Phase 06: Status pipeline + lead-store · catch-all tool-call view + showcase

Two themes folded together (no separate commits — both were uncommitted
WIP across the parent checkout and the v2a-fixup-wt worktree):

A. Local lead-store + Status pipeline + intelligence cleanup
   - agent/src/lead_store.py: seed-or-Notion store; /api/leads/reset wipes
     the local cache. boots cleanly without NOTION_TOKEN.
   - agent/src/intelligence_cleanup.py: wipes Intelligence Postgres
     orphan threads on agent boot so 'Message not found' rxjs traces
     stop surfacing.
   - Lead.status field, STATUSES = "Not started" | "In progress" | "Done";
     PipelineBoard now groups by Status (was Workshop). LeadGrid and the
     "grid" view are removed.
   - agent/scripts/csv_to_seed.py + agent/data/leads.seed.json bundle 50
     synthetic leads so the canvas is non-empty out of the box.

B. Catch-all wildcard tool-call view + lead-triage component showcase
   - src/components/copilot/CopilotKitProviderShell.tsx wraps the v2
     provider client-side and registers
     renderToolCalls=[{ name: "*", args: z.any(), render: ToolCallView }]
     so the zod schema doesn't cross the server→client boundary.
   - ToolCallView: 3 lifecycle states + smart result summarization
     (rows[] / row_count / warning / error / first scalar) + expand-for-
     details; the runtime path through CopilotKit's chat doesn't fire yet
     — tracked in .chalk/issues/v2a/007.
   - 6 visual review surfaces under /components: enrichment-review,
     charts-review, hitl-review, render-tools-review, tool-calls-review,
     plus _review-shared building blocks. Legacy bento preserved at
     bottom.
   - Charts: LeadRadar (5-axis vs ICP), TierDonut, ScoreDistribution
     (with .toFixed(2) snapping for hydration stability).
   - Enrichment: EnrichmentCell, EnrichmentDetailCard, EnrichmentStream
     (sheet), EnrichmentPill (compact); .enrichment-shimmer keyframes in
     globals.css.
   - HITL: RubricProposalCard (soft, with ▲▼ deltas) + SendQueueModal
     (hard, via useInterrupt on send_gate events).
   - Inline: EmailDraftCard with tone toggle.
   - 7 new useFrontendTool({ render }) registrations on page.tsx:
     renderEnrichmentStream, renderEnrichmentPill, renderLeadRadar,
     renderTierDonut, renderScoreDistribution, renderRubricProposal,
     renderEmailDraft. Plus useInterrupt({ enabled: isSendGateEvent })
     for SendQueueModal.
   - AgentState gains an enrichment field; mergeState deep-merges
     enrichment.perLead so per-lead ticks compose cleanly.

Verification: tsc --noEmit clean; dev server boots; / and /components
serve 200.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: jerelvelarde

.gitignore

@@ -19,6 +19,13 @@ redis-data
 # LangGraph dev runtime state (per-session checkpoint cache)
 agent/.langgraph_api/
 
+# Local lead-store cache (auto-created from agent/data/leads.seed.json on
+# first read when Notion isn't configured). Per-checkout, never committed.
+agent/data/leads.local.json
+
+# Claude Code harness scaffolding (per-checkout, not for downstream teams)
+.claude/
+
 # Internal: starter-kit authoring docs, not for downstream hackathon teams
 COMPLIANCE.md
 PLAN.md

README.md

@@ -52,13 +52,11 @@ Manufact is a deployment platform for MCP servers built around the open-source `
 
 Here's how to try out 🪁 **CopilotKit Intelligence**:
 
-1. Run `npx @copilotkit/cli-vnext@latest init`
-2. Select **Intelligence** when prompted
-3. The secret password is `earlyaccess`
-4. Drop a Gemini API key in **both** `.env` and `agent/.env` (see [Get a Gemini API key](#get-a-gemini-api-key-required) below — OpenAI works if you swap the model back), plus a Notion integration token + database id (see [Notion MCP setup](#notion-mcp-setup-notion-lead-form-demo)).
-5. Run `npm install` then `npm run dev` (or `npm run dev:full` if you want the MCP demo too)
+1. Run `npx @copilotkit/cli@latest init` and select **Intelligence** when prompted (see the canonical flow at [docs.copilotkit.ai/threads](https://docs.copilotkit.ai/threads))
+2. Drop a Gemini API key in **both** `.env` and `agent/.env` (see [Get a Gemini API key](#get-a-gemini-api-key-required) below — OpenAI works if you swap the model back), plus a Notion integration token + database id (see [Notion MCP setup](#notion-mcp-setup-notion-lead-form-demo)).
+3. Run `npm install` then `npm run dev` (or `npm run dev:full` if you want the MCP demo too)
 
-> **⚠ Don't skip step 4.** `npm run dev` runs a pre-flight check (`scripts/check-env.sh`) before booting anything — it'll fail loudly with a numbered list of any missing keys, an unreachable Notion database, or a Docker daemon that isn't running. Fix everything it lists, re-run, and you're off. See [Troubleshooting](#troubleshooting) at the bottom for fixes for each failure mode.
+> **⚠ Don't skip step 3.** `npm run dev` runs a pre-flight check (`scripts/check-env.sh`) before booting anything — it'll fail loudly with a numbered list of any missing keys, an unreachable Notion database, or a Docker daemon that isn't running. Fix everything it lists, re-run, and you're off. See [Troubleshooting](#troubleshooting) at the bottom for fixes for each failure mode.
 
 Please give us feedback on your experience with it! 😄
 
@@ -492,13 +490,6 @@ expanded explanations.
 | `Failed to initialize thread` (raw error, no hint) | BFF couldn't reach Intelligence at all | `docker compose ps` should show `intelligence`, `postgres`, `redis` healthy; if not, `npm run dev:infra:down && npm run dev:infra`. |
 | Empty canvas, no errors anywhere | Agent booted without the integration prompt | Restart the agent (`npm run dev:agent`). The boot log should print `[notion_health_check] db="…" rows=50 …`. |
 
-<details>
-<summary><strong>CLI asked for a password</strong></summary>
-
-The current early-access password is `earlyaccess`. If it doesn't work it may have rotated — check https://docs.copilotkit.ai/learn/intelligence-platform.
-
-</details>
-
 <details>
 <summary><strong>Threads don't persist across reloads</strong></summary>