];
  directories: string[];
  commands: {
    dev: string | null;
    build: string | null;
    test: string | null;
    typecheck: string | null;
    lint: string | null;
  };
  packages: ScannerPackageSummary[];
  inventory: {
    totalFiles: number;
    totalDirectories: number;
    truncated: boolean;
    maxDepth: number;
    maxEntries: number;
  };
}

export interface ReadinessReport {
  score: number;
  checks: ReadinessCheck[];
  issues: string[];
  signals?: ScannerSignals;
}

export type TargetTool = "codex" | "cursor" | "claude" | "gemini" | "generic";

export type TemplateTaskType =
  | "general"
  | "ui"
  | "backend"
  | "fullstack"
  | "build"
  | "bugfix"
  | "refactor"
  | "docs"
  | "tests";

export type RuleCategory =
  | "general"
  | "ui"
  | "backend"
  | "bugfix"
  | "refactor"
  | "docs"
  | "tests"
  | "assets"
  | "verification";

export interface PromptTemplate {
  id: string;
  name: string;
  description: string;
  targetTool: TargetTool;
  taskType: TemplateTaskType;
  content: string;
  isBuiltin: boolean;
  createdAt?: string;
  updatedAt?: string;
}

export interface RuleItem {
  id: string;
 


<!-- Snippet truncated. Inspect the full file before editing. -->

---

ContextForge Assisted Notes

Task Intent Analysis

• Source: ollama
• Task area: backend
• Risk level: medium
• Intent confidence: 0.95
• Intent tags: form-flow, backend-flow
• Domain terms: генерации, task, pack, булево, поле, показывающее, был, refinement, получен
• File role hints: component, state, style, api, route, service
• Structured targets: none
• Edit scope: target_with_supporting_context
• Understanding readiness: ready
• Understanding action: update
• Understanding goal: Add a boolean field to the Task Pack generation API indicating if refinement was retrieved from cache.
• Interpretation risk: objective
• Change definition: bounded

Execution Contract

• Mode: implementation
• Implementation guidance allowed: yes
• Required layers: backend
• Confirmed targets: server/src/routes/taskPacks.ts
• Proposed targets: none
• Target evidence: server/src/ollama/taskPackGenerationReliability.ts=graph_supported, server/src/routes/taskPacks.ts=graph_supported, apps/desktop/renderer/src/types/index.ts=graph_supported
• Implementation gate reasons: none
• Unresolved decisions: none
• Candidate layer coverage: backend
• Confirmed layer coverage: backend
• Missing confirmed layers: none
• Missing required layers (candidate-level): none

AI File Selection

• Source: deterministic
• Selection origin: final-decision
• Used fallback: no
• Duration: 4791 ms
• Effective task area: backend
• Asset mode: none
• Rejected model paths: none
• Evidence summary: user_confirmed=0, inventory_exact=0, graph_supported=3, model_proposed=0, ranked_candidate=0

Project Inventory

• Total files found: 261
• Files kept in inventory: 261
• Truncated: no

Investigation Trace

• Trigger: Existing implementation candidates require trace verification.

• Seeds: server/src/routes/taskPacks.ts, server/src/contextComposer/contextComposerService.ts, server/src/ollama/taskPackGenerationReliability.ts, apps/desktop/renderer/src/api/client.ts, apps/desktop/renderer/src/App.backup.txt, apps/desktop/renderer/src/components/modals/GeneratedTaskPackModal.tsx, apps/desktop/renderer/src/components/modals/GlobalSearchModal.tsx, apps/desktop/renderer/src/components/ui/Modal.tsx, apps/desktop/renderer/src/hooks/useDashboardController.ts, apps/desktop/renderer/src/i18n/index.ts, apps/desktop/renderer/src/pages/DashboardHomePage.tsx, apps/desktop/renderer/src/pages/DashboardPage.tsx

• Inspected files: 16; edges followed: 64; hops: 1; duration: 506.6 ms; cache reused: no

• Confirmed owner candidates: none

• Probable owner candidates: server/src/routes/taskPacks.ts

• Reference/display candidates: server/src/contextComposer/contextComposerService.ts, server/src/ollama/taskPackGenerationReliability.ts, apps/desktop/renderer/src/api/client.ts, apps/desktop/renderer/src/App.backup.txt, apps/desktop/renderer/src/components/modals/GeneratedTaskPackModal.tsx, apps/desktop/renderer/src/components/modals/GlobalSearchModal.tsx, apps/desktop/renderer/src/components/ui/Modal.tsx, apps/desktop/renderer/src/hooks/useDashboardController.ts

• Unresolved trace points: Trace stopped at file limit.

• server/src/routes/taskPacks.ts: route-owner, probable; symbols=refinement, validation, taskpackgenerationreliability, boolean, generation, taskpack, taskpackgeneration, buildstabletaskpackrefinementcacheidentity, refinementcacheidentity, taskpackrefinement

• server/src/contextComposer/contextComposerService.ts: reference, reference; symbols=validation, boolean

• server/src/ollama/taskPackGenerationReliability.ts: reference, reference; symbols=refinement, validation, boolean, generation, taskpackrefinementpolicydiagnostics, refinementarraylimits, policyrefinementarraylimits, taskpackrefinementschema, taskpackrefinement, refinementitems, validationissuecodes, parsedrefinement

• apps/desktop/renderer/src/api/client.ts: reference, reference; symbols=boolean, apirequesterror, getgithubintegrationstatus, startgithubdeviceauth, pollgithubdeviceauth, signoutgithub, getprojects, addproject, rescanproject, getprojectgithubrepositorylink, initializeprojectgitrepository, setprojectgithuboriginremote, linkprojectgithubrepository, getagentspreview, saveagentsfile, getstorageaudit

• apps/desktop/renderer/src/App.backup.txt: consumer-display, reference; symbols=project, taskpack, getscorelabel, loadprojects, loadtaskpacks, handleselectproject, handlerescanproject, handlegenerateagentspreview, handlesaveagentsfile, handlecreatetaskpack, selectedpath, readinessscore, isexpanded, classname, onclick, disabled

• apps/desktop/renderer/src/components/modals/GeneratedTaskPackModal.tsx: consumer-display, reference; symbols=generatedtaskpackmodal, generatedtaskpackmodalprops, promptviewmode, formatduration, formatdate, gettaskpackbodylabel, gettaskpackbodydescription, infotile, handlecopyprompt, viewswitchtransition, markdownpreviewstyles, generatedprompt, bodylabel, bodydescription, classname, eyebrow

• apps/desktop/renderer/src/components/modals/GlobalSearchModal.tsx: consumer-display, reference; symbols=globalsearchmodal, lucideicon, apppageid, globalsearchmodalprops, searchitem, normalize, getprojectstack, getresulticon, getresultkind, handlesubmitfirstresult, handlecopyworkspaceresult, inputref, normalizedquery, timeoutid, searchitems, visiblelocalresults

• apps/desktop/renderer/src/components/ui/Modal.tsx: consumer-display, reference; symbols=boolean, reactnode, modalprops, requestclose, handlekeydown, modalexitms, maxwidth, scrollable, classname, onclick, isvisible, isclosing, timeout, eyebrow, children, onclose

• apps/desktop/renderer/src/hooks/useDashboardController.ts: reference, reference; symbols=boolean, usedashboardcontroller, parsemultilinerules, getclarificationsignature, getblockedcontextmessage, settaskpackdraft, loadprojects, loadtaskpacks, refreshdashboard, handleselectproject, handlerescanproject, handlegenerateagentspreview, handleopenprojectcontextfile, handleregenerateagentspreview, handlesaveagentsfile, generatetaskpackfromdraft

• apps/desktop/renderer/src/i18n/index.ts: reference, reference; symbols=applanguage, resolveapplanguage, applyapplanguage, preferences, language, resources, resolvedlanguage, checkspassed, tokengeneration, translation, addproject, scanning, refresh, refreshing, savechanges, unsaved

• imports: server/src/routes/taskPacks.ts -> server/src/storage/index.ts

• imports: server/src/routes/taskPacks.ts -> server/src/settings/settingsService.ts

• imports: server/src/routes/taskPacks.ts -> server/src/rules/rulesService.ts

• imports: server/src/routes/taskPacks.ts -> server/src/ollama/taskIntentAnalyzer.ts

• imports: server/src/routes/taskPacks.ts -> server/src/ollama/taskPackGenerationReliability.ts

• imports: server/src/routes/taskPacks.ts -> server/src/ollama/taskFileSelector.ts

• imports: server/src/routes/taskPacks.ts -> server/src/selection/contextQuality.ts

• imports: server/src/routes/taskPacks.ts -> server/src/selection/safetyPolicy.ts

• imported_by: server/src/index.ts -> server/src/routes/taskPacks.ts

• imported_by: server/src/ollama/taskPackGenerationReliability.smoke.ts -> server/src/routes/taskPacks.ts

Notes

• Project inventory was collected by ContextForge before selecting files.
• Files were selected from real inventory paths and validated before being added to this Task Pack.
• Protected context sections were generated by the backend and restored after local AI generation.
• Task intent source: ollama; area: backend; confidence: 0.95.
• Structured intent: 0 primary target(s); edit scope target_with_supporting_context.
• Task understanding: ready; action update; can proceed yes.
• Effective task area: backend.
• Asset mode: none.
• File selection source: deterministic; selection origin: final-decision; selected files: 3.
• Context quality: ready; score: 100/100.
• Asset files were detected and kept in inventory for asset-related tasks.
• Source files were detected.
• Style files were detected.
• Config files were detected.
• Documentation files were detected.
• Inventory includes dynamic text hints extracted from real file names and readable file contents.
• Inventory includes generic technical file roles inferred from paths and framework conventions.
• Selector engine version: 2026-07-16.canonical-final-decision-v1.
• Selector safety profile: canonical-selection-gate-v1.
• Final selection was rebuilt from API contract evidence: server/src/ollama/taskPackGenerationReliability.ts already produces cached, and server/src/routes/taskPacks.ts owns the API boundary.
• Reuse/expose operation proven: surface the existing producer value through the API boundary instead of adding a second semantic field to the refinement payload.
• Existing public contract retained as reference: apps/desktop/renderer/src/types/index.ts (generationcached).
• Execution mode: implementation.
• Confirmed 1 implementation target(s) from current user/code evidence.
• No unconfirmed target proposal was retained.
• Required technical layers: backend.
• No unresolved execution decision was retained.
• Selector prompt shortlist included 24 of 261 real inventory files and omitted 237 lower-ranked candidates.
• No task type conflict detected.
• No semantically weak model-selected paths were accepted.
• All selected paths were validated against project inventory and semantic quality gates.

---

Agent Instructions

Use a concise implementation plan, make focused edits, and keep the final response review-friendly.

Handle the task with the smallest practical set of changes. Avoid unrelated restructuring.

Before editing:

• Inspect the relevant file candidates and snippets provided by ContextForge.
• Treat inspect-only files as reference context.
• Treat asset-reference files as binary/reference context unless the task explicitly requires asset changes.
• Use only real files from the project.
• Keep the implementation focused on the user's actual task.

AI-refined implementation guidance

• In server/src/routes/taskPacks.ts, modify the API handler that generates the Task Pack response to expose this new generationCached boolean field in the public API contract.
• Ensure that the client-side type definition in apps/desktop/renderer/src/types/index.ts is updated (if necessary, though it's marked inspect-only) to reflect the presence of the generationCached field for consumers.

Constraints

• Do not invent project files: Do not invent files, folders, APIs, scripts, dependencies, environment variables, or implementation details that are not present in the provided context.
• Inspect before editing: Inspect the selected files and snippets before editing. If snippets are partial or truncated, inspect the full file first.
• Focused review scope: Keep changes focused and reviewable. Do not perform unrelated rewrites, formatting sweeps, dependency changes, or architecture changes.
• Respect inspect-only files: Files marked as inspect-only are reference context. Do not edit them unless the user explicitly changes the task scope.
• Asset reference safety: Files marked as asset-reference or non-text context should be treated as binary/reference context unless the task explicitly requests asset replacement or asset editing.
• No fake verification: Do not claim tests, builds, or checks were run unless they were actually run. If verification was not run, state what should be checked.

Task-specific safeguards

• The new boolean field must be sourced from the existing cache status logic within server/src/ollama/taskPackGenerationReliability.ts.
• Only modify the API boundary (server/src/routes/taskPacks.ts) to expose the cached status, avoiding duplication of the source value.

Known AI-Readiness Issues

• A test script exists, but no test files or test config were detected in the scanned project paths.

Acceptance Criteria

• The requested task is implemented using real project files only.
• Changes are focused and do not introduce unrelated behavior.
• The final response explains what changed and how to verify it.

Task-specific acceptance checks

• The Task Pack generation payload in server/src/ollama/taskPackGenerationReliability.ts must include a boolean field indicating cache usage.
• The API endpoint defined in server/src/routes/taskPacks.ts must successfully return the new generationCached boolean field when generating a Task Pack.
• Client-side consumers relying on apps/desktop/renderer/src/types/index.ts should be able to access the generationCached field without type errors.

Verification

• Run the existing build script if the change can affect production output.
• If a command cannot be run, explain why and provide manual verification steps.

Suggested verification

• Run npm run build and verify that the compilation succeeds after adding the new field.
• Manually test the Task Pack generation API endpoint, ensuring the response payload includes a boolean value for generationCached (both true and false cases).

Expected Final Response

Return a concise final response with:

• Files changed
• What was changed
• Verification performed
• Any risks, limitations, or manual checks still needed

---

Additional response requirements

• List all files modified.
• Describe how the cache status is now exposed via the API.
• Confirm that the new field type is correctly reported in the response schema.
• Report any manual checks required for client integration.

ContextForge Rules & Criteria

This section is generated and validated by ContextForge. Preserve it exactly.

Selected Template

• ID: template.codex.general
• Name: Codex · general
• Target tool: codex
• Task type: general
• Built-in: yes

Selected Rule Profile

• ID: profile.safe-general
• Name: Safe general task
• Task type: general
• Built-in: yes

Enabled Toggle Rules

• Do not invent project files (general): Do not invent files, folders, APIs, scripts, dependencies, environment variables, or implementation details that are not present in the provided context.
• Inspect before editing (general): Inspect the selected files and snippets before editing. If snippets are partial or truncated, inspect the full file first.
• Focused review scope (general): Keep changes focused and reviewable. Do not perform unrelated rewrites, formatting sweeps, dependency changes, or architecture changes.
• Respect inspect-only files (general): Files marked as inspect-only are reference context. Do not edit them unless the user explicitly changes the task scope.
• Asset reference safety (assets): Files marked as asset-reference or non-text context should be treated as binary/reference context unless the task explicitly requests asset replacement or asset editing.
• No fake verification (verification): Do not claim tests, builds, or checks were run unless they were actually run. If verification was not run, state what should be checked.

Custom User Rules

• No custom user rules provided.

Acceptance Criteria Preset

• ID: criteria.general-done
• Name: General done
• Task type: general

Final Acceptance Criteria

• The requested task is implemented using real project files only.
• Changes are focused and do not introduce unrelated behavior.
• The final response explains what changed and how to verify it.

---

Created from ContextForge. Project source files stay local; this issue contains only the generated task brief.