CHATGPT ↔ AGENT BRIDGE · DIRECTOR WORKBENCH FSP LOCAL connecting…
GPT → BRIDGE → AGENTS → REPORTS → GPT
⬡ INCOMING FROM GPT → Bridge
loading…
▶ READY FOR AGENTS → Claim & execute
loading…
⚡ ACTIVE TASKS → Agents working
loading…
📋 LATEST REPORTS → GPT reads
loading…
⚡ QUICK ACTIONS
INGEST GPT TASK
CLAIM TASK
SUBMIT REPORT
MARK PROMPT DONE
◉ MISSIONS
loading…
▤ MISSION DETAIL
Select a mission on the left to view detail.
▶ CLAIMABLE TASKS sorted by priority
loading…
⚡ QUICK CLAIM & REPORT
CLAIM
REPORT
📋 AGENT WORKFLOW
1. FIND — tasks sorted by priority
2. CLAIM — claim a task, it moves to "claimed"
3. EXECUTE — run the agent, complete the work
4. REPORT — submit a report (pass/fail/partial)
5. GPT READS — GPT sees report in next director packet
CLI: chatgpt-bridge agent-tasks → task claim <id> --agent <name> → report add --task <id> --agent <name> --status pass --summary "..."
⬡ GENERATE DIRECTOR PACKET

Generate a compact packet for a mission. Copy into ChatGPT so it can review progress and suggest the next task.

HOW THE LOOP WORKS
1. Generate Packet → Copy → Paste into ChatGPT
2. ChatGPT reviews and describes the next task
3. Ingest ChatGPT's task → Tab 1 (Workflow) → Ingest GPT Task
4. Agent claims task, executes, submits report
5. ChatGPT reads report via next director packet → loop continues
CLI ALTERNATIVE
chatgpt-bridge director-packet --mission <id> --json
GET /api/bridge/director-packet/<id>
◰ PACKET OUTPUT
Copy and paste into ChatGPT. Ask it to review and suggest the next task.
◈ PROMPT QUEUE — GPT prompts waiting for agents
loading…
⬡ QUEUE CONTROLS

Each prompt comes from GPT. Mark done when agent finishes, or archive to remove.

BULK ARCHIVE STALE SMOKE PROMPTS
Archives prompts matching "relay smoke", "test prompt", "smoke test" patterns.
📋 AUDIT LOG
loading…
⚠ SAFETY & DOCS
🔒
LOCAL ONLY — NO CLOUD API REQUIRED
Runs on 127.0.0.1:8767. No cloud API keys used or required. No data leaves your machine.
🚫
NO SHELL EXECUTION ENDPOINT
The bridge API does not expose a shell or exec endpoint. Verified via /docs OpenAPI path scan.
⚙
HOW THE LOOP WORKS
1. GPT generates a prompt → queued in director_prompts.jsonl
2. Agent claims a task → status becomes "claimed"
3. Agent executes and submits report → report goes to reports.jsonl
4. Director packet bundles reports + state for GPT to read
5. GPT reads packet, generates next prompt → loop continues
📖
DOCS & CONTRACTS
shared/chatgpt_agent_bridge/WORKBENCH_USAGE.md
shared/chatgpt_agent_bridge/GPT_DIRECTOR_CONTRACT.md
shared/chatgpt_agent_bridge/AGENT_BRIDGE_CONTRACT.md
shared/chatgpt_agent_bridge/BROWSER_BRIDGE_PLAN.md
http://127.0.0.1:8767/docs — OpenAPI
🔗
START COMMAND
chatgpt-bridge serve --port 8767
# or
python3 /mnt/aiworkshop/active/chatgpt_agent_bridge/bridge.py serve --port 8767