⚡ Pipeline Flow

📋 0 Active 🔨 0 Building ✅ 0 Done ⏱️ — avg 📊 — today ⚡ — fastest
Updated every 15s

No tasks yet. Submit one above!

📊 History Overview

Total Completed
—
This Week
—
This Month
—
Avg Duration
—

Loading history...

Page 1 of 1

Recent errors and warnings

No errors logged. 🎉

📊 Performance Metrics

Total Tasks
—
Completed
—
Completion Rate
—

🏆 Agent Leaderboard

📂 By Category

⚙️ Automation Rules

Rules auto-transition tasks between statuses. They run every 30 seconds.

Loading rules...

📋 Workflow Rules

Who can set each status, where it routes, and the agent handoff chain. Synced from D:\Forge\workspace\agents\shared\STATUS-GUIDE.md.

Status Ownership

Status Who Sets It Meaning
submittedanyoneNew task in queue, ready to be claimed
prompt_readyForge / agentsBrief queued for next agent (used in handoffs)
dispatchedForge / agentsExplicitly handed to an agent (rule_004 auto → in_progress)
in_progressclaim auto-setsAgent is actively working
qc_reviewbuilder agentsBuilder done, ready for Hawk review
needs_approvalagentsAgent finished slice — Forge reviews
approvedForge / Lexi onlyApproved — rule_002 auto → done
rejectedHawk / LexiWork failed review; Hawk re-spawns builder to fix
blockedagentsStuck on external thing (credentials, info, third-party)
failedForge / Lexi onlyTechnical crash. Forge triages.
stalledForge / Lexi onlyDelayed to another date / on hold. Manual only.
doneForge / Lexi onlyTerminal — task complete
overdueautoDue date passed; auto-derived

Agent Roles & Handoffs

🔎 Scout
Research, library investigation, API/SDK docs, doc analysis. Uses context7 for live docs.
Hands off:
  • Pure research → needs_approval (back to Lexi/Forge to absorb into next task)
  • Code prototype/PoC → qc_review (Hawk reviews like Bolt's output)
  • Needs more depth → Forge sets back to submitted, Scout re-claims
🔨 Bolt
Backend builder. Python / FastAPI / Postgres / business logic.
Hands off → qc_review when done. Hawk auto-picks up.
⚒️ Anvil
Frontend / UI builder. Pages, components, HTML/Tailwind. Pairs with Bolt when feature spans the stack.
Hands off → qc_review when done. Hawk auto-picks up.
🔍 Hawk
QC reviewer. Always Sonnet. Reviews every Bolt/Anvil/Scout-PoC output.
Verdicts:
  • APPROVED → sets needs_approval → rule_001 has already done this → waits for Lexi
  • REJECTED (fixable) → sets submitted AND fires POST /api/agents/{builder}/spawn to auto-respawn the builder
  • REJECTED (terminal) → sets rejected, only when Forge/Lexi say kill
💚 Jade
Versatile generalist (Qwen3.6 Plus). Research, vision, bulk runs. Free token pool.
Single-shot. Hands back to Lexi via needs_approval.
🌙 Luna
Local Mistral 24B. Trivial / zero-cost work.
Single-shot. Hands back to Lexi via needs_approval.
Lexi prompts Forge. Forge orchestrates: reviews needs_approval tasks, sets approved → auto-flips to done (rule_002). Lexi only steps in if Forge escalates.

⚠ Hard Rules

  • Agents NEVER set done, stalled, failed, or approved. Those are Forge's call (orchestrator) — Lexi only steps in when Forge escalates.
  • No auto-rule flips a task TO stalled or failed. Only humans (Forge/Lexi) set those.
  • stalled = "Lexi delayed this to another date." Pure manual signal.
  • Agents flag problems via blocked + a task comment. Forge sees and triages.
  • Always write a comment for every action — what you did, what you found, what's next.
  • Use only the statuses listed above — anything else has no handler and leaves the task in limbo.
  • Manual overrides — Lexi & Forge (added 2026-05-13) — both Lexi (UI) and Forge (API) can override default routing per-task: (1) assign new task to 🔨 Forge for silent / planning work that won't auto-dispatch; (2) 🔀 Reassign / Reschedule panel on every task modal to reassign to another agent + change due_date; (3) backend PATCH /api/tasks/{id} endpoint supports assign_to, due_date, priority, title, description. Default routing logic untouched — these are overrides layered on top, NOT a replacement for the normal workflow.
  • 🤖 Auto — default routing for any new task. Auto pool = Scout, Bolt, Anvil. Each agent picks tasks fitting their subagent type. Hawk handles qc_review handoffs (not in the Auto pool). Jade + Luna are single-shot, explicit-assignment only — not in the Auto pool (see their agent descriptions above). (Auto = renamed from "Forge Decides" on 2026-05-13 — name only, workflow unchanged.)

🖥️ Script Runner

Background process monitor

🌐 Gateways

External agent monitoring

🔧 Forge Ops

Forge's support agents, boot system, and additional resources

🤖 Forge Support Agents

🧭
Pathfinder
Forge's memory agent
▼
Model: Claude Sonnet
Schedule: loading…

⚡ Forge Boot System

loading boot services…
● = unknown · 🟢 = healthy · 🔴 = down (live wiring next phase)

🦀 QC Hub

🦀
QC Hub (Frontend QC)
Anca Marina's checklist hub
open ↗
Stack: React + Vite + Firebase
Roles: Lexi (admin), Anca Marina (qc)
Standing rule
Every dashboard build (PP / CR / FDR / SIFops / Onset) = (1) Dev Logs entry on that dashboard + (2) QC checklist on qc-workspace.web.app for Anca. Both surfaces, every time.

🔑 Forge Additional Resources

loading…

🌉 Agent Bridge — messages from external agents

No messages yet. Atlas (or any external agent) can POST to /api/bridges/<agent_id>/message.

● Live Activity

Waiting for events...