P3: meta + primitive tool group (think/now/cite + classify/extract/summarize)

Grow executus/tools into a real generic tool library:

- Register(reg): the always-available, zero-config tools — think, now (UTC
  unless a CurrentTimeProvider is wired), cite (inert unless a CitationStorage
  is wired). All nil-safe; a light host calls Register and is useful.
- RegisterMeta(reg, MetaDeps): the LLM-backed meta tools — classify,
  extract_entities, summarize — over the llmmeta helper. Budget defaults to the
  shipped in-memory per-run cap; Files optional; caps default.
- Seams moved (interface/type-only, no host coupling): research_providers.go
  (CurrentTimeProvider/CitationStorage/SearchBudget/PageExtractor/PDFFetcher/…)
  and file_storage.go (FileStorage + FileDomainMeta). Plus the in-memory budget
  default (research_defaults.go) and scope_validate.go.

calculate deferred (drags github.com/Krognol/go-wolfram + a module-path replace
— not worth it in the lean core for one tool). Core go.sum still free of
gorm/redis/discordgo/sqlite/wolfram.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tools/cite.go

@@ -0,0 +1,128 @@
+// Package tools — v11 cite.
+//
+// Anti-hallucination forcing function. The convention: agents call
+// cite(claim, url) for every numbered reference in their final
+// answer. The tool verifies the URL appears in the run's
+// touched-URL set (populated by web_search results +
+// read_page/read_pdf/read_video). If yes → write to
+// skill_run_sources, return {ok: true}. If no → return
+// {ok: false, reason: "url_not_in_run_history"} and DO NOT write.
+//
+// Skills authored without this discipline don't lose anything;
+// skills WITH it produce more reliable citations. The webui
+// renders the skill_run_sources rows as a Sources panel on the
+// run trace page — invisible to skills that don't use cite().
+package tools
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"gitea.stevedudenhoeffer.com/steve/executus/tool"
+)
+
+// citeParams is the LLM-facing param struct.
+type citeParams struct {
+	Claim string `json:"claim" description:"The claim or fact you are asserting (e.g. 'Mort was published in 1987')."`
+	URL   string `json:"url" description:"The URL that supports the claim. MUST be a URL the agent has previously read via read_page/read_pdf/read_video or seen as a web_search result."`
+}
+
+// citeResponse is the JSON envelope returned to the agent.
+//
+// On success: ok=true, the skill_run_sources row was written.
+// On failure: ok=false, reason=<one of the documented sentinels>.
+type citeResponse struct {
+	OK     bool   `json:"ok"`
+	Reason string `json:"reason,omitempty"`
+	Claim  string `json:"claim,omitempty"`
+	URL    string `json:"url,omitempty"`
+}
+
+// NewCite constructs the v11 cite tool. cs may be nil — handler
+// returns "not configured" at first call.
+//
+// The "anyone author / share-safe" permission shape matches every
+// other v11 research-class tool. Skills that adopt cite() get the
+// Sources panel automatically; skills that don't are unaffected.
+func NewCite(cs CitationStorage) tool.Tool {
+	return tool.NewGatedTool[citeParams](
+		"cite",
+		"Record a citation: a claim + the URL that supports it. The URL MUST be one the agent has actually fetched via read_page/read_pdf/read_video or seen as a web_search result — citing a URL the agent never visited is rejected with reason 'url_not_in_run_history'. Successful citations populate the run's Sources panel.",
+		tool.Permission{
+			AuthoringRequirement: tool.RequirementAnyone,
+			OperatesOn:           tool.ScopeGlobal,
+			SafeForShare:         true,
+			Categories:           []string{"citation"},
+		},
+		func(ctx context.Context, inv tool.Invocation, p citeParams) (string, error) {
+			if cs == nil {
+				return "", fmt.Errorf("cite: citation storage not configured")
+			}
+			claim := strings.TrimSpace(p.Claim)
+			urlStr := strings.TrimSpace(p.URL)
+			if claim == "" {
+				return marshalCite(citeResponse{
+					OK:     false,
+					Reason: "claim_empty",
+				}), nil
+			}
+			if urlStr == "" {
+				return marshalCite(citeResponse{
+					OK:     false,
+					Reason: "url_empty",
+				}), nil
+			}
+			if inv.RunID == "" {
+				// No run id → cite() can't verify history. Bail loud.
+				return marshalCite(citeResponse{
+					OK:     false,
+					Reason: "no_run_context",
+					Claim:  claim,
+					URL:    urlStr,
+				}), nil
+			}
+
+			touched, err := cs.GetTouchedURLs(ctx, inv.RunID)
+			if err != nil {
+				return marshalCite(citeResponse{
+					OK:     false,
+					Reason: fmt.Sprintf("touched_lookup_failed: %v", err),
+					Claim:  claim,
+					URL:    urlStr,
+				}), nil
+			}
+			if _, ok := touched[urlStr]; !ok {
+				return marshalCite(citeResponse{
+					OK:     false,
+					Reason: "url_not_in_run_history",
+					Claim:  claim,
+					URL:    urlStr,
+				}), nil
+			}
+
+			if err := cs.RecordCitation(ctx, inv.RunID, urlStr, claim); err != nil {
+				return marshalCite(citeResponse{
+					OK:     false,
+					Reason: fmt.Sprintf("record_failed: %v", err),
+					Claim:  claim,
+					URL:    urlStr,
+				}), nil
+			}
+			return marshalCite(citeResponse{
+				OK:    true,
+				Claim: claim,
+				URL:   urlStr,
+			}), nil
+		},
+	)
+}
+
+func marshalCite(r citeResponse) string {
+	b, err := json.Marshal(r)
+	if err != nil {
+		return fmt.Sprintf(`{"ok":false,"reason":"marshal_failed: %v"}`, err)
+	}
+	return string(b)
+}