-
Notifications
You must be signed in to change notification settings - Fork 4
Limits and Glossary
Language: English · 简体中文
LWC bounds inputs, retrieval, graph traversal, and Agent context so one command cannot accidentally load an entire dense knowledge space. These values describe the current release line; the installed command's --help and returned request_limits remain authoritative.
| Surface | Default | Hard boundary |
|---|---|---|
| Source or Page file input | — | 64 MiB |
| Directory or manifest import | — | 100,000 documents |
search --limit |
20 | 1–1,000 |
context --limit |
50 | 1–1,000 per Store |
lint --limit |
100 | 1–1,000 |
log --limit |
20 | 1–1,000 |
| Source, Page, and ingest list limit | 100 | 1–1,000 |
source diff input |
— | 8 MiB and 200,000 lines per side |
source diff --max-chars |
20,000 | 1–100,000 |
| Changeset or checkpoint name | — | One safe filename segment, at most 80 bytes |
| SQLite busy boundary | — | 10 seconds |
Pagination offsets are zero-based and must fit a signed 64-bit integer. A response with has_more=true is an instruction to request the next page, not to raise the limit.
| Surface | Default | Hard boundary |
|---|---|---|
| Passage span | — | 64 KiB |
| Sentence span | — | 16 KiB |
span expand --before / --after
|
1 / 1 | 0–20 each |
span expand --children |
20 | 1–200 |
Oversized logical sentences or passages are split at valid UTF-8 boundaries. Byte bounds protect indexing; CLI window offsets remain Unicode-character based where documented.
| Surface | Default | Hard boundary |
|---|---|---|
| Tag name | — | 1–128 characters, no control characters |
| Membership or policy reason | — | 1–512 characters, no control characters |
load tag --limit |
10 | 1–100 Pages |
| Autoload Page limit | 10 | 1–100 Pages |
| Autoload character budget | 50,000 | 1–100,000 characters |
| Hook autoload policies considered | — | 100 policies |
The character budget applies across returned complete Page bodies. When the budget is exhausted, LWC reports omission diagnostics rather than truncating a Page into misleading partial rules.
| Command | Default | Hard boundary |
|---|---|---|
graph related --limit |
20 | 1–1,000 |
graph explore --depth |
2 | 1–10 |
graph explore --limit |
100 | 1–5,000 nodes |
graph neighbors --limit |
100 | 1–5,000 nodes |
graph path --max-depth |
6 | 1–10 |
graph path --limit |
200 | 1–5,000 nodes |
graph impact --max-depth |
4 | 1–10 |
graph impact --limit |
100 | 1–5,000 nodes |
graph overview --limit |
10 | 1–1,000 |
graph relation list --limit |
100 | 1–1,000 |
Traversal limits bound returned work, not graph size. Use filters and smaller starting nodes instead of repeatedly requesting the maximum.
| Budget | Hard boundary |
|---|---|
| Searchable query terms | 8 |
| Candidate documents per page | 25 |
| Terms per response | 30 |
| Nodes per response | 200 |
| Edges per response | 500 |
| Inspected text | 4 MiB |
| Spans inspected per document | 4 |
| Metadata inspected per document | 16 KiB |
| Pagination offset | 10,000 |
The Viewer clamps document and term page sizes to their caps and returns request_limits plus truncation diagnostics. Empty queries and queries with more than eight searchable terms are rejected. This is the central performance contract: Word Graph samples a bounded neighborhood and never materializes every token edge.
| Surface | Hard boundary |
|---|---|
| Viewer Page or Source metadata | 1,000 items |
| Viewer knowledge-graph nodes / edges | 1,000 / 5,000 |
| CodeGraph overview nodes / edges | 1,000 / 5,000 |
| CodeGraph prompt-Hook output | 20 KiB within 2 seconds |
| CodeGraph runtime install-lock wait | 30 seconds |
Viewer bounds are read-only presentation limits. They do not cap canonical Store size. CodeGraph query commands apply their own upstream result budgets in addition to LWC's bridge bounds.
| Surface | Default | Hard boundary |
|---|---|---|
| Conversion timeout | 120 seconds | 1–900 seconds |
| Input file | — | 64 MiB |
| Markdown output | — | 64 MiB and valid UTF-8 |
| Captured error preview | — | 8 KiB |
LWC creates a new output path and never overwrites an existing file. Increasing timeout does not relax path, argument, secret, or size validation.
| Surface | Default | Hard boundary |
|---|---|---|
| Hook input envelope | — | 64 KiB |
| Hook rendered context | — | 100,000 characters |
| MCP JSON-RPC frame | — | 64 KiB |
lwc_explore.query |
— | 1–10,000 characters |
projectPath |
— | 4,096 characters and an absolute project directory |
maxDocuments |
8 | 1–20 |
maxFiles |
12 | 1–20 |
| CodeGraph MCP request | — | 10 seconds |
MCP budgets protect the Agent transport. They do not authorize writes, downloads, graph initialization, root filesystem access, or relative project paths.
Graph and maintenance Work emits a heartbeat every 5 seconds. A running state becomes eligible for resume recovery after 30 seconds without an update. These values prevent two live runners from treating the same short pause as abandonment; they are not task-duration limits.
| Term | Meaning |
|---|---|
| Agent | The AI coding or research host that invokes LWC through CLI, Hook, Skill, Instructions, or MCP |
| AgentTarget | One handwritten adapter for an Agent host's official paths, formats, and capability gaps |
| canonical | Authoritative state; for Wiki knowledge this is SQLite, not generated Markdown or a graph sidecar |
| changeset | An isolated sparse draft that publishes several supported mutations atomically |
| checkpoint | A complete validated SQLite recovery point for one Store |
| CodeGraph | The project-local structural code index for symbols, calls, dependencies, and impact |
| document knowledge graph | The physical Page, Source, citation, link, provenance, and semantic-relation projection; also called the memory graph |
| derived state | Rebuildable material such as Markdown, FTS, spans, graph sidecars, and indexes |
| FTS | SQLite full-text search indexes used for deterministic lexical retrieval |
| global scope | User-level reusable memory stored under the current user's LWC home |
| Hook | A bounded lifecycle adapter that reports readiness and selected strong context to an Agent host |
| ingest | The persistent Source integration state machine from collection through analysis and Page updates |
| Instructions | Marker-bounded durable guidance installed in the Agent host's official instruction surface |
| lint | Deterministic structural validation of canonical Wiki and search-index invariants |
| MCP | Model Context Protocol transport exposing the single read-only lwc_explore tool |
| operation | A durable audit record for a successful state transition or explicitly recorded read |
| Page | Maintained, Agent-written knowledge with a stable slug, body, links, citations, and provenance |
| projection | A canonical document rendered into a rebuildable secondary representation |
| project scope | The nearest authorized ancestor project Wiki discovered from the current directory |
| provenance | Evidence classification explaining where a Page claim comes from |
| receipt | Agent integration ownership record used by status, refresh, and precise uninstall |
| Source | An immutable snapshot of authorized evidence collected for later integration |
| span | An indexed passage or sentence with a stable locator into a Source or Page revision |
| strong context | A deliberately small set of complete Pages loaded by tag policy rather than search ranking |
| tag | A curated membership relation over core Pages, optionally with an autoload policy |
| trans | Optional MarkItDown or AnyDoc subprocess adaptation that creates Markdown from a document |
| Viewer | Loopback-only foreground UI that reads current state without refreshing or mutating it |
| Wiki | One project or global body of canonical Sources, Pages, governance, indexes, and history |
| Word Graph | A bounded on-demand graph connecting documents through shared searchable terms |
| Work | A durable background job with progress, cancellation, resume, result, and error state |
For practical use, continue with Common workflows and recipes. For exact command families, see CLI and configuration reference.
LWC Wiki
- Home · 首页
- Project overview · 项目简介
- Basic concepts · 基本概念
- Use cases · 应用场景
- Installation and upgrades · 安装与升级
- Quick start · 快速开始
- Persistent memory · 持久记忆体系
- Agent workflow and memory policy · Agent 工作流与主动记忆策略
- Sources and ingestion · 来源与知识整合
- Wiki pages and provenance · Wiki 页面与来源证明
- Search and context · 搜索与上下文载入
- Document knowledge graph · 文档知识图(记忆图网)
- Code graph · 代码图
- Word graph · 词图
- Tags and strong context · 标签与强上下文
- Document conversion · 文档转换
-
MCP server and
lwc_explore· MCP 服务与lwc_explore - Skills, Hooks, and Instructions · Skills、Hooks 与 Instructions
- AgentTarget installation · AgentTarget 安装与集成
- Changesets · Changeset 原子变更
- Work system · Work 任务系统
- Checkpoints and rollback · Checkpoint、恢复与回滚
- Read-only Viewer · 只读可视化界面
- Architecture overview · 总体架构
- Storage and data model · 存储与数据模型
- Retrieval and indexing · 检索与索引设计
- Graph projection and performance · 图投影与性能设计
- MCP, Hooks, and AgentTarget design · MCP、Hook 与 AgentTarget 设计
- Safety and trust boundaries · 安全模型与信任边界
- Maintenance and diagnostics · 维护与诊断
- Troubleshooting and FAQ · 故障排查与常见问题
- Migration and compatibility · 迁移与版本兼容
- Support and issue reporting · 获取帮助与问题反馈
- CLI and configuration reference · CLI 与配置参考
- JSON output and error contract · JSON 输出与错误契约
- Limits and glossary · 系统限制与术语表
- Contributing and development · 贡献与开发指南
- Testing and release process · 测试与发布流程
- Wiki style guide · Wiki 编写规范