Skip to content

DEMO.md

Latest commit

 

History

History
88 lines (61 loc) · 3.47 KB

DEMO.md

File metadata and controls

88 lines (61 loc) · 3.47 KB

DashClaw Demo: Market Intelligence Briefing

A full-stack demo that exercises every major DashClaw feature in one workflow execution.

What It Creates

Feature What's Seeded
Knowledge Collection 3 strategy documents (roadmap, competitors, markets)
Capabilities 5 real HTTP APIs at different risk levels
Policies 3 guard policies (auto-allow, warn, require approval)
Model Strategy Balanced analysis strategy (Claude Sonnet)
Workflow Template 5-step "Daily Market Briefing"

Prerequisites

  1. DashClaw running locally: npm run dev
  2. API key configured (check /setup page)

Run the Demo

Step 1: Seed demo data

node scripts/seed-demo-capabilities.mjs

This creates the knowledge collection, 5 capabilities, 3 policies, a model strategy, and the workflow template. Safe to re-run (idempotent).

Step 2: Execute the workflow

Open your DashClaw instance and navigate to Workflows. Find "Daily Market Briefing" and click Run.

Or execute via API:

curl -X POST http://localhost:3000/api/workflows/templates/<TEMPLATE_ID>/execute \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_KEY" \
  -d '{"agent_id": "demo-agent"}'

Step 3: Watch Mission Control

Open /mission-control and watch the operations feed in real-time:

  1. Knowledge Search — Searches your strategy docs (internal, no governance)
  2. HN News Fetch — Auto-allowed (risk 10, green)
  3. LLM Analysis — Runs via model strategy, produces the briefing
  4. Team Notification — Guard warns (risk 55, yellow) but proceeds
  5. Publish Briefing — Requires approval (risk 80, orange) — pauses here

Step 4: Approve or deny

Navigate to Approvals (or click the pending item in Mission Control). You'll see the publish action waiting for your decision:

  • Approve — The briefing is published to the demo endpoint (jsonplaceholder.typicode.com/posts) and the workflow completes
  • Deny — The publish step fails but the workflow still completes (continue_on_failure: true)

Step 5: Review the trail

Open Decisions to see the full audit trail: every guard evaluation, every action record, every artifact captured from each workflow step.

What This Exercises

DashClaw Feature How
Knowledge Collections Semantic search in Step 1
Capability Registry 5 HTTP APIs at different risk levels
Capability Invoke Steps 2, 4, 5 call real external APIs
Workflow Engine All 3 step types (knowledge_search, capability_invoke, prompt)
Workflow Variables ${steps.search_strategy.output} in Step 3
Guard Evaluation Risk 10 → allow, 55 → warn, 80 → require_approval
HITL Approvals Step 5 pauses for human decision
Model Strategies Step 3 uses configured analysis strategy
Artifacts Each step output auto-captured
Policies 3 threshold policies at different levels
continue_on_failure Workflow completes even if publish denied
Mission Control Real-time operations feed
Decisions Ledger Full audit trail

Troubleshooting

"Cannot reach DashClaw" — Make sure npm run dev is running and the URL is correct.

Knowledge search returns empty — Embeddings require an OpenAI key configured in org settings. The workflow still runs; the analysis step just won't have strategy context.

"Model strategy execution failed" — The prompt step requires a BYOK provider key (Anthropic or OpenAI) in org settings. Configure at /settings.