ci: build ui before go tests

feat(tools): add duplicate audit report
Merge pull request 'Link thoughts and learnings' (#36 ) from issue-32-link-thoughts-learnings into main
2026-07-14 16:32:00 +02:00 · 2026-07-14 16:28:06 +02:00 · 2026-07-14 14:42:05 +00:00 · 2026-07-14 15:18:57 +02:00
12 changed files with 873 additions and 73 deletions
@@ -31,6 +31,20 @@ jobs:
      - name: Download dependencies
        run: go mod download

+      - name: Set up Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: 'lts/*'
+
+      - name: Install pnpm
+        run: npm install -g pnpm
+
+      - name: Build UI
+        run: |
+          cd ui
+          pnpm install --frozen-lockfile
+          pnpm run build
+
      - name: Tidy modules
        run: go mod tidy

@@ -36,3 +36,4 @@ ui/.svelte-kit/
 internal/app/ui/dist/*
 !internal/app/ui/dist/placeholder.txt
 .codex
+.worktrees/
@@ -194,28 +194,30 @@ func routes(logger *slog.Logger, cfg *config.Config, info buildinfo.Info, db *st
 	adminActions := newAdminActions(backfillTool, enrichmentRetryer, logger)

 	toolSet := mcpserver.ToolSet{
-		Capture:         tools.NewCaptureTool(db, embeddings, cfg.Capture, activeProjects, enrichmentRetryer, backfillTool),
-		Search:          tools.NewSearchTool(db, embeddings, cfg.Search, activeProjects),
-		List:            tools.NewListTool(db, cfg.Search, activeProjects),
-		Stats:           tools.NewStatsTool(db),
-		Get:             tools.NewGetTool(db),
-		Update:          tools.NewUpdateTool(db, embeddings, metadata, cfg.Capture, logger),
-		Delete:          tools.NewDeleteTool(db),
-		Archive:         tools.NewArchiveTool(db),
-		Projects:        tools.NewProjectsTool(db, activeProjects),
-		Version:         tools.NewVersionTool(cfg.MCP.ServerName, info),
-		Learnings:       tools.NewLearningsTool(db, activeProjects, cfg.Search),
-		Plans:           tools.NewPlansTool(db, activeProjects, cfg.Search),
-		ProjectPersonas: tools.NewProjectPersonasTool(db, activeProjects),
-		WorldModel:      tools.NewWorldModelTool(db, activeProjects),
-		Context:         tools.NewContextTool(db, embeddings, cfg.Search, activeProjects),
-		Recall:          tools.NewRecallTool(db, embeddings, cfg.Search, activeProjects),
-		Summarize:       tools.NewSummarizeTool(db, embeddings, metadata, cfg.Search, activeProjects),
-		Links:           tools.NewLinksTool(db, embeddings, cfg.Search),
-		Files:           filesTool,
-		Backfill:        backfillTool,
-		Reparse:         tools.NewReparseMetadataTool(db, bgMetadata, cfg.Capture, activeProjects, logger),
-		RetryMetadata:   tools.NewRetryEnrichmentTool(enrichmentRetryer),
+		Capture:              tools.NewCaptureTool(db, embeddings, cfg.Capture, activeProjects, enrichmentRetryer, backfillTool),
+		Search:               tools.NewSearchTool(db, embeddings, cfg.Search, activeProjects),
+		List:                 tools.NewListTool(db, cfg.Search, activeProjects),
+		Stats:                tools.NewStatsTool(db),
+		Get:                  tools.NewGetTool(db),
+		Update:               tools.NewUpdateTool(db, embeddings, metadata, cfg.Capture, logger),
+		Delete:               tools.NewDeleteTool(db),
+		Archive:              tools.NewArchiveTool(db),
+		DuplicateAudit:       tools.NewDuplicateAuditTool(db, cfg.Search, activeProjects),
+		Projects:             tools.NewProjectsTool(db, activeProjects),
+		Version:              tools.NewVersionTool(cfg.MCP.ServerName, info),
+		Learnings:            tools.NewLearningsTool(db, activeProjects, cfg.Search),
+		Plans:                tools.NewPlansTool(db, activeProjects, cfg.Search),
+		ProjectPersonas:      tools.NewProjectPersonasTool(db, activeProjects),
+		WorldModel:           tools.NewWorldModelTool(db, activeProjects),
+		ThoughtLearningLinks: tools.NewThoughtLearningLinksTool(db),
+		Context:              tools.NewContextTool(db, embeddings, cfg.Search, activeProjects),
+		Recall:               tools.NewRecallTool(db, embeddings, cfg.Search, activeProjects),
+		Summarize:            tools.NewSummarizeTool(db, embeddings, metadata, cfg.Search, activeProjects),
+		Links:                tools.NewLinksTool(db, embeddings, cfg.Search),
+		Files:                filesTool,
+		Backfill:             backfillTool,
+		Reparse:              tools.NewReparseMetadataTool(db, bgMetadata, cfg.Capture, activeProjects, logger),
+		RetryMetadata:        tools.NewRetryEnrichmentTool(enrichmentRetryer),
 		//Maintenance:   tools.NewMaintenanceTool(db),
 		Skills:      tools.NewSkillsTool(db, activeProjects),
 		Personas:    tools.NewAgentPersonasTool(db),
@@ -19,33 +19,35 @@ const (
 )

 type ToolSet struct {
-	Version       *tools.VersionTool
-	Capture       *tools.CaptureTool
-	Search        *tools.SearchTool
-	List          *tools.ListTool
-	Stats         *tools.StatsTool
-	Get           *tools.GetTool
-	Update        *tools.UpdateTool
-	Delete        *tools.DeleteTool
-	Archive       *tools.ArchiveTool
-	Projects      *tools.ProjectsTool
-	Context       *tools.ContextTool
-	Recall        *tools.RecallTool
-	Summarize     *tools.SummarizeTool
-	Links         *tools.LinksTool
-	Files         *tools.FilesTool
-	Backfill      *tools.BackfillTool
-	Reparse       *tools.ReparseMetadataTool
-	RetryMetadata *tools.RetryEnrichmentTool
+	Version        *tools.VersionTool
+	Capture        *tools.CaptureTool
+	Search         *tools.SearchTool
+	List           *tools.ListTool
+	Stats          *tools.StatsTool
+	Get            *tools.GetTool
+	Update         *tools.UpdateTool
+	Delete         *tools.DeleteTool
+	Archive        *tools.ArchiveTool
+	DuplicateAudit *tools.DuplicateAuditTool
+	Projects       *tools.ProjectsTool
+	Context        *tools.ContextTool
+	Recall         *tools.RecallTool
+	Summarize      *tools.SummarizeTool
+	Links          *tools.LinksTool
+	Files          *tools.FilesTool
+	Backfill       *tools.BackfillTool
+	Reparse        *tools.ReparseMetadataTool
+	RetryMetadata  *tools.RetryEnrichmentTool
 	//Maintenance   *tools.MaintenanceTool
-	Skills          *tools.SkillsTool
-	Personas        *tools.AgentPersonasTool
-	ChatHistory     *tools.ChatHistoryTool
-	Describe        *tools.DescribeTool
-	Learnings       *tools.LearningsTool
-	Plans           *tools.PlansTool
-	ProjectPersonas *tools.ProjectPersonasTool
-	WorldModel      *tools.WorldModelTool
+	Skills               *tools.SkillsTool
+	Personas             *tools.AgentPersonasTool
+	ChatHistory          *tools.ChatHistoryTool
+	Describe             *tools.DescribeTool
+	Learnings            *tools.LearningsTool
+	Plans                *tools.PlansTool
+	ProjectPersonas      *tools.ProjectPersonasTool
+	WorldModel           *tools.WorldModelTool
+	ThoughtLearningLinks *tools.ThoughtLearningLinksTool
 }

 // Handlers groups the HTTP handlers produced for an MCP server instance.
@@ -91,6 +93,7 @@ func NewHandlers(cfg config.MCPConfig, logger *slog.Logger, toolSet ToolSet, onS
 		registerProjectTools,
 		registerWorldModelTools,
 		registerLearningTools,
+		registerThoughtLearningLinkTools,
 		registerPlanTools,
 		registerFileTools,
 		registerMaintenanceTools,
@@ -225,6 +228,12 @@ func registerThoughtTools(server *mcp.Server, logger *slog.Logger, toolSet ToolS
 	}, toolSet.Archive.Handle); err != nil {
 		return err
 	}
+	if err := addTool(server, logger, &mcp.Tool{
+		Name:        "audit_duplicates",
+		Description: "Dry-run duplicate audit for projects, thoughts, and metadata normalization candidates; performs no writes.",
+	}, toolSet.DuplicateAudit.Handle); err != nil {
+		return err
+	}
 	if err := addTool(server, logger, &mcp.Tool{
 		Name:        "summarize_thoughts",
 		Description: "LLM summary of a filtered set of thoughts.",
@@ -308,6 +317,34 @@ func registerLearningTools(server *mcp.Server, logger *slog.Logger, toolSet Tool
 	return nil
 }

+func registerThoughtLearningLinkTools(server *mcp.Server, logger *slog.Logger, toolSet ToolSet) error {
+	if err := addTool(server, logger, &mcp.Tool{
+		Name:        "link_thought_learning",
+		Description: "Create or update an explicit association between a raw thought and a curated learning.",
+	}, toolSet.ThoughtLearningLinks.Link); err != nil {
+		return err
+	}
+	if err := addTool(server, logger, &mcp.Tool{
+		Name:        "unlink_thought_learning",
+		Description: "Remove an explicit association between a thought and a learning.",
+	}, toolSet.ThoughtLearningLinks.Unlink); err != nil {
+		return err
+	}
+	if err := addTool(server, logger, &mcp.Tool{
+		Name:        "get_thought_learnings",
+		Description: "List curated learnings explicitly linked to a thought.",
+	}, toolSet.ThoughtLearningLinks.GetThoughtLearnings); err != nil {
+		return err
+	}
+	if err := addTool(server, logger, &mcp.Tool{
+		Name:        "get_learning_thoughts",
+		Description: "List raw thoughts explicitly linked to a learning.",
+	}, toolSet.ThoughtLearningLinks.GetLearningThoughts); err != nil {
+		return err
+	}
+	return nil
+}
+
 func registerPlanTools(server *mcp.Server, logger *slog.Logger, toolSet ToolSet) error {
 	if err := addTool(server, logger, &mcp.Tool{
 		Name:        "create_plan",
@@ -734,6 +771,7 @@ func BuildToolCatalog() []tools.ToolEntry {
 		{Name: "update_thought", Description: "Update thought content or merge metadata.", Category: "thoughts"},
 		{Name: "delete_thought", Description: "Hard-delete a thought by id.", Category: "thoughts"},
 		{Name: "archive_thought", Description: "Archive a thought so it is hidden from default search and listing.", Category: "thoughts"},
+		{Name: "audit_duplicates", Description: "Dry-run duplicate audit for exact/normalized project names, thought content, and metadata value variants. Reports candidates and recommendations; performs no writes.", Category: "admin"},
 		{Name: "summarize_thoughts", Description: "Produce an LLM prose summary of a filtered or searched set of thoughts.", Category: "thoughts"},
 		{Name: "recall_context", Description: "Recall semantically relevant and recent context for prompt injection. Combines vector similarity with recency. Falls back to full-text search when no embeddings exist.", Category: "thoughts"},
 		{Name: "link_thoughts", Description: "Create a typed relationship between two thoughts.", Category: "thoughts"},
@@ -755,6 +793,10 @@ func BuildToolCatalog() []tools.ToolEntry {
 		{Name: "add_learning", Description: "Create a curated learning record distinct from raw thoughts.", Category: "projects"},
 		{Name: "get_learning", Description: "Retrieve a structured learning by id.", Category: "projects"},
 		{Name: "list_learnings", Description: "List structured learnings with optional project, category, area, status, priority, tag, and text filters.", Category: "projects"},
+		{Name: "link_thought_learning", Description: "Create or update a lightweight explicit association between a raw thought and a curated learning.", Category: "projects"},
+		{Name: "unlink_thought_learning", Description: "Remove a lightweight explicit association between a thought and a learning.", Category: "projects"},
+		{Name: "get_thought_learnings", Description: "List curated learnings explicitly associated with a thought.", Category: "projects"},
+		{Name: "get_learning_thoughts", Description: "List raw source thoughts explicitly associated with a learning.", Category: "projects"},

 		// plans
 		{Name: "create_plan", Description: "Create a structured plan with status, priority, owner, due date, and optional project link.", Category: "plans"},
@@ -189,30 +189,31 @@ func TestStreamableHTTPReturnsStructuredToolErrors(t *testing.T) {

 func streamableTestToolSet() ToolSet {
 	return ToolSet{
-		Version:         tools.NewVersionTool("test", buildinfo.Info{Version: "0.0.1", TagName: "v0.0.1", Commit: "test", BuildDate: "2026-03-31T00:00:00Z"}),
-		Capture:         new(tools.CaptureTool),
-		Search:          new(tools.SearchTool),
-		List:            new(tools.ListTool),
-		Stats:           new(tools.StatsTool),
-		Get:             new(tools.GetTool),
-		Update:          new(tools.UpdateTool),
-		Delete:          new(tools.DeleteTool),
-		Archive:         new(tools.ArchiveTool),
-		Projects:        new(tools.ProjectsTool),
-		Context:         new(tools.ContextTool),
-		Recall:          new(tools.RecallTool),
-		Summarize:       new(tools.SummarizeTool),
-		Links:           new(tools.LinksTool),
-		Files:           new(tools.FilesTool),
-		Backfill:        new(tools.BackfillTool),
-		Reparse:         new(tools.ReparseMetadataTool),
-		RetryMetadata:   new(tools.RetryEnrichmentTool),
-		Skills:          new(tools.SkillsTool),
-		ChatHistory:     new(tools.ChatHistoryTool),
-		Describe:        new(tools.DescribeTool),
-		Learnings:       new(tools.LearningsTool),
-		Plans:           new(tools.PlansTool),
-		ProjectPersonas: new(tools.ProjectPersonasTool),
-		WorldModel:      new(tools.WorldModelTool),
+		Version:              tools.NewVersionTool("test", buildinfo.Info{Version: "0.0.1", TagName: "v0.0.1", Commit: "test", BuildDate: "2026-03-31T00:00:00Z"}),
+		Capture:              new(tools.CaptureTool),
+		Search:               new(tools.SearchTool),
+		List:                 new(tools.ListTool),
+		Stats:                new(tools.StatsTool),
+		Get:                  new(tools.GetTool),
+		Update:               new(tools.UpdateTool),
+		Delete:               new(tools.DeleteTool),
+		Archive:              new(tools.ArchiveTool),
+		Projects:             new(tools.ProjectsTool),
+		Context:              new(tools.ContextTool),
+		Recall:               new(tools.RecallTool),
+		Summarize:            new(tools.SummarizeTool),
+		Links:                new(tools.LinksTool),
+		Files:                new(tools.FilesTool),
+		Backfill:             new(tools.BackfillTool),
+		Reparse:              new(tools.ReparseMetadataTool),
+		RetryMetadata:        new(tools.RetryEnrichmentTool),
+		Skills:               new(tools.SkillsTool),
+		ChatHistory:          new(tools.ChatHistoryTool),
+		Describe:             new(tools.DescribeTool),
+		Learnings:            new(tools.LearningsTool),
+		Plans:                new(tools.PlansTool),
+		ProjectPersonas:      new(tools.ProjectPersonasTool),
+		WorldModel:           new(tools.WorldModelTool),
+		ThoughtLearningLinks: new(tools.ThoughtLearningLinksTool),
 	}
 }
@@ -0,0 +1,148 @@
+package store
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/google/uuid"
+
+	"git.warky.dev/wdevs/amcs/internal/generatedmodels"
+	thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
+)
+
+// LinkThoughtLearning creates or updates an association between a thought (by GUID) and a learning (by numeric ID).
+// If the pair already exists, the relation label is updated.
+func (db *DB) LinkThoughtLearning(ctx context.Context, thoughtGUID uuid.UUID, learningID int64, relation string) error {
+	_, err := db.pool.Exec(ctx, `
+		insert into thought_learning_links (thought_id, learning_id, relation)
+		select t.id, $2, $3
+		from thoughts t
+		where t.guid = $1
+		on conflict (thought_id, learning_id) do update set relation = excluded.relation
+	`, thoughtGUID, learningID, relation)
+	if err != nil {
+		return fmt.Errorf("link thought learning: %w", err)
+	}
+	return nil
+}
+
+// UnlinkThoughtLearning removes the association between a thought (by GUID) and a learning (by numeric ID).
+func (db *DB) UnlinkThoughtLearning(ctx context.Context, thoughtGUID uuid.UUID, learningID int64) error {
+	_, err := db.pool.Exec(ctx, `
+		delete from thought_learning_links
+		where thought_id = (select id from thoughts where guid = $1)
+		  and learning_id = $2
+	`, thoughtGUID, learningID)
+	if err != nil {
+		return fmt.Errorf("unlink thought learning: %w", err)
+	}
+	return nil
+}
+
+// GetLinkedLearnings returns all learnings explicitly associated with a thought (by GUID).
+func (db *DB) GetLinkedLearnings(ctx context.Context, thoughtGUID uuid.UUID) ([]thoughttypes.LinkedLearning, error) {
+	rows, err := db.pool.Query(ctx, `
+		select l.id, l.guid, l.summary, l.details, l.category, l.area, l.status, l.priority, l.confidence,
+		       l.action_required, l.source_type, l.source_ref, l.project_id, l.related_thought_id,
+		       l.related_skill_id, l.reviewed_by, l.reviewed_at, l.duplicate_of_learning_id,
+		       l.supersedes_learning_id, l.tags::text[], l.created_at, l.updated_at,
+		       tll.relation, tll.created_at as linked_at
+		from thought_learning_links tll
+		join learnings l on l.id = tll.learning_id
+		where tll.thought_id = (select id from thoughts where guid = $1)
+		order by tll.created_at desc
+	`, thoughtGUID)
+	if err != nil {
+		return nil, fmt.Errorf("query linked learnings: %w", err)
+	}
+	defer rows.Close()
+
+	items := make([]thoughttypes.LinkedLearning, 0)
+	for rows.Next() {
+		var tags []string
+		var linked thoughttypes.LinkedLearning
+		var m generatedmodels.ModelPublicLearnings
+
+		if err := rows.Scan(
+			&m.ID,
+			&m.GUID,
+			&m.Summary,
+			&m.Details,
+			&m.Category,
+			&m.Area,
+			&m.Status,
+			&m.Priority,
+			&m.Confidence,
+			&m.ActionRequired,
+			&m.SourceType,
+			&m.SourceRef,
+			&m.ProjectID,
+			&m.RelatedThoughtID,
+			&m.RelatedSkillID,
+			&m.ReviewedBy,
+			&m.ReviewedAt,
+			&m.DuplicateOfLearningID,
+			&m.SupersedesLearningID,
+			&tags,
+			&m.CreatedAt,
+			&m.UpdatedAt,
+			&linked.Relation,
+			&linked.CreatedAt,
+		); err != nil {
+			return nil, fmt.Errorf("scan linked learning: %w", err)
+		}
+		linked.Learning = learningFromModel(m, tags)
+		items = append(items, linked)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("iterate linked learnings: %w", err)
+	}
+	return items, nil
+}
+
+// GetLinkedThoughtsForLearning returns all thoughts explicitly associated with a learning (by numeric ID).
+func (db *DB) GetLinkedThoughtsForLearning(ctx context.Context, learningID int64) ([]thoughttypes.LinkedLearningThought, error) {
+	rows, err := db.pool.Query(ctx, `
+		select t.id, t.guid, t.content, t.metadata, t.project_id, t.archived_at, t.created_at, t.updated_at,
+		       tll.relation, tll.created_at as linked_at
+		from thought_learning_links tll
+		join thoughts t on t.id = tll.thought_id
+		where tll.learning_id = $1
+		order by tll.created_at desc
+	`, learningID)
+	if err != nil {
+		return nil, fmt.Errorf("query linked thoughts for learning: %w", err)
+	}
+	defer rows.Close()
+
+	items := make([]thoughttypes.LinkedLearningThought, 0)
+	for rows.Next() {
+		var linked thoughttypes.LinkedLearningThought
+		var m generatedmodels.ModelPublicThoughts
+
+		if err := rows.Scan(
+			&m.ID,
+			&m.GUID,
+			&m.Content,
+			&m.Metadata,
+			&m.ProjectID,
+			&m.ArchivedAt,
+			&m.CreatedAt,
+			&m.UpdatedAt,
+			&linked.Relation,
+			&linked.CreatedAt,
+		); err != nil {
+			return nil, fmt.Errorf("scan linked thought for learning: %w", err)
+		}
+		thought, err := thoughtFromModel(m)
+		if err != nil {
+			return nil, fmt.Errorf("map linked thought for learning: %w", err)
+		}
+		linked.Thought = thought
+		items = append(items, linked)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("iterate linked thoughts for learning: %w", err)
+	}
+	return items, nil
+}
@@ -0,0 +1,305 @@
+package tools
+
+import (
+	"context"
+	"regexp"
+	"sort"
+	"strings"
+
+	"github.com/google/uuid"
+	"github.com/modelcontextprotocol/go-sdk/mcp"
+
+	"git.warky.dev/wdevs/amcs/internal/config"
+	"git.warky.dev/wdevs/amcs/internal/session"
+	"git.warky.dev/wdevs/amcs/internal/store"
+	thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
+)
+
+const defaultDuplicateAuditLimit = 50
+
+var duplicateWhitespace = regexp.MustCompile(`\s+`)
+
+type DuplicateAuditTool struct {
+	store    *store.DB
+	search   config.SearchConfig
+	sessions *session.ActiveProjects
+}
+
+type DuplicateAuditInput struct {
+	Project         string `json:"project,omitempty" jsonschema:"optional project name or id to scope thought duplicate scanning"`
+	Limit           int    `json:"limit,omitempty" jsonschema:"maximum candidate groups to return"`
+	ThoughtLimit    int    `json:"thought_limit,omitempty" jsonschema:"maximum thoughts to scan for duplicate content; defaults to search limit"`
+	IncludeArchived bool   `json:"include_archived,omitempty" jsonschema:"include archived thoughts in the scan"`
+}
+
+type DuplicateAuditOutput struct {
+	Report DuplicateAuditReport `json:"report"`
+}
+
+type DuplicateAuditReport struct {
+	Summary             DuplicateAuditSummary       `json:"summary"`
+	ProjectCandidates   []DuplicateProjectCandidate `json:"project_candidates"`
+	ThoughtCandidates   []DuplicateThoughtCandidate `json:"thought_candidates"`
+	MetadataCandidates  []MetadataCleanupCandidate  `json:"metadata_candidates"`
+	RecommendedNextStep string                      `json:"recommended_next_step"`
+}
+
+type DuplicateAuditSummary struct {
+	DryRun                  bool `json:"dry_run"`
+	ScannedProjects         int  `json:"scanned_projects"`
+	ScannedThoughts         int  `json:"scanned_thoughts"`
+	ProjectCandidateGroups  int  `json:"project_candidate_groups"`
+	ThoughtCandidateGroups  int  `json:"thought_candidate_groups"`
+	MetadataCandidateGroups int  `json:"metadata_candidate_groups"`
+	Truncated               bool `json:"truncated"`
+}
+
+type DuplicateProjectCandidate struct {
+	MatchType              string                        `json:"match_type"`
+	NormalizedValue        string                        `json:"normalized_value"`
+	Confidence             float64                       `json:"confidence"`
+	RecommendedCanonicalID uuid.UUID                     `json:"recommended_canonical_id"`
+	Projects               []thoughttypes.ProjectSummary `json:"projects"`
+}
+
+type DuplicateThoughtCandidate struct {
+	MatchType              string                 `json:"match_type"`
+	NormalizedValue        string                 `json:"normalized_value"`
+	Confidence             float64                `json:"confidence"`
+	RecommendedCanonicalID uuid.UUID              `json:"recommended_canonical_id"`
+	Thoughts               []thoughttypes.Thought `json:"thoughts"`
+}
+
+type MetadataCleanupCandidate struct {
+	Field             string   `json:"field"`
+	NormalizedValue   string   `json:"normalized_value"`
+	Values            []string `json:"values"`
+	Occurrences       int      `json:"occurrences"`
+	RecommendedValue  string   `json:"recommended_value"`
+	RecommendedAction string   `json:"recommended_action"`
+}
+
+func NewDuplicateAuditTool(db *store.DB, search config.SearchConfig, sessions *session.ActiveProjects) *DuplicateAuditTool {
+	return &DuplicateAuditTool{store: db, search: search, sessions: sessions}
+}
+
+func (t *DuplicateAuditTool) Handle(ctx context.Context, req *mcp.CallToolRequest, in DuplicateAuditInput) (*mcp.CallToolResult, DuplicateAuditOutput, error) {
+	project, err := resolveProject(ctx, t.store, t.sessions, req, in.Project, false)
+	if err != nil {
+		return nil, DuplicateAuditOutput{}, err
+	}
+	var projectID *int64
+	if project != nil {
+		projectID = &project.NumericID
+	}
+
+	projects, err := t.store.ListProjects(ctx)
+	if err != nil {
+		return nil, DuplicateAuditOutput{}, err
+	}
+
+	thoughtLimit := normalizeLimit(in.ThoughtLimit, t.search)
+	if in.ThoughtLimit <= 0 {
+		thoughtLimit = normalizeLimit(0, t.search)
+	}
+	thoughts, err := t.store.ListThoughts(ctx, thoughttypes.ListFilter{Limit: thoughtLimit, ProjectID: projectID, IncludeArchived: in.IncludeArchived})
+	if err != nil {
+		return nil, DuplicateAuditOutput{}, err
+	}
+
+	if project != nil {
+		_ = t.store.TouchProject(ctx, project.NumericID)
+	}
+	return nil, DuplicateAuditOutput{Report: buildDuplicateAuditReport(projects, thoughts, in)}, nil
+}
+
+func buildDuplicateAuditReport(projects []thoughttypes.ProjectSummary, thoughts []thoughttypes.Thought, in DuplicateAuditInput) DuplicateAuditReport {
+	limit := in.Limit
+	if limit <= 0 {
+		limit = defaultDuplicateAuditLimit
+	}
+
+	projectCandidates, projectTruncated := duplicateProjectCandidates(projects, limit)
+	thoughtCandidates, thoughtTruncated := duplicateThoughtCandidates(thoughts, limit)
+	metadataCandidates, metadataTruncated := metadataCleanupCandidates(thoughts, limit)
+
+	return DuplicateAuditReport{
+		Summary: DuplicateAuditSummary{
+			DryRun:                  true,
+			ScannedProjects:         len(projects),
+			ScannedThoughts:         len(thoughts),
+			ProjectCandidateGroups:  len(projectCandidates),
+			ThoughtCandidateGroups:  len(thoughtCandidates),
+			MetadataCandidateGroups: len(metadataCandidates),
+			Truncated:               projectTruncated || thoughtTruncated || metadataTruncated,
+		},
+		ProjectCandidates:   projectCandidates,
+		ThoughtCandidates:   thoughtCandidates,
+		MetadataCandidates:  metadataCandidates,
+		RecommendedNextStep: "Review candidates manually, then use explicit archive/update/merge tools; this audit performs no writes.",
+	}
+}
+
+func duplicateProjectCandidates(projects []thoughttypes.ProjectSummary, limit int) ([]DuplicateProjectCandidate, bool) {
+	groups := map[string][]thoughttypes.ProjectSummary{}
+	for _, project := range projects {
+		key := normalizeDuplicateText(project.Name)
+		if key != "" {
+			groups[key] = append(groups[key], project)
+		}
+	}
+	keys := sortedDuplicateKeys(groups)
+	out := make([]DuplicateProjectCandidate, 0, min(len(keys), limit))
+	for _, key := range keys {
+		items := groups[key]
+		sort.SliceStable(items, func(i, j int) bool {
+			if items[i].ThoughtCount != items[j].ThoughtCount {
+				return items[i].ThoughtCount > items[j].ThoughtCount
+			}
+			return items[i].CreatedAt.Before(items[j].CreatedAt)
+		})
+		matchType := "normalized_name"
+		if allProjectNamesExact(items) {
+			matchType = "exact_name"
+		}
+		out = append(out, DuplicateProjectCandidate{MatchType: matchType, NormalizedValue: key, Confidence: 1.0, RecommendedCanonicalID: items[0].ID, Projects: items})
+		if len(out) == limit {
+			return out, len(keys) > limit
+		}
+	}
+	return out, false
+}
+
+func duplicateThoughtCandidates(thoughts []thoughttypes.Thought, limit int) ([]DuplicateThoughtCandidate, bool) {
+	groups := map[string][]thoughttypes.Thought{}
+	for _, thought := range thoughts {
+		key := normalizeDuplicateText(thought.Content)
+		if key != "" {
+			groups[key] = append(groups[key], thought)
+		}
+	}
+	keys := sortedDuplicateKeys(groups)
+	out := make([]DuplicateThoughtCandidate, 0, min(len(keys), limit))
+	for _, key := range keys {
+		items := groups[key]
+		sort.SliceStable(items, func(i, j int) bool { return items[i].CreatedAt.Before(items[j].CreatedAt) })
+		matchType := "normalized_content"
+		if allThoughtContentExact(items) {
+			matchType = "exact_content"
+		}
+		out = append(out, DuplicateThoughtCandidate{MatchType: matchType, NormalizedValue: key, Confidence: 1.0, RecommendedCanonicalID: items[0].GUID, Thoughts: items})
+		if len(out) == limit {
+			return out, len(keys) > limit
+		}
+	}
+	return out, false
+}
+
+func metadataCleanupCandidates(thoughts []thoughttypes.Thought, limit int) ([]MetadataCleanupCandidate, bool) {
+	groups := map[string]map[string]int{}
+	add := func(field, value string) {
+		trimmed := strings.TrimSpace(value)
+		key := normalizeDuplicateText(trimmed)
+		if key == "" || trimmed == "" {
+			return
+		}
+		bucketKey := field + "\x00" + key
+		if groups[bucketKey] == nil {
+			groups[bucketKey] = map[string]int{}
+		}
+		groups[bucketKey][trimmed]++
+	}
+	for _, thought := range thoughts {
+		add("type", thought.Metadata.Type)
+		add("source", thought.Metadata.Source)
+		for _, topic := range thought.Metadata.Topics {
+			add("topics", topic)
+		}
+		for _, person := range thought.Metadata.People {
+			add("people", person)
+		}
+	}
+
+	keys := make([]string, 0, len(groups))
+	for key, values := range groups {
+		if len(values) > 1 {
+			keys = append(keys, key)
+		}
+	}
+	sort.Strings(keys)
+
+	out := make([]MetadataCleanupCandidate, 0, min(len(keys), limit))
+	for _, key := range keys {
+		parts := strings.SplitN(key, "\x00", 2)
+		values, occurrences, recommended := valuesByCount(groups[key])
+		out = append(out, MetadataCleanupCandidate{
+			Field:             parts[0],
+			NormalizedValue:   parts[1],
+			Values:            values,
+			Occurrences:       occurrences,
+			RecommendedValue:  recommended,
+			RecommendedAction: "preview bulk metadata remap before applying; no changes made by audit",
+		})
+		if len(out) == limit {
+			return out, len(keys) > limit
+		}
+	}
+	return out, false
+}
+
+func sortedDuplicateKeys[T any](groups map[string][]T) []string {
+	keys := make([]string, 0, len(groups))
+	for key, items := range groups {
+		if len(items) > 1 {
+			keys = append(keys, key)
+		}
+	}
+	sort.Strings(keys)
+	return keys
+}
+
+func valuesByCount(counts map[string]int) ([]string, int, string) {
+	values := make([]string, 0, len(counts))
+	occurrences := 0
+	for value, count := range counts {
+		values = append(values, value)
+		occurrences += count
+	}
+	sort.Slice(values, func(i, j int) bool {
+		if counts[values[i]] != counts[values[j]] {
+			return counts[values[i]] > counts[values[j]]
+		}
+		return values[i] < values[j]
+	})
+	return values, occurrences, values[0]
+}
+
+func normalizeDuplicateText(value string) string {
+	return duplicateWhitespace.ReplaceAllString(strings.ToLower(strings.TrimSpace(value)), " ")
+}
+
+func allProjectNamesExact(projects []thoughttypes.ProjectSummary) bool {
+	if len(projects) == 0 {
+		return false
+	}
+	first := strings.TrimSpace(projects[0].Name)
+	for _, project := range projects[1:] {
+		if strings.TrimSpace(project.Name) != first {
+			return false
+		}
+	}
+	return true
+}
+
+func allThoughtContentExact(thoughts []thoughttypes.Thought) bool {
+	if len(thoughts) == 0 {
+		return false
+	}
+	first := strings.TrimSpace(thoughts[0].Content)
+	for _, thought := range thoughts[1:] {
+		if strings.TrimSpace(thought.Content) != first {
+			return false
+		}
+	}
+	return true
+}
@@ -0,0 +1,76 @@
+package tools
+
+import (
+	"testing"
+	"time"
+
+	"github.com/google/uuid"
+
+	thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
+)
+
+func TestBuildDuplicateAuditGroupsProjectsAndThoughtsSafely(t *testing.T) {
+	now := time.Date(2026, 4, 12, 20, 5, 0, 0, time.UTC)
+	projects := []thoughttypes.ProjectSummary{
+		{Project: thoughttypes.Project{ID: uuid.MustParse("11111111-1111-1111-1111-111111111111"), NumericID: 1, Name: "AMCS", CreatedAt: now}, ThoughtCount: 3},
+		{Project: thoughttypes.Project{ID: uuid.MustParse("22222222-2222-2222-2222-222222222222"), NumericID: 2, Name: "amcs ", CreatedAt: now.Add(time.Hour)}, ThoughtCount: 1},
+		{Project: thoughttypes.Project{ID: uuid.MustParse("33333333-3333-3333-3333-333333333333"), NumericID: 3, Name: "other", CreatedAt: now}, ThoughtCount: 1},
+	}
+	thoughts := []thoughttypes.Thought{
+		{ID: 10, GUID: uuid.MustParse("aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa"), Content: "Same text", Metadata: thoughttypes.ThoughtMetadata{Type: "Note", Topics: []string{"Go", "go"}, People: []string{"Alice"}}, CreatedAt: now},
+		{ID: 11, GUID: uuid.MustParse("bbbbbbbb-bbbb-bbbb-bbbb-bbbbbbbbbbbb"), Content: " same   text ", Metadata: thoughttypes.ThoughtMetadata{Type: "note", Topics: []string{"go"}, People: []string{"alice"}}, CreatedAt: now.Add(time.Hour)},
+		{ID: 12, GUID: uuid.MustParse("cccccccc-cccc-cccc-cccc-cccccccccccc"), Content: "unique", Metadata: thoughttypes.ThoughtMetadata{Type: "Task", Topics: []string{"Tasks"}}, CreatedAt: now},
+	}
+
+	report := buildDuplicateAuditReport(projects, thoughts, DuplicateAuditInput{})
+
+	if len(report.ProjectCandidates) != 1 {
+		t.Fatalf("project candidates = %d, want 1: %#v", len(report.ProjectCandidates), report.ProjectCandidates)
+	}
+	projectCandidate := report.ProjectCandidates[0]
+	if projectCandidate.MatchType != "normalized_name" || projectCandidate.NormalizedValue != "amcs" || len(projectCandidate.Projects) != 2 {
+		t.Fatalf("project candidate mismatch: %#v", projectCandidate)
+	}
+	if projectCandidate.RecommendedCanonicalID != projects[0].ID {
+		t.Fatalf("recommended project = %s, want %s", projectCandidate.RecommendedCanonicalID, projects[0].ID)
+	}
+
+	if len(report.ThoughtCandidates) != 1 {
+		t.Fatalf("thought candidates = %d, want 1: %#v", len(report.ThoughtCandidates), report.ThoughtCandidates)
+	}
+	thoughtCandidate := report.ThoughtCandidates[0]
+	if thoughtCandidate.MatchType != "normalized_content" || thoughtCandidate.NormalizedValue != "same text" || len(thoughtCandidate.Thoughts) != 2 {
+		t.Fatalf("thought candidate mismatch: %#v", thoughtCandidate)
+	}
+	if thoughtCandidate.RecommendedCanonicalID != thoughts[0].GUID {
+		t.Fatalf("recommended thought = %s, want %s", thoughtCandidate.RecommendedCanonicalID, thoughts[0].GUID)
+	}
+
+	if len(report.MetadataCandidates) != 3 {
+		t.Fatalf("metadata candidates = %d, want 3: %#v", len(report.MetadataCandidates), report.MetadataCandidates)
+	}
+	if report.Summary.ProjectCandidateGroups != 1 || report.Summary.ThoughtCandidateGroups != 1 || report.Summary.MetadataCandidateGroups != 3 {
+		t.Fatalf("summary mismatch: %#v", report.Summary)
+	}
+	if report.Summary.DryRun != true {
+		t.Fatalf("dry run = false, want true")
+	}
+}
+
+func TestBuildDuplicateAuditRespectsLimit(t *testing.T) {
+	projects := []thoughttypes.ProjectSummary{
+		{Project: thoughttypes.Project{ID: uuid.MustParse("11111111-1111-1111-1111-111111111111"), Name: "Alpha"}},
+		{Project: thoughttypes.Project{ID: uuid.MustParse("22222222-2222-2222-2222-222222222222"), Name: "alpha"}},
+		{Project: thoughttypes.Project{ID: uuid.MustParse("33333333-3333-3333-3333-333333333333"), Name: "Beta"}},
+		{Project: thoughttypes.Project{ID: uuid.MustParse("44444444-4444-4444-4444-444444444444"), Name: "beta"}},
+	}
+
+	report := buildDuplicateAuditReport(projects, nil, DuplicateAuditInput{Limit: 1})
+
+	if len(report.ProjectCandidates) != 1 {
+		t.Fatalf("project candidates = %d, want 1", len(report.ProjectCandidates))
+	}
+	if report.Summary.Truncated != true {
+		t.Fatalf("truncated = false, want true")
+	}
+}
@@ -0,0 +1,152 @@
+package tools
+
+import (
+	"context"
+	"strconv"
+	"strings"
+
+	"github.com/modelcontextprotocol/go-sdk/mcp"
+
+	"git.warky.dev/wdevs/amcs/internal/store"
+	thoughttypes "git.warky.dev/wdevs/amcs/internal/types"
+)
+
+type ThoughtLearningLinksTool struct {
+	store *store.DB
+}
+
+type LinkThoughtLearningInput struct {
+	ThoughtID  string `json:"thought_id" jsonschema:"UUID of the thought to link"`
+	LearningID int64  `json:"learning_id" jsonschema:"numeric id of the learning to link"`
+	Relation   string `json:"relation,omitempty" jsonschema:"relationship label, e.g. source, derived_from, related; defaults to source"`
+}
+
+type LinkThoughtLearningOutput struct {
+	Linked bool `json:"linked"`
+}
+
+type UnlinkThoughtLearningInput struct {
+	ThoughtID  string `json:"thought_id" jsonschema:"UUID of the thought"`
+	LearningID int64  `json:"learning_id" jsonschema:"numeric id of the learning"`
+}
+
+type UnlinkThoughtLearningOutput struct {
+	Unlinked bool `json:"unlinked"`
+}
+
+type GetThoughtLearningsInput struct {
+	ThoughtID string `json:"thought_id" jsonschema:"UUID of the thought"`
+}
+
+type GetThoughtLearningsOutput struct {
+	Learnings []thoughttypes.LinkedLearning `json:"learnings"`
+}
+
+type GetLearningThoughtsInput struct {
+	LearningID int64 `json:"learning_id" jsonschema:"numeric id of the learning"`
+}
+
+type GetLearningThoughtsOutput struct {
+	Thoughts []thoughttypes.LinkedLearningThought `json:"thoughts"`
+}
+
+func NewThoughtLearningLinksTool(db *store.DB) *ThoughtLearningLinksTool {
+	return &ThoughtLearningLinksTool{store: db}
+}
+
+func (t *ThoughtLearningLinksTool) Link(ctx context.Context, _ *mcp.CallToolRequest, in LinkThoughtLearningInput) (*mcp.CallToolResult, LinkThoughtLearningOutput, error) {
+	if err := t.ensureConfigured(); err != nil {
+		return nil, LinkThoughtLearningOutput{}, err
+	}
+
+	thoughtGUID, err := parseUUID(in.ThoughtID)
+	if err != nil {
+		return nil, LinkThoughtLearningOutput{}, err
+	}
+	if in.LearningID <= 0 {
+		return nil, LinkThoughtLearningOutput{}, errRequiredField("learning_id")
+	}
+	relation := strings.TrimSpace(in.Relation)
+	if relation == "" {
+		relation = "source"
+	}
+
+	if _, err := t.store.GetThought(ctx, thoughtGUID); err != nil {
+		return nil, LinkThoughtLearningOutput{}, errEntityNotFound("thought", "thought_id", in.ThoughtID)
+	}
+	if _, err := t.store.GetLearning(ctx, in.LearningID); err != nil {
+		return nil, LinkThoughtLearningOutput{}, errEntityNotFound("learning", "learning_id", strconv.FormatInt(in.LearningID, 10))
+	}
+
+	if err := t.store.LinkThoughtLearning(ctx, thoughtGUID, in.LearningID, relation); err != nil {
+		return nil, LinkThoughtLearningOutput{}, err
+	}
+	return nil, LinkThoughtLearningOutput{Linked: true}, nil
+}
+
+func (t *ThoughtLearningLinksTool) Unlink(ctx context.Context, _ *mcp.CallToolRequest, in UnlinkThoughtLearningInput) (*mcp.CallToolResult, UnlinkThoughtLearningOutput, error) {
+	if err := t.ensureConfigured(); err != nil {
+		return nil, UnlinkThoughtLearningOutput{}, err
+	}
+
+	thoughtGUID, err := parseUUID(in.ThoughtID)
+	if err != nil {
+		return nil, UnlinkThoughtLearningOutput{}, err
+	}
+	if in.LearningID <= 0 {
+		return nil, UnlinkThoughtLearningOutput{}, errRequiredField("learning_id")
+	}
+
+	if err := t.store.UnlinkThoughtLearning(ctx, thoughtGUID, in.LearningID); err != nil {
+		return nil, UnlinkThoughtLearningOutput{}, err
+	}
+	return nil, UnlinkThoughtLearningOutput{Unlinked: true}, nil
+}
+
+func (t *ThoughtLearningLinksTool) GetThoughtLearnings(ctx context.Context, _ *mcp.CallToolRequest, in GetThoughtLearningsInput) (*mcp.CallToolResult, GetThoughtLearningsOutput, error) {
+	if err := t.ensureConfigured(); err != nil {
+		return nil, GetThoughtLearningsOutput{}, err
+	}
+
+	thoughtGUID, err := parseUUID(in.ThoughtID)
+	if err != nil {
+		return nil, GetThoughtLearningsOutput{}, err
+	}
+
+	if _, err := t.store.GetThought(ctx, thoughtGUID); err != nil {
+		return nil, GetThoughtLearningsOutput{}, errEntityNotFound("thought", "thought_id", in.ThoughtID)
+	}
+
+	learnings, err := t.store.GetLinkedLearnings(ctx, thoughtGUID)
+	if err != nil {
+		return nil, GetThoughtLearningsOutput{}, err
+	}
+	return nil, GetThoughtLearningsOutput{Learnings: learnings}, nil
+}
+
+func (t *ThoughtLearningLinksTool) GetLearningThoughts(ctx context.Context, _ *mcp.CallToolRequest, in GetLearningThoughtsInput) (*mcp.CallToolResult, GetLearningThoughtsOutput, error) {
+	if err := t.ensureConfigured(); err != nil {
+		return nil, GetLearningThoughtsOutput{}, err
+	}
+
+	if in.LearningID <= 0 {
+		return nil, GetLearningThoughtsOutput{}, errRequiredField("learning_id")
+	}
+
+	if _, err := t.store.GetLearning(ctx, in.LearningID); err != nil {
+		return nil, GetLearningThoughtsOutput{}, errEntityNotFound("learning", "learning_id", strconv.FormatInt(in.LearningID, 10))
+	}
+
+	thoughts, err := t.store.GetLinkedThoughtsForLearning(ctx, in.LearningID)
+	if err != nil {
+		return nil, GetLearningThoughtsOutput{}, err
+	}
+	return nil, GetLearningThoughtsOutput{Thoughts: thoughts}, nil
+}
+
+func (t *ThoughtLearningLinksTool) ensureConfigured() error {
+	if t == nil || t.store == nil {
+		return errInvalidInput("thought learning links tool is not configured")
+	}
+	return nil
+}
@@ -39,3 +39,22 @@ type LinkedThought struct {
 	Direction string    `json:"direction"`
 	CreatedAt time.Time `json:"created_at"`
 }
+
+type ThoughtLearningLink struct {
+	ThoughtID  int64     `json:"thought_id"`
+	LearningID int64     `json:"learning_id"`
+	Relation   string    `json:"relation"`
+	CreatedAt  time.Time `json:"created_at"`
+}
+
+type LinkedLearning struct {
+	Learning  Learning  `json:"learning"`
+	Relation  string    `json:"relation"`
+	CreatedAt time.Time `json:"created_at"`
+}
+
+type LinkedLearningThought struct {
+	Thought   Thought   `json:"thought"`
+	Relation  string    `json:"relation"`
+	CreatedAt time.Time `json:"created_at"`
+}
@@ -0,0 +1,26 @@
+-- Many-to-many join table between thoughts and learnings.
+-- Allows a learning to reference multiple source thoughts and a thought to
+-- expose multiple related learnings, queryable in both directions.
+
+CREATE SEQUENCE IF NOT EXISTS public.identity_thought_learning_links_id
+  INCREMENT 1
+  MINVALUE 1
+  MAXVALUE 9223372036854775807
+  START 1
+  CACHE 1;
+
+CREATE TABLE IF NOT EXISTS public.thought_learning_links (
+  id bigint NOT NULL DEFAULT nextval('public.identity_thought_learning_links_id'),
+  thought_id bigint NOT NULL REFERENCES public.thoughts(id) ON DELETE CASCADE,
+  learning_id bigint NOT NULL REFERENCES public.learnings(id) ON DELETE CASCADE,
+  relation text NOT NULL DEFAULT 'source',
+  created_at timestamptz NOT NULL DEFAULT now(),
+  CONSTRAINT thought_learning_links_pkey PRIMARY KEY (id),
+  CONSTRAINT thought_learning_links_unique UNIQUE (thought_id, learning_id)
+);
+
+CREATE INDEX IF NOT EXISTS idx_thought_learning_links_thought_id
+  ON public.thought_learning_links (thought_id);
+
+CREATE INDEX IF NOT EXISTS idx_thought_learning_links_learning_id
+  ON public.thought_learning_links (learning_id);
@@ -32,6 +32,20 @@ Table thought_links {
  }
 }

+Table thought_learning_links {
+  id bigserial [pk]
+  thought_id bigint [not null, ref: > thoughts.id]
+  learning_id bigint [not null, ref: > learnings.id]
+  relation text [not null, default: `'source'`]
+  created_at timestamptz [not null, default: `now()`]
+
+  indexes {
+    (thought_id, learning_id) [unique]
+    thought_id
+    learning_id
+  }
+}
+
 Table embeddings {
  id bigserial [pk]
  guid uuid [unique, not null, default: `gen_random_uuid()`]
Author	SHA1	Message	Date
warkanum	8db2141d45	ci: build ui before go tests CI / build-and-test (push) Successful in 7m10s Details CI / build-and-test (pull_request) Successful in 7m13s Details	2026-07-14 16:32:00 +02:00
warkanum	3198600031	feat(tools): add duplicate audit report CI / build-and-test (push) Failing after 1m24s Details CI / build-and-test (pull_request) Failing after 59s Details	2026-07-14 16:28:06 +02:00
warkanum	5c899d1635	Merge pull request 'Link thoughts and learnings' (#36 ) from issue-32-link-thoughts-learnings into main CI / build-and-test (push) Failing after 1m6s Details Reviewed-on: #36 Reviewed-by: Warky <2+warkanum@noreply@warky.dev>	2026-07-14 14:42:05 +00:00
warkanum	81a3470407	feat(tools): link thoughts and learnings CI / build-and-test (pull_request) Failing after 1m2s Details CI / build-and-test (push) Failing after 3m10s Details	2026-07-14 15:18:57 +02:00