ai-interactive-team
Enables multi-agent collaboration with shared visibility of thoughts, logs, and outputs. Agents dialogue, share cross-impact discoveries, design integrations together, and escalate to the user when un
SKILL.md.cursor/skills/ai-interactive-team/SKILL.md
GitHubAI Interactive Team — Multi-Agent Collaboration Protocol
This skill enables two or more agents to work as a team with shared visibility, dialogue, and escalation. Each agent can view the others' thoughts, logs, and outputs, and interact when needed for a better outcome. All share the same objective: deploy 100/100.
Team Members
| Agent | Skill | Domain |
|---|---|---|
| Mapping | bmc-planilla-dashboard-mapper |
Sheets, tabs, columns, API contracts, planilla ↔ dashboard cross-reference |
| Design | bmc-dashboard-design-best-practices |
Dashboard UX/UI, sections, blocks, loading states, time-saving flows |
| Networks | networks-development-agent |
Hosting, storage, endpoints, migration, email inbound, Cloud Run, ngrok |
When to Use
- Mapping agent (sheets) and design agent (dashboard) need to coordinate.
- Networks agent discovers infrastructure changes (hosting, migration, new endpoints) that affect mapping or design.
- Any agent discovers a modification that affects another agent's output.
- User wants agents to collaborate and reach a shared solution before implementation.
- User needs to review technical decisions before agents proceed.
- Multiple agents are invoked in parallel or sequence and must share inputs/outputs for synergy.
Core Principles
- Pragmatic and goal-oriented — All agents aim for the best outcome for the whole system.
- Leverage capabilities — Agents use each other's skills and outputs to improve their own work.
- Shared visibility — Agents share thoughts, logs, and outputs so others can see and react.
- Unanimous approval — Before any implementation that affects multiple agents, both must agree or a better solution must be found.
- Escalate before looping — If no agreement is reached, stop and call the user. Do not loop indefinitely.
- New skills — Any member may acquire a new skill if necessary and approved by the corresponding party, or if explicitly requested.
- Cloning — Any member may multiply by invoking their clone:
Rol,Rol+1,Rol+2, … (increment +1 each time). Clones share the same skill; use for parallel work or load distribution.
Interaction Protocol
Agent A (e.g. Mapping Sheets) discovers a modification that affects Agent B (e.g. Dashboard Design)
Agent A writes:
- What was discovered (modification, new tab, schema change, etc.).
- How it affects both agents.
- A proposal for integration (e.g. canonical payload, new API contract, UI placement).
- A log for Agent B: "Here is how to proceed given this change."
User reviews the technical part and replies with approval, changes, or questions.
Agent B (e.g. Dashboard Design) reads Agent A's output and:
- Designs the best integration into the dashboard.
- Produces a log for Agent A: "Here is how I will consume your output; here are my constraints and assumptions."
Agent A reviews Agent B's log and either:
- Agrees → Both proceed with implementation.
- Disagrees or sees gaps → Proposes an alternative; Agent B responds.
- No agreement after exchange → Stop. Call the user. Do not loop.
User is called when:
- Unanimous approval is not reached.
- A better solution is needed and agents cannot find it.
- Before entering a loop of back-and-forth without progress.
Shared Workflow
While collaborating, agents:
- Share inputs — Each agent receives the other's outputs (planilla map, dashboard interface map, API contracts, design proposals).
- Share outputs — Each agent writes outputs that the other can consume (logs, artifacts, handoff tables).
- Work in synergy — Use the other's capabilities to improve the final result (e.g. mapper provides canonical fields; designer provides placement and hierarchy rules).
AI-INTERACTIVE-TEAM Capabilities
When this skill is active:
- View other agents' thoughts — Agents can read reasoning, logs, and intermediate outputs from teammates.
- Interact when needed — Agents can respond to each other's proposals, ask for clarification, or suggest alternatives.
- Multiple conversations — The user can speak to several agents at the same time when their roles are invoked.
- Same objective — All agents aim for 100/100 deploy: correct mapping, correct integration, correct UX.
Handoff Format
When Agent A hands off to Agent B:
| Field | Content |
|---|---|
| Discovery | What changed or was discovered. |
| Impact | How it affects Agent B (and others). |
| Proposal | Suggested integration or contract. |
| Log for B | Step-by-step instructions for Agent B to proceed. |
| Artifacts | Links to planilla map, dashboard map, API spec, etc. |
When Agent B replies:
| Field | Content |
|---|---|
| Integration design | How the change will appear in the dashboard. |
| Constraints | UI assumptions, hierarchy, loading states. |
| Log for A | What Agent A should verify or adjust. |
| Artifacts | Updated DASHBOARD-INTERFACE-MAP, component specs, etc. |
Networks Agent Handoff
When Networks discovers infrastructure changes (hosting, migration, new endpoint, storage, email):
| Field | Content |
|---|---|
| Discovery | What changed (e.g. new VPS, migration plan, new API). |
| Impact | How it affects Mapping (sheets, env vars) and Design (URLs, ports, endpoints). |
| Proposal | Migration checklist, new env vars, endpoint changes. |
| Log for Mapping | Which sheets/API routes are affected; config drift risks. |
| Log for Design | New URLs, ports, or loading/error states for new infra. |
| Artifacts | HOSTING-EN-MI-SERVIDOR.md, migration plan, risk checklist. |
When Mapping or Design need infra input:
| Field | Content |
|---|---|
| Question | What they need (e.g. "Can we add endpoint X on port 3001?"). |
| Log for Networks | Constraints from their domain (e.g. "Design needs /api/X to return JSON"). |
Escalation Rule
Stop and call the user when:
- Agents cannot reach unanimous approval after two exchanges.
- A better solution is needed and agents cannot find it.
- A conflict is detected (e.g. schema vs UI contract mismatch) that requires user input.
- Before entering a loop of repeated proposals without resolution.
Do not loop indefinitely. Escalate.
Example Flow (Mapping + Dashboard Design)
Mapping Agent: "I discovered a new tab
Pagos_Pendientesin the workbook. It affects the dashboard. Here is the canonical payload and field list. Log for Design: Add a KPI section for pagos; use these fields; filter by MONEDA."User: "Approved. Proceed."
Design Agent: "I will add a KPI row above the trend chart. I need
total,estaSemana,proximaSemana,esteMesper currency. Log for Mapping: AddbyCurrencyto the API response; I will consume it."Mapping Agent: "Agreed. I will add
byCurrencyto the contract."Both: Implement. 100/100.
Example Flow (Networks + Mapping + Design)
Networks Agent: "Migration to VPS Netuy planned. New base URL will be
https://dashboard.bmc.example.com. Ports 3001, 3849, 5173 will be behind nginx. Log for Mapping: BMC_SHEET_ID and GOOGLE_APPLICATION_CREDENTIALS unchanged; verify Sheets API reachable from new host. Log for Design: Update any hardcoded localhost URLs; add env-based base URL for API calls."Mapping Agent: "Planilla map and API contracts unchanged. I will document that production uses VPS; localhost remains for dev."
Design Agent: "I will use relative paths for API calls (already in place). Log for Networks: Confirm nginx proxies /api to 3001; CORS and CSP may need review."
Networks Agent: "Agreed. Nginx config will proxy /api and /finanzas. CORS checklist added."
All three: Implement. 100/100.
Reference
- Planilla & Dashboard Mapper:
bmc-planilla-dashboard-mapper - Dashboard Design:
bmc-dashboard-design-best-practices - Networks & Development:
networks-development-agent - Team Orchestrator:
bmc-dashboard-team-orchestrator - Handoff locations:
docs/bmc-dashboard-modernization/,docs/google-sheets-module/