69607b3a65
Implementa una base de conocimiento persistente por agente siguiendo el patrón pure core / impure shell: - pkg/knowledge/: tipos puros (Document, Store interface) - shell/knowledge/: FileStore con SQLite para indexación y archivos .md - tools/knowledge.go: 4 tools LLM (search, read, write, list) - tools/knowledge_test.go: tests unitarios de las tools - internal/config/schema.go: nuevo KnowledgeToolCfg en ToolsCfg - agents/runtime.go: inicialización del store y registro de tools - agents/*/knowledge/about-me.md: documentos semilla para cada agente Cada agente puede buscar, leer, crear y actualizar documentos de conocimiento. Los archivos .md viven en agents/<id>/knowledge/ y se indexan en SQLite (agents/<id>/data/knowledge.db). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
292 lines
7.9 KiB
Go
292 lines
7.9 KiB
Go
// Package shellknowledge implements the knowledge store using files + SQLite FTS5.
|
|
package shellknowledge
|
|
|
|
import (
|
|
"context"
|
|
"database/sql"
|
|
"fmt"
|
|
"log/slog"
|
|
"os"
|
|
"path/filepath"
|
|
"regexp"
|
|
"strings"
|
|
"time"
|
|
|
|
"github.com/enmanuel/agents/pkg/knowledge"
|
|
)
|
|
|
|
const ftsSchema = `
|
|
CREATE VIRTUAL TABLE IF NOT EXISTS documents USING fts5(
|
|
slug,
|
|
title,
|
|
content,
|
|
updated_at UNINDEXED
|
|
);
|
|
`
|
|
|
|
var slugRe = regexp.MustCompile(`^[a-z0-9][a-z0-9-]{0,62}[a-z0-9]$`)
|
|
|
|
// ValidSlug returns true if s is a valid document slug.
|
|
func ValidSlug(s string) bool {
|
|
if len(s) < 2 || len(s) > 64 {
|
|
return false
|
|
}
|
|
return slugRe.MatchString(s)
|
|
}
|
|
|
|
// FileStore implements knowledge.Store using markdown files + SQLite FTS5 index.
|
|
type FileStore struct {
|
|
dir string // path to agents/<id>/knowledge/
|
|
dbPath string // path to agents/<id>/data/knowledge.db
|
|
db *sql.DB
|
|
logger *slog.Logger
|
|
}
|
|
|
|
// New creates a FileStore. It ensures the knowledge dir and DB dir exist,
|
|
// opens the SQLite database, and creates the FTS5 table if needed.
|
|
func New(dir, dbPath string, logger *slog.Logger) (*FileStore, error) {
|
|
log := logger.With("component", "knowledge", "dir", dir, "db_path", dbPath)
|
|
|
|
if err := os.MkdirAll(dir, 0o755); err != nil {
|
|
return nil, fmt.Errorf("create knowledge dir: %w", err)
|
|
}
|
|
if err := os.MkdirAll(filepath.Dir(dbPath), 0o755); err != nil {
|
|
return nil, fmt.Errorf("create knowledge db dir: %w", err)
|
|
}
|
|
|
|
db, err := sql.Open("sqlite3", dbPath)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("open knowledge db: %w", err)
|
|
}
|
|
if _, err := db.Exec(ftsSchema); err != nil {
|
|
db.Close()
|
|
return nil, fmt.Errorf("create knowledge fts5 table: %w", err)
|
|
}
|
|
|
|
log.Info("knowledge_store_ready")
|
|
return &FileStore{dir: dir, dbPath: dbPath, db: db, logger: log}, nil
|
|
}
|
|
|
|
// Sync re-indexes all .md files from disk into the FTS5 table.
|
|
func (s *FileStore) Sync(ctx context.Context) error {
|
|
entries, err := os.ReadDir(s.dir)
|
|
if err != nil {
|
|
return fmt.Errorf("read knowledge dir: %w", err)
|
|
}
|
|
|
|
tx, err := s.db.BeginTx(ctx, nil)
|
|
if err != nil {
|
|
return fmt.Errorf("begin sync tx: %w", err)
|
|
}
|
|
defer tx.Rollback()
|
|
|
|
// Clear existing index
|
|
if _, err := tx.ExecContext(ctx, `DELETE FROM documents`); err != nil {
|
|
return fmt.Errorf("clear fts5 index: %w", err)
|
|
}
|
|
|
|
count := 0
|
|
for _, e := range entries {
|
|
if e.IsDir() || !strings.HasSuffix(e.Name(), ".md") {
|
|
continue
|
|
}
|
|
slug := strings.TrimSuffix(e.Name(), ".md")
|
|
if !ValidSlug(slug) {
|
|
s.logger.Warn("skipping invalid slug", "file", e.Name())
|
|
continue
|
|
}
|
|
|
|
content, err := os.ReadFile(filepath.Join(s.dir, e.Name()))
|
|
if err != nil {
|
|
s.logger.Warn("skipping unreadable file", "file", e.Name(), "err", err)
|
|
continue
|
|
}
|
|
|
|
info, err := e.Info()
|
|
if err != nil {
|
|
s.logger.Warn("skipping file without info", "file", e.Name(), "err", err)
|
|
continue
|
|
}
|
|
|
|
title := extractTitle(string(content), slug)
|
|
mtime := info.ModTime().UTC().Format(time.RFC3339)
|
|
|
|
if _, err := tx.ExecContext(ctx,
|
|
`INSERT INTO documents (slug, title, content, updated_at) VALUES (?, ?, ?, ?)`,
|
|
slug, title, string(content), mtime,
|
|
); err != nil {
|
|
s.logger.Warn("failed to index file", "slug", slug, "err", err)
|
|
continue
|
|
}
|
|
count++
|
|
}
|
|
|
|
if err := tx.Commit(); err != nil {
|
|
return fmt.Errorf("commit sync tx: %w", err)
|
|
}
|
|
|
|
s.logger.Info("knowledge_sync", "count", count)
|
|
return nil
|
|
}
|
|
|
|
// Search performs full-text search on the FTS5 index.
|
|
func (s *FileStore) Search(ctx context.Context, query string, limit int) ([]knowledge.SearchResult, error) {
|
|
if limit <= 0 {
|
|
limit = 5
|
|
}
|
|
|
|
rows, err := s.db.QueryContext(ctx,
|
|
`SELECT slug, title, snippet(documents, 2, '**', '**', '…', 32), rank
|
|
FROM documents WHERE documents MATCH ?
|
|
ORDER BY rank LIMIT ?`,
|
|
query, limit,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("knowledge search: %w", err)
|
|
}
|
|
defer rows.Close()
|
|
|
|
var results []knowledge.SearchResult
|
|
for rows.Next() {
|
|
var r knowledge.SearchResult
|
|
if err := rows.Scan(&r.Slug, &r.Title, &r.Snippet, &r.Rank); err != nil {
|
|
return nil, err
|
|
}
|
|
results = append(results, r)
|
|
}
|
|
return results, rows.Err()
|
|
}
|
|
|
|
// Get reads a document from disk by slug.
|
|
func (s *FileStore) Get(ctx context.Context, slug string) (*knowledge.Document, error) {
|
|
if !ValidSlug(slug) {
|
|
return nil, fmt.Errorf("invalid slug: %q", slug)
|
|
}
|
|
|
|
path := filepath.Join(s.dir, slug+".md")
|
|
content, err := os.ReadFile(path)
|
|
if err != nil {
|
|
if os.IsNotExist(err) {
|
|
return nil, fmt.Errorf("document not found: %q", slug)
|
|
}
|
|
return nil, fmt.Errorf("read document: %w", err)
|
|
}
|
|
|
|
info, err := os.Stat(path)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("stat document: %w", err)
|
|
}
|
|
|
|
return &knowledge.Document{
|
|
Slug: slug,
|
|
Title: extractTitle(string(content), slug),
|
|
Content: string(content),
|
|
UpdatedAt: info.ModTime().UTC(),
|
|
}, nil
|
|
}
|
|
|
|
// Put writes a document to disk and updates the FTS5 index.
|
|
func (s *FileStore) Put(ctx context.Context, doc knowledge.Document) error {
|
|
if !ValidSlug(doc.Slug) {
|
|
return fmt.Errorf("invalid slug: %q", doc.Slug)
|
|
}
|
|
if len(doc.Content) > 64*1024 {
|
|
return fmt.Errorf("document too large: %d bytes (max 65536)", len(doc.Content))
|
|
}
|
|
|
|
path := filepath.Join(s.dir, doc.Slug+".md")
|
|
if err := os.WriteFile(path, []byte(doc.Content), 0o644); err != nil {
|
|
return fmt.Errorf("write document: %w", err)
|
|
}
|
|
|
|
title := extractTitle(doc.Content, doc.Slug)
|
|
now := time.Now().UTC().Format(time.RFC3339)
|
|
|
|
// Upsert: delete old + insert new (FTS5 doesn't support UPDATE well)
|
|
tx, err := s.db.BeginTx(ctx, nil)
|
|
if err != nil {
|
|
return fmt.Errorf("begin put tx: %w", err)
|
|
}
|
|
defer tx.Rollback()
|
|
|
|
if _, err := tx.ExecContext(ctx, `DELETE FROM documents WHERE slug = ?`, doc.Slug); err != nil {
|
|
return fmt.Errorf("delete old index: %w", err)
|
|
}
|
|
if _, err := tx.ExecContext(ctx,
|
|
`INSERT INTO documents (slug, title, content, updated_at) VALUES (?, ?, ?, ?)`,
|
|
doc.Slug, title, doc.Content, now,
|
|
); err != nil {
|
|
return fmt.Errorf("insert index: %w", err)
|
|
}
|
|
|
|
if err := tx.Commit(); err != nil {
|
|
return fmt.Errorf("commit put tx: %w", err)
|
|
}
|
|
|
|
s.logger.Debug("knowledge_put", "slug", doc.Slug, "size", len(doc.Content))
|
|
return nil
|
|
}
|
|
|
|
// Delete removes a document from disk and the FTS5 index.
|
|
func (s *FileStore) Delete(ctx context.Context, slug string) error {
|
|
if !ValidSlug(slug) {
|
|
return fmt.Errorf("invalid slug: %q", slug)
|
|
}
|
|
|
|
path := filepath.Join(s.dir, slug+".md")
|
|
if err := os.Remove(path); err != nil && !os.IsNotExist(err) {
|
|
return fmt.Errorf("remove document: %w", err)
|
|
}
|
|
|
|
if _, err := s.db.ExecContext(ctx, `DELETE FROM documents WHERE slug = ?`, slug); err != nil {
|
|
return fmt.Errorf("delete from index: %w", err)
|
|
}
|
|
|
|
s.logger.Debug("knowledge_delete", "slug", slug)
|
|
return nil
|
|
}
|
|
|
|
// List returns all documents from the FTS5 index.
|
|
func (s *FileStore) List(ctx context.Context) ([]knowledge.Document, error) {
|
|
rows, err := s.db.QueryContext(ctx,
|
|
`SELECT slug, title, updated_at FROM documents ORDER BY slug`)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("knowledge list: %w", err)
|
|
}
|
|
defer rows.Close()
|
|
|
|
var docs []knowledge.Document
|
|
for rows.Next() {
|
|
var d knowledge.Document
|
|
var updatedAt string
|
|
if err := rows.Scan(&d.Slug, &d.Title, &updatedAt); err != nil {
|
|
return nil, err
|
|
}
|
|
d.UpdatedAt, _ = time.Parse(time.RFC3339, updatedAt)
|
|
docs = append(docs, d)
|
|
}
|
|
return docs, rows.Err()
|
|
}
|
|
|
|
// Close releases the SQLite database.
|
|
func (s *FileStore) Close() error {
|
|
s.logger.Info("knowledge_store_closed")
|
|
return s.db.Close()
|
|
}
|
|
|
|
// extractTitle returns the first H1 heading from markdown content, or a humanized slug.
|
|
func extractTitle(content, slug string) string {
|
|
for _, line := range strings.SplitN(content, "\n", 20) {
|
|
line = strings.TrimSpace(line)
|
|
if strings.HasPrefix(line, "# ") {
|
|
return strings.TrimPrefix(line, "# ")
|
|
}
|
|
}
|
|
// Humanize slug: "go-patterns" → "Go patterns"
|
|
humanized := strings.ReplaceAll(slug, "-", " ")
|
|
if len(humanized) > 0 {
|
|
humanized = strings.ToUpper(humanized[:1]) + humanized[1:]
|
|
}
|
|
return humanized
|
|
}
|