@vpxa/kb 0.1.13 → 0.1.16

package/packages/chunker/dist/treesitter-chunker.js

@@ -1,8 +0,0 @@
-import{extname as m}from"node:path";import{CHUNK_SIZES as k}from"../../core/dist/index.js";import{CodeChunker as y}from"./code-chunker.js";const f={".ts":{package:"tree-sitter-typescript",subLanguage:"typescript"},".tsx":{package:"tree-sitter-typescript",subLanguage:"tsx"},".mts":{package:"tree-sitter-typescript",subLanguage:"typescript"},".cts":{package:"tree-sitter-typescript",subLanguage:"typescript"},".js":{package:"tree-sitter-javascript"},".jsx":{package:"tree-sitter-javascript"},".mjs":{package:"tree-sitter-javascript"},".cjs":{package:"tree-sitter-javascript"},".py":{package:"tree-sitter-python"},".go":{package:"tree-sitter-go"},".rs":{package:"tree-sitter-rust"},".java":{package:"tree-sitter-java"}},x=new Set(["function_declaration","class_declaration","interface_declaration","type_alias_declaration","enum_declaration","lexical_declaration","export_statement","variable_declaration","abstract_class_declaration","function_definition","class_definition","decorated_definition","function_declaration","method_declaration","type_declaration","function_item","impl_item","struct_item","enum_item","trait_item","mod_item","type_item","class_declaration","interface_declaration","method_declaration","enum_declaration"]);class u{static instance=null;parser=null;languages=new Map;static async initialize(){if(u.instance)return u.instance;const t=new u;return await t.init()?(u.instance=t,t):null}static get(){return u.instance}async init(){try{const t=await import("tree-sitter"),r=t.default??t;this.parser=new r;for(const[c,s]of Object.entries(f))try{const o=await import(s.package),e=o.default??o,n=s.subLanguage?e[s.subLanguage]:e;n&&this.languages.set(c,n)}catch{}const i=this.languages.size>0;return i&&console.error(`[KB] Tree-sitter initialized with ${this.languages.size} languages: ${[...this.languages.keys()].join(", ")}`),i}catch{return!1}}parse(t,r){const i=this.languages.get(r);return i?(this.parser.setLanguage(i),this.parser.parse(t)):null}hasLanguage(t){return this.languages.has(t)}}async function w(){return await u.initialize()!==null}class L{maxChunkSize;runtime;regexFallback;constructor(t,r){this.runtime=t,this.maxChunkSize=r?.maxChunkSize??k.code.max,this.regexFallback=new y(r)}chunk(t,r){const i=m(r.sourcePath).toLowerCase(),c=this.runtime.parse(t,i);if(!c)return this.regexFallback.chunk(t,r);try{return this.chunkFromAst(c,t,r)}catch{return this.regexFallback.chunk(t,r)}}chunkFromAst(t,r,i){const c=t.rootNode,s=[],o=[];for(let e=0;e<c.childCount;e++){const n=c.child(e);if(n&&x.has(n.type)){let h=n.startPosition?n.startIndex:n.startByte??0,l=n.startPosition?.row??0,a=n.previousSibling;for(;a&&(a.type==="comment"||a.type==="line_comment"||a.type==="block_comment"||a.type==="decorator");)h=a.startIndex??a.startByte??h,l=a.startPosition?.row??l,a=a.previousSibling;o.push({startByte:h,endByte:n.endIndex??n.endByte??r.length,startRow:l,endRow:n.endPosition?.row??r.split(`
-`).length-1})}}if(o.length===0)return this.regexFallback.chunk(r,i);if(o[0].startByte>0){const e=r.slice(0,o[0].startByte).trim();e.length>0&&s.push({text:`// File: ${i.sourcePath}
-${e}`,sourcePath:i.sourcePath,contentType:i.contentType,chunkIndex:0,totalChunks:0,startLine:1,endLine:o[0].startRow+1})}for(const e of o){let n=r.slice(e.startByte,e.endByte).trim();if(n=`// File: ${i.sourcePath}
-`+n,n.length>this.maxChunkSize){const l=this.splitByLines(n,this.maxChunkSize);let a=e.startRow+1;for(const p of l){const g=p.split(`
-`).length;s.push({text:p,sourcePath:i.sourcePath,contentType:i.contentType,chunkIndex:s.length,totalChunks:0,startLine:a,endLine:a+g-1}),a+=g}}else s.push({text:n,sourcePath:i.sourcePath,contentType:i.contentType,chunkIndex:s.length,totalChunks:0,startLine:e.startRow+1,endLine:e.endRow+1})}return s.map((e,n)=>({...e,chunkIndex:n,totalChunks:s.length}))}splitByLines(t,r){const i=t.split(`
-`),c=[];let s=[],o=0;for(const e of i)o+e.length+1>r&&s.length>0?(c.push(s.join(`
-`)),s=[e],o=e.length):(s.push(e),o+=e.length+1);return s.length>0&&c.push(s.join(`
-`)),c}}export{L as TreeSitterChunker,u as TreeSitterRuntime,w as initializeTreeSitter};

package/packages/cli/dist/commands/init.d.ts

@@ -1,10 +0,0 @@
-/**
- * `kb init` — scaffold a knowledge base in the current directory.
- * Creates kb.config.json, .gitignore entry, optional IDE config,
- * agent instruction files (.github/copilot-instructions.md, AGENTS.md),
- * and skills/knowledge-base/SKILL.md.
- */
-export declare function initProject(options: {
-    force: boolean;
-}): Promise<void>;
-//# sourceMappingURL=init.d.ts.map

package/packages/cli/dist/commands/init.js

@@ -1,308 +0,0 @@
-import{appendFileSync as h,copyFileSync as y,existsSync as t,mkdirSync as n,readFileSync as v,writeFileSync as r}from"node:fs";import{basename as w,dirname as S,resolve as e}from"node:path";import{fileURLToPath as x}from"node:url";const C={sources:[{path:".",excludePatterns:["**/node_modules/**","**/dist/**","**/build/**","**/.git/**","**/.kb-data/**","**/coverage/**","**/*.min.js","**/package-lock.json","**/pnpm-lock.yaml"]}],indexing:{chunkSize:1500,chunkOverlap:200,minChunkSize:100},embedding:{model:"mixedbread-ai/mxbai-embed-large-v1",dimensions:1024},store:{backend:"lancedb",path:".kb-data/lance"},curated:{path:"curated"}};function A(o){return{servers:{[`${o}-kb`]:{type:"stdio",command:"npx",args:["@vpxa/kb","serve"]}}}}function T(o){return`# ${o} \u2014 Copilot Instructions
-
-This project uses **@vpxa/kb** as an MCP knowledge base server. The KB provides search, context compression, persistent memory, and validation tools that **MUST** be used to maintain quality and performance.
-
-## Mandatory Rules
-
-### 1. ALWAYS search before acting
-Before writing or modifying code, you MUST recall prior decisions and understand the affected area:
-\`\`\`
-kb_search({ query: "<task keywords>" })
-kb_scope_map({ task: "<what you are doing>" })
-\`\`\`
-
-### 2. ALWAYS compress context before reading large files
-Do NOT read entire files into context. Use these tools to reduce tokens and focus on what matters:
-
-| Situation | Tool | Why |
-|-----------|------|-----|
-| Need relevant sections of a large file | \`kb_compact\` | Extracts only semantically relevant chunks \u2014 5-20x reduction |
-| Need to summarize multiple sources | \`kb_digest\` | Compresses multiple text blocks into a token-budgeted summary |
-| Need a quick structural overview of a file | \`kb_file_summary\` | Returns exports, imports, functions without reading the full file |
-| Need a reusable context card | \`kb_stratum_card\` | Generates T1/T2 cards \u2014 10-100x token reduction |
-
-**Example \u2014 reading a 400-line file:**
-\`\`\`
-# BAD: read_file entire file (wastes ~3000 tokens)
-# GOOD: get structure first, then compact only what you need
-kb_file_summary({ path: "src/http-client.ts" })
-kb_compact({ path: "src/http-client.ts", query: "retry logic" })
-\`\`\`
-
-### 3. ALWAYS use persistent memory
-- **Start of session**: Read prior knowledge with \`kb_list()\` and \`kb_search({ query: "SESSION CHECKPOINT", origin: "curated" })\`
-- **During session**: Use \`kb_stash\` for temporary working data, \`kb_checkpoint\` to save progress
-- **End of session**: MUST call \`kb_remember\` to persist decisions, patterns, and lessons learned
-- **Before proposing a new approach**: MUST \`kb_search\` to check if a decision already exists
-
-| Tool | When | Persistence |
-|------|------|-------------|
-| \`kb_stash\` | Temporary data within a session (intermediate results, scratch notes) | Session only |
-| \`kb_checkpoint\` | Save/restore session progress (resumable work) | Session only |
-| \`kb_remember\` | Architecture decisions, conventions, patterns, troubleshooting | Permanent (survives reindex) |
-
-### 4. ALWAYS validate before committing
-\`\`\`
-kb_check({})                    # Typecheck + lint (returns summary by default)
-kb_test_run({})                 # Run tests
-kb_blast_radius({ changed_files: ["..."] })  # Impact analysis
-kb_audit({})                    # Unified project audit with score and recommendations
-\`\`\`
-
-### 5. Use FORGE for complex tasks
-For tasks involving multiple files or architectural decisions:
-\`\`\`
-kb_forge_classify({ task: "<description>" })   # Classify complexity tier
-kb_forge_ground({ task: "<description>" })     # Full analysis: scope + constraints + evidence
-kb_evidence_map({ claims: [...] })             # Track verified vs assumed claims
-\`\`\`
-
-## Quick Workflow
-
-\`\`\`
-# 1. Orient
-kb_search({ query: "task keywords" })
-kb_scope_map({ task: "what you are doing" })
-
-# 2. Read efficiently (NEVER raw-read large files)
-kb_file_summary({ path: "relevant-file.ts" })
-kb_compact({ path: "relevant-file.ts", query: "specific concern" })
-
-# 3. Do the work
-
-# 4. Validate
-kb_check({})
-kb_test_run({})
-
-# 5. Persist
-kb_remember({ title: "What I learned", content: "...", category: "decisions" })
-\`\`\`
-
-## Context Budget Rules
-
-- **Prefer \`kb_file_summary\` over \`read_file\`** for understanding file structure
-- **Prefer \`kb_compact\` over full file reads** when you need specific sections (use \`path\` param to avoid read_file round-trip)
-- **Prefer \`kb_search\` over \`grep_search\`** for semantic/conceptual queries
-- **Prefer \`kb_symbol\` over manual grep** for finding definitions and references
-- **Use \`kb_digest\`** when combining information from 3+ sources
-- **Use \`kb_stratum_card\`** to create reusable compressed context for repeated reference
-`}function P(o){return`# ${o} \u2014 Agent Instructions
-
-## KB Knowledge Base
-
-This project has a **@vpxa/kb** MCP server providing search, analysis, memory, and developer tools.
-
-### Skills Reference
-
-| Context | Skill | Details |
-|---------|-------|---------|
-| KB search, analysis, memory | \`kb\` | See [skills/knowledge-base/SKILL.md](skills/knowledge-base/SKILL.md) or run \`kb_status({})\` |
-
-### Available Tool Categories
-
-| Category | Tools | Purpose |
-|----------|-------|---------|
-| Search & Discovery | \`kb_search\`, \`kb_find\`, \`kb_symbol\`, \`kb_trace\`, \`kb_scope_map\`, \`kb_lookup\`, \`kb_dead_symbols\`, \`kb_file_summary\` | Find code, symbols, data flow, reading plans |
-| Code Analysis | \`kb_analyze_structure\`, \`kb_analyze_dependencies\`, \`kb_analyze_symbols\`, \`kb_analyze_patterns\`, \`kb_analyze_entry_points\`, \`kb_analyze_diagram\`, \`kb_blast_radius\` | Structure, deps, patterns, impact, diagrams |
-| Knowledge | \`kb_remember\`, \`kb_read\`, \`kb_update\`, \`kb_forget\`, \`kb_list\`, \`kb_produce_knowledge\` | Persistent cross-session memory |
-| Execution | \`kb_check\`, \`kb_test_run\`, \`kb_eval\`, \`kb_batch\`, \`kb_audit\` | Typecheck, lint, test, run code, unified audit. \`check\` defaults to summary output (~300 tokens) |
-| Code Manipulation | \`kb_rename\`, \`kb_codemod\`, \`kb_diff_parse\`, \`kb_data_transform\` | Safe renames, transforms, diff parsing |
-| Context | \`kb_compact\`, \`kb_workset\`, \`kb_stash\`, \`kb_checkpoint\`, \`kb_parse_output\` | Manage working sets, save progress. \`compact\` accepts \`path\` for server-side file read |
-| FORGE | \`kb_forge_ground\`, \`kb_forge_classify\`, \`kb_evidence_map\`, \`kb_digest\`, \`kb_stratum_card\` | Quality gates, context compression |
-| Web & API | \`kb_web_fetch\`, \`kb_web_search\`, \`kb_http\` | Fetch pages, search web, test APIs |
-| Lanes | \`kb_lane\` | Isolated file copies for parallel exploration (create/list/status/diff/merge/discard) |
-| Git & Environment | \`kb_git_context\`, \`kb_process\`, \`kb_watch\`, \`kb_delegate\` | Git info, process management |
-| Utilities | \`kb_regex_test\`, \`kb_encode\`, \`kb_measure\`, \`kb_changelog\`, \`kb_schema_validate\`, \`kb_snippet\`, \`kb_env\`, \`kb_time\` | Regex, encoding, metrics, validation |
-| System | \`kb_status\`, \`kb_reindex\`, \`kb_health\`, \`kb_guide\`, \`kb_onboard\`, \`kb_graph\`, \`kb_queue\`, \`kb_replay\` | Index management, health checks, tool discovery, knowledge graph |
-
----
-
-## MANDATORY: Context Reduction Protocol
-
-**Every agent interaction MUST minimize context window usage.** Raw file reads waste tokens and degrade LLM output quality. Use these tools instead:
-
-### Decision Tree \u2014 How to Read Code
-
-\`\`\`
-Need to understand a file?
-\u251C\u2500 Just structure? \u2192 kb_file_summary (exports, imports, functions \u2014 ~50 tokens)
-\u251C\u2500 Specific section? \u2192 kb_compact({ path: "file.ts", query: "topic" }) \u2014 5-20x reduction
-\u251C\u2500 Multiple files? \u2192 kb_digest (multi-source compression \u2014 token-budgeted)
-\u251C\u2500 Repeated reference? \u2192 kb_stratum_card (T1/T2 card \u2014 10-100x reduction)
-\u2514\u2500 Full file needed? \u2192 ONLY as last resort, and compact it after reading
-\`\`\`
-
-### Rules
-1. **NEVER read a file >100 lines without compressing it first**
-2. **ALWAYS use \`kb_file_summary\` before \`read_file\`** \u2014 often the summary is sufficient
-3. **ALWAYS use \`kb_compact\` when you only need specific sections** of a file (use \`path\` param to read server-side)
-4. **Use \`kb_digest\` when synthesizing from 3+ sources** \u2014 don't accumulate raw text
-5. **Use \`kb_stratum_card\` for files you'll reference repeatedly** in a session
-
----
-
-## MANDATORY: Memory Protocol
-
-**Every session MUST read and write persistent memory.** Without this, every conversation starts from zero.
-
-### Session Start (MUST do ALL of these)
-\`\`\`
-kb_status({})                                              # Verify KB is ready
-kb_list()                                                  # See what knowledge exists
-kb_search({ query: "SESSION CHECKPOINT", origin: "curated" })  # Resume prior work
-\`\`\`
-
-### During Session
-| Situation | Action |
-|-----------|--------|
-| Found a useful intermediate result | \`kb_stash({ key: "name", value: "data" })\` |
-| Completed a milestone | \`kb_checkpoint({ action: "save", name: "milestone" })\` |
-| Made an architecture decision | \`kb_remember({ title: "...", category: "decisions" })\` |
-| Discovered a pattern or convention | \`kb_remember({ title: "...", category: "patterns" })\` |
-| Found a non-obvious solution | \`kb_remember({ title: "...", category: "troubleshooting" })\` |
-| About to propose a new approach | \`kb_search({ query: "..." })\` \u2014 check if decided before |
-
-### Session End (MUST do this)
-\`\`\`
-kb_remember({
-  title: "Session checkpoint: <topic>",
-  content: "<what was done, decisions made, blockers, next steps>",
-  category: "conventions"
-})
-\`\`\`
-
-### Memory Decision Tree
-\`\`\`
-Is this data temporary (scratch, intermediate)?
-\u251C\u2500 Yes \u2192 kb_stash (session-scoped key-value)
-\u2514\u2500 No \u2192 Is it resumable progress?
-   \u251C\u2500 Yes \u2192 kb_checkpoint (session-scoped snapshot)
-   \u2514\u2500 No \u2192 kb_remember (permanent, survives reindex)
-       Categories: decisions | patterns | conventions | troubleshooting
-\`\`\`
-
----
-
-## MANDATORY: Search-Before-Act Protocol
-
-**NEVER write or modify code without first searching for context.**
-
-\`\`\`
-# Before ANY code change:
-kb_search({ query: "<what you're about to change>" })     # Prior decisions?
-kb_scope_map({ task: "<description>" })                    # What files to read?
-kb_symbol({ name: "<key symbol>" })                        # Where is it defined/used?
-\`\`\`
-
-If \`kb_search\` returns a prior decision about the topic, you MUST follow it or explicitly explain why you're deviating.
-
----
-
-## MANDATORY: Validation Protocol
-
-**NEVER commit or present code without validation.**
-
-\`\`\`
-kb_check({})                                    # Typecheck + lint (tsc + biome)
-kb_test_run({})                                 # Run tests
-kb_blast_radius({ changed_files: ["..."] })     # Impact analysis
-kb_audit({})                                    # Unified project audit (structure, deps, patterns, health, dead symbols, entry points)
-\`\`\`
-
----
-
-## FORGE Protocol (for complex tasks)
-
-For tasks touching 3+ files or involving architectural decisions:
-
-\`\`\`
-kb_forge_classify({ task: "<description>" })        # Quick: Floor/Standard/Critical tier
-kb_forge_ground({ task: "<description>" })           # Full: scope + constraints + evidence
-kb_evidence_map({ claims: ["claim1", "claim2"] })   # Track verified vs assumed
-\`\`\`
-
----
-
-## Search Modes
-
-| Mode | When | Example |
-|------|------|---------|
-| \`hybrid\` (default) | General queries | \`kb_search({ query: "error handling" })\` |
-| \`semantic\` | Conceptual/meaning-based | \`kb_search({ query: "retry with backoff", search_mode: "semantic" })\` |
-| \`keyword\` | Exact identifiers | \`kb_search({ query: "CircuitBreaker", search_mode: "keyword" })\` |
-
-Filters: \`origin\` (\`indexed\`/\`curated\`/\`produced\`), \`category\`, \`content_type\`, \`tags\`, \`min_score\`.
-
----
-
-## Workflow Chains
-
-**Codebase onboarding:**
-\`\`\`
-kb_onboard({ path: "." }) \u2192 kb_produce_knowledge({ path: "src/" }) \u2192 kb_remember(...)
-\`\`\`
-
-**Planning a task:**
-\`\`\`
-kb_search({ query: "task keywords" })
-\u2192 kb_scope_map({ task: "description" })
-\u2192 kb_file_summary for each file in scope
-\u2192 kb_compact({ path: "relevant-file.ts", query: "detail needed" }) for files needing detail
-\u2192 kb_workset({ action: "save", name: "task", files: [...] })
-\`\`\`
-
-**Bug investigation:**
-\`\`\`
-kb_parse_output({ output: "<error>" })
-\u2192 kb_symbol({ name: "failingFn" })
-\u2192 kb_trace({ symbol: "failingFn", direction: "backward" })
-\u2192 kb_blast_radius({ changed_files: ["suspect.ts"] })
-\`\`\`
-
-**Safe refactor with lanes:**
-\`\`\`
-kb_lane({ action: "create", name: "refactor", files: [...] })
-\u2192 [make changes]
-\u2192 kb_lane({ action: "diff", name: "refactor" })
-\u2192 kb_check({}) \u2192 kb_test_run({})
-\u2192 kb_lane({ action: "merge", name: "refactor" })
-\`\`\`
-
-**After making changes:**
-\`\`\`
-kb_blast_radius({ changed_files: ["src/file.ts"] })
-\u2192 kb_check({}) \u2192 kb_test_run({})
-\u2192 kb_reindex({})
-\u2192 kb_remember(...)
-\`\`\`
-
----
-
-## Knowledge Categories
-
-| Category | What to store |
-|----------|---------------|
-| \`decisions\` | Architecture choices, trade-offs, rejected alternatives |
-| \`patterns\` | Code patterns, naming conventions, structural patterns |
-| \`conventions\` | Session checkpoints, workflow conventions, team agreements |
-| \`troubleshooting\` | Non-obvious fixes, debugging strategies, workarounds |
-
----
-
-## Core Rules Summary
-
-1. **Search KB before proposing anything new** \u2014 prior decisions exist
-2. **Compress context aggressively** \u2014 \`file_summary\` \u2192 \`compact\` \u2192 \`digest\`
-3. **Use persistent memory** \u2014 \`remember\` decisions, \`stash\` temporary data
-4. **Validate before committing** \u2014 \`check\` + \`test_run\` + \`blast_radius\`
-5. **Follow \`_Next:\` hints** in tool responses for guided workflow
-6. **Use FORGE for complex tasks** \u2014 \`forge_classify\` \u2192 \`forge_ground\` \u2192 \`evidence_map\`
-`}async function R(o){const s=process.cwd(),c=e(s,"kb.config.json");if(t(c)&&!o.force){console.log("kb.config.json already exists. Use --force to overwrite.");return}r(c,`${JSON.stringify(C,null,2)}
-`,"utf-8"),console.log("  Created kb.config.json");const d=e(s,"curated");t(d)||(n(d,{recursive:!0}),console.log("  Created curated/"));const a=e(s,".gitignore");t(a)?v(a,"utf-8").includes(".kb-data")||(h(a,`
-# Knowledge base vector store
-.kb-data/
-`,"utf-8"),console.log("  Added .kb-data/ to .gitignore")):(r(a,`# Knowledge base vector store
-.kb-data/
-`,"utf-8"),console.log("  Created .gitignore with .kb-data/"));const i=w(s),l=e(s,".vscode"),k=e(l,"mcp.json");t(l)&&!t(k)&&(r(k,`${JSON.stringify(A(i),null,2)}
-`,"utf-8"),console.log("  Created .vscode/mcp.json"));const u=e(s,".github"),b=e(u,"copilot-instructions.md");t(b)||(n(u,{recursive:!0}),r(b,T(i),"utf-8"),console.log("  Created .github/copilot-instructions.md"));const m=e(s,"AGENTS.md");t(m)||(r(m,P(i),"utf-8"),console.log("  Created AGENTS.md"));const _=e(s,"skills","knowledge-base"),p=e(_,"SKILL.md");if(!t(p)){const f=S(x(import.meta.url)),g=e(f,"..","..","..","skills","knowledge-base","SKILL.md");t(g)&&(n(_,{recursive:!0}),y(g,p),console.log("  Created skills/knowledge-base/SKILL.md"))}console.log(`
-Knowledge base initialized! Next steps:`),console.log("  kb reindex     Index your codebase"),console.log("  kb search      Search indexed content"),console.log("  kb serve       Start MCP server for IDE integration")}export{R as initProject};