一键导入
init-sync
Scan codebase and fill/update CLAUDE.md and AGENTS.md with real project data. Produces deep architectural documentation matching production quality.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Scan codebase and fill/update CLAUDE.md and AGENTS.md with real project data. Produces deep architectural documentation matching production quality.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
Deep research pipeline for thorough investigation of any topic. Multi-agent system with Scout, Investigators, Deep Diver, Verifier, Synthesizer, and iterative Critic review. Produces verified, sourced research reports.
Use for complex features needing bulletproof plans. Automated 10-actor verification pipeline (Researcher → Planner → [Skeptic + TDD + Completeness + Integration + Security* + Performance*] → Critic) that iterates up to 10 times until the Critic approves (≥75%), producing a one-shot-ready plan.
Execute bd tasks with Agent Teams. Lead → Dev → 3 Reviewers (Bug + Security + Compliance) → Validate → QA.
Full website audit — UX/UI design, marketing, SEO, accessibility, performance, conversion. Crawls the site (up to 100 pages) via cmux browser, generates detailed visual critique with actionable recommendations.
Deep system testing — auto-detects stack, discovers components, runs unit/integration/E2E/security/performance tests, reports bugs to bd with full reproduction steps.
Generates a structured planning prompt for bishx-plan. Analyzes the user's idea, discovers relevant skills, and builds a detailed prompt for execution.
基于 SOC 职业分类
| name | init-sync |
| description | Scan codebase and fill/update CLAUDE.md and AGENTS.md with real project data. Produces deep architectural documentation matching production quality. |
You scan the current project and fill CLAUDE.md and AGENTS.md with real data from the codebase. This works both on fresh templates (with <!-- bishx:init:xxx --> markers) and on previously filled files (updates existing content).
The output quality must match production-grade project documentation — deep, specific, with exact file paths, counts, and lists.
Core principle: Document what ACTUALLY exists. Don't force the project into a predefined template. If the project has Controllers — write Controllers, not Routes. If it has Repositories — write Repositories, not DAOs. Adapt section names, structure, and depth to the real architecture. The examples below are for quality reference, not for copying verbatim.
CLAUDE.md must exist in project root. If not — tell user to run /bishx:init first.AGENTS.md must exist in project root. If not — tell user to run /bishx:init first.Search for project manifests to identify the stack:
package.json, pyproject.toml, requirements.txt, go.mod, Cargo.toml,
composer.json, Gemfile, pom.xml, build.gradle, mix.exs, pubspec.yaml,
tsconfig.json, vite.config.*, next.config.*, nuxt.config.*,
docker-compose.yml, Dockerfile, Makefile
Read each found manifest. Extract: language, framework, ALL dependencies with versions.
This is NOT a surface scan. You must understand the codebase deeply.
Run ls (depth 2-3) on the project root. Map every significant directory.
Find all model definitions:
class.*Base, SQLAlchemy models, Pydantic models, dataclassesFor each: note file path, count, list key model names.
Find all route/endpoint definitions:
@router.get/post/put/delete, APIRouterrouter.get/post, app.get/postmux.HandleFunc, gin.GET/POSTapp/ or pages/ directory structureFor each: note file paths, count, list key endpoints, identify namespaces (admin, public, internal).
Find service layer files. Note: location, count, key services.
Find data access layer. Note: base class (if any), file paths, count, examples.
Find auth mechanisms: middleware, decorators, guards, permission checkers. Note: auth method (JWT/JWE/sessions/OAuth), permission model, key files.
Find standardized response wrappers (envelope, DTO, serializers).
Find all middleware. List each with one-line description.
Find background tasks, cron jobs, queue consumers. Note mechanism and examples.
Find config management approach (env, runtime config, feature flags).
Find existing docs (docs/, README.md, API specs, architecture docs).
Identify domain-specific terms from model names, comments, README.
Read CLAUDE.md. Replace placeholder sections AND add new sections if they don't exist.
Replace <!-- bishx:init:project_description --> with 1-2 sentence description.
Source: README.md, package.json description, pyproject.toml metadata.
Replace <!-- bishx:init:stack --> with compact bullet list:
- Backend: Python 3.13, FastAPI, SQLAlchemy async, Alembic, Pydantic v2
- Frontend: React 19, TypeScript, Vite, Tabler (HTML→React conversion)
- DB: PostgreSQL, Redis
- Auth: LDAP/AD → JWE tokens
Only include categories that actually exist. Be specific with versions.
Insert AFTER Stack, BEFORE Security. Content:
Example:
## Environment
- venv: `.venv` in project root, activate via `python3.13`
- Full codebase context: `docs/CODEBASE_CONTEXT.md`
- Architecture details: `docs/plans/v1/implementation-plan.md`
Insert AFTER Environment, BEFORE Security. Brief (2-3 sentences):
Example:
## Backend Architecture (`backend/`)
Base backend architecture in `backend/app/`. New modules follow these patterns. Architecture can evolve if justified — but don't break the foundation (DAO, DTO, ResponseEnvelope, async everywhere, session injection)
If current content is only generic OWASP rules, enhance with project-specific security concerns detected from the codebase (encryption, compliance requirements, network restrictions, etc.).
Insert AFTER Security. Brief notes:
Example:
## Frontend
- Tabler HTML → HTML→React conversion, no Tailwind
- Avoid generic AI aesthetics. Follow the framework's style.
This is the DEEP documentation file. It must be thorough enough that a new agent can understand the entire codebase without reading source files.
Replace {PROJECT_NAME} with actual project name.
Replace <!-- bishx:init:project_description --> with 2-3 sentence description.
Replace <!-- bishx:init:stack_detailed -->.
Create subsections per layer with **bold labels** and exact versions:
### Backend
- **Runtime:** Python 3.13
- **Framework:** FastAPI
- **ORM:** SQLAlchemy 2.x async
- **Migrations:** Alembic
- **Validation:** Pydantic v2
- **Cache/Messaging:** Redis 7+
- **DB:** PostgreSQL 16+
### Frontend
- **Runtime:** React 19
- **Language:** TypeScript 5
- **Bundler:** Vite 6
- **UI Framework:** Tabler (HTML→React conversion, no Tailwind)
- **Router:** React Router v7
- **State:** Zustand
- **Data Fetching:** TanStack React Query v5
- **HTTP Client:** Axios
### Auth
- **Source:** LDAP/AD bind
- **Tokens:** JWE (access 15min / refresh 30d HttpOnly)
Include ALL layers: Backend, Frontend, ML/AI, Auth, Monitoring, Infrastructure — whatever exists.
Replace <!-- bishx:init:project_structure --> with FULL architectural breakdown.
This is NOT just a directory tree. Create subsections for EACH architectural layer found:
Discover the actual layers the project uses. Common ones: Models, DAO/Repository, Schemas/DTOs, Routes/Controllers, Services, Auth, Response Format, Config, Middleware, Workers, Scheduler — but only document what's actually there. Skip what doesn't exist. Add layers not listed here if the project has them.
For each layer found, include: Location, Count (if countable), key names/examples.
Example for one subsection:
### Models
- **Location:** `backend/app/main_dao/models.py`
- **Count:** ~20 models in one file
- **Key models:** User, Role, UserRole, Conference, Transcript, AudioFile, Tag, Notification, AuditLog
### DAO Layer
- **BaseDAO[T]:** `backend/app/main_dao/base.py` — Generic CRUD with cursor pagination
- **DAO files:** `backend/app/dao/` — 24 files, inherit BaseDAO
- **Examples:** UserDAO, ConferenceDAO, TranscriptDAO, TagDAO, NotificationDAO
Discover the actual frontend structure. Common layers: Features/Pages, UI Components, Layout, State Management, Hooks/Composables, Routes — but adapt to what's actually there (Vue has composables, not hooks; Angular has modules; etc.).
For each layer found, include: Location, Count, lists.
Routes — prefer table format:
| Path | Component | Feature |
|------|-----------|---------|
| `/` | HomePage | home |
| `/login` | LoginPage | auth |
Replace with table of ALL found documentation files:
| Document | Description |
|----------|-------------|
| [README.md](README.md) | Navigation hub |
| [docs/architecture.md](docs/architecture.md) | Services architecture |
Extract domain-specific terms from models, README, comments:
| Term | Meaning |
|------|---------|
| Conference | TrueConf video call (2-60 min) |
If no clear domain terms found — omit this section entirely.
Replace <!-- bishx:init:patterns -->.
Per-layer list of ACTUAL patterns found in code (not recommendations):
### Backend
- DAO pattern (not Repository, not Active Record)
- Async everywhere (asyncpg, aiohttp, aioredis)
- Pydantic v2 schemas for validation
- Cursor-based pagination (not offset)
- Soft delete (deleted_at, deleted_by)
- ResponseEnvelope for all API responses
### Frontend
- Feature-based structure (not by file type)
- Zustand for state management
- React Query for server state
Be specific — mention what pattern IS used and what it's NOT (like the voicer example).
[bishx:init-sync] Done
- CLAUDE.md: N sections filled, M sections added
- AGENTS.md: N sections filled, M sections added
- Stack: {one-line stack summary}
- Architecture: {count} layers documented
- Patterns: {count} patterns identified
app/models.py". Don't write "some routes" — write "20+ files in api/v1/routes/".