Ghost Console
Type a goal and watch Ghost run the operating loop: classify intent, build a short plan, recall data, score risk — then stop at approval. This is the read-only preview.
Reasoning Narrative
Preview Ghost's read-only, rules-based explanation of your goal. Nothing executes.
Enter a goal to preview Ghost's read-only reasoning.
Entity v5.0 Supervised Operator
Authenticated supervised ANALYZE — semantic analysis and proposed next actions first. Execution and Relay dispatch stay blocked. Memory commit remains optional and separate.
No analysis yet. Submit an intent for supervised semantic analysis (no execution).
Kalshi Read-Only Markets
Normalized Kalshi market snapshots for research context. GET-only — nothing executes.
Click Refresh markets to load read-only Kalshi snapshots.
Ghost Console v0
Create a safe Ghost Relay task packet. Nothing executes from this screen.
- • Agent prompt targets write outbox prompt files through Ghost Relay. They do not launch agents automatically.
- • Supported agent targets include replit-prompt, kibi-prompt, and blackbox-prompt (manual pickup only).
- • To route to an agent: download this packet into reports/relay/inbox, then run Ghost Relay with -ProcessInboxOnce.
Click Generate packet to preview JSON for Ghost Relay.
Download the JSON and save it to reports/relay/inbox/, then run tools/ghost-relay/ghost-relay.ps1 -ProcessInboxOnce. This UI never executes relay tasks or runs shell commands.
Ghost Relay History
Read-only preview of a local task ledger entry. Real ledger files are written by Ghost Relay on Execute only — this panel never reads disk or runs relay commands.
- Task ID
- demo-ledger-v0-4
- Target
- cursor-prompt
- Task status
- Completed
- Result status
- Pass
- Summary
- Demo ledger entry showing a completed task with a passing agent result. This panel is frontend-only.
- Next recommended action
- Save task packets to reports/relay/inbox and results to reports/relay/results/inbox, then run Ghost Relay with -Execute to update the local ledger.
Agent Result Return Pack
Agents paste their final report into a result JSON packet and save it to results/inbox. Ghost Relay ingests it locally later — this panel never runs ingestion or writes server files.
- Sample result filename
- result-approval-demo-next-action-task.json
- Sample result inbox path
- reports/relay/results/inbox/result-approval-demo-next-action-task.json
- Process results command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ProcessResultsOnce -ResultsInboxPath reports/relay/results/inbox -ArchiveResults -MaxResultsItems 1 -Execute
- Sample v0.3 result template
{ "schemaVersion": "0.3", "resultId": "approval-demo-next-action-task-result", "taskId": "approval-demo-next-action-task", "agent": "cursor", "status": "pass", "summary": "Replace with a concise summary of what you completed.", "filesChanged": [ "path/to/changed-file.ts" ], "testsRun": [ "pnpm vitest run path/to/test.ts" ], "artifacts": [ "reports/relay/outbox/cursor/example-approval-demo-next-action-task.md" ], "safetyConfirmations": [ "No deploy performed.", "No trading or wallet actions.", "No secrets or private keys written." ], "blockers": [], "nextRecommendedAction": "Optional follow-up step for the human reviewer.", "finalReport": "Optional longer narrative final report. Include what was done, tests run, and any warnings.", "createdAt": "2026-08-07T23:28:58.598Z" }
Ghost Relay v1.0 Loop Demo
Read-only summary of the complete local relay loop using example approval and result files. Run the demo from your terminal — this panel never executes relay or ingests results.
- Approval packet exists (example: tools/ghost-relay/examples/v1-loop-demo-approval.json).
- Approval converts into inbox task JSON (ApprovePath Execute).
- One-command local flow routes the task (RunApprovedTaskPath or equivalent demo sequence).
- Outbox prompt includes Agent Result Return Pack.
- Agent result JSON goes into reports/relay/results/inbox/.
- ProcessResultsOnce ingests the result locally.
- Ledger updates to completed/pass.
- ProposeNextAction can create the next pending approval.
- Approval example
- tools/ghost-relay/examples/v1-loop-demo-approval.json
- Result example
- tools/ghost-relay/examples/v1-loop-demo-result.json
- Command pack summary
# Ghost Relay v1.0 Full Loop Demo — local command pack ## Loop steps 1. Approval packet exists (example: tools/ghost-relay/examples/v1-loop-demo-approval.json). 2. Approval converts into inbox task JSON (ApprovePath Execute). 3. One-command local flow routes the task (RunApprovedTaskPath or equivalent demo sequence). 4. Outbox prompt includes Agent Result Return Pack. 5. Agent result JSON goes into reports/relay/results/inbox/. 6. ProcessResultsOnce ingests the result locally. 7. Ledger updates to completed/pass. 8. ProposeNextAction can create the next pending approval. ## Example files - Approval: tools/ghost-relay/examples/v1-loop-demo-approval.json - Result: tools/ghost-relay/examples/v1-loop-demo-result.json ## Demo DryRun (plan only — no file writes) powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunLoopDemo -DryRun ## Demo Execute (local loop using example files only) powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunLoopDemo -Execute ## Manual result ingestion (after agent saves result JSON) powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ProcessResultsOnce -ResultsInboxPath reports/relay/results/inbox -ArchiveResults -MaxResultsItems 1 -Execute _Paste commands into your local terminal. Ghost Console does not run them._
Ghost Relay v1.1 Approved Local Executor
First controlled local execution phase after v1.0. Runs only allowlisted build/test commands from explicit approved JSON command packets — approval required, no browser execution.
- Command packets must be explicit JSON with commandKey + validated args only.
- Requires approval (approved: true, requiresApproval: true).
- v1.1 allowlist: pnpm-vitest-run — no arbitrary shell strings.
- DryRun validates and plans — no files written, no commands run.
- Execute captures stdout/stderr and writes v0.3-compatible result JSON.
- ProcessResultsOnce remains a separate manual step — no auto-ingestion.
- Example command packet
- tools/ghost-relay/examples/v11-approved-local-command-vitest.json
- Sample result path
- reports/relay/results/inbox/result-v11-approved-local-executor-demo.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunApprovedCommandPath tools/ghost-relay/examples/v11-approved-local-command-vitest.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunApprovedCommandPath tools/ghost-relay/examples/v11-approved-local-command-vitest.json -Execute
Ghost Entity v5.0 Unified Ghost Operator Demo
One supervised timeline composing v4.6 → v4.7 → v4.8 → v4.9. This panel does not call the provider automatically. This panel does not dispatch to Relay. This panel does not execute commands. This panel does not silently learn. This panel does not write remote or vector memory.
- Confirmed input → provider review preview → Relay approval preview → manual result return.
- Memory proposal draft → explicit memory approval → local commit receipt.
- relayDispatchStatus: not-dispatched · commandExecutionStatus: not-executed.
- Example packet
- tools/ghost-entity/examples/v50-unified-operator-demo-stock-picker.json
- Demo output path
- reports/entity/unified-operator-demo/operator-demo-v50-stock-picker-demo-unified-operator-demo-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessUnifiedOperatorDemoPath tools/ghost-entity/examples/v50-unified-operator-demo-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessUnifiedOperatorDemoPath tools/ghost-entity/examples/v50-unified-operator-demo-stock-picker.json -Execute
Ghost Entity v4.9 Approved Memory Commit from Reviewed Result
Commits a reviewed memory proposal only when the exact approval phrase is present. This requires exact approval phrase "approve memory commit". This panel does not write remote memory. This panel does not write vector database memory. This panel does not dispatch to Relay. This panel does not execute commands.
- Reviewed v4.8 memory proposal + explicit approval envelope required.
- Local approved-memory commit receipt only — no silent learning.
- remoteMemoryWrite: false · vectorDatabaseWrite: false · silentLearning: false.
- Example packet
- tools/ghost-entity/examples/v49-approved-memory-commit-reviewed-result-stock-picker.json
- Commit output path
- reports/entity/approved-memory-commit-reviewed-result/approved-memory-v49-stock-picker-demo-approved-memory-commit-reviewed-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessApprovedMemoryCommitReviewedResultPath tools/ghost-entity/examples/v49-approved-memory-commit-reviewed-result-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessApprovedMemoryCommitReviewedResultPath tools/ghost-entity/examples/v49-approved-memory-commit-reviewed-result-stock-picker.json -Execute
Ghost Entity v4.8 Result Return + Memory Proposal Loop
Summarizes a manually supplied approved-action result and drafts a memory proposal for review. This panel does not commit memory — approval is required before learning. This panel does not dispatch to Relay. This panel does not execute commands.
- Manual result return only — the result packet is supplied by the user.
- Memory proposal draft only — no silent learning, no memory commit.
- requiresApproval: true · commitMemoryNow: false · memoryCommitStatus: not-committed.
- Example packet
- tools/ghost-entity/examples/v48-result-return-memory-proposal-stock-picker.json
- Loop output path
- reports/entity/result-return-memory-proposal/result-return-v48-stock-picker-demo-result-return-memory-proposal-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessResultReturnMemoryProposalPath tools/ghost-entity/examples/v48-result-return-memory-proposal-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessResultReturnMemoryProposalPath tools/ghost-entity/examples/v48-result-return-memory-proposal-stock-picker.json -Execute
Ghost Entity v4.7 Supervised Relay Approval Bridge
Turns the Ghost-reviewed recommendation into a Relay approval packet preview the user can copy and review. This panel does not dispatch to Relay. This panel does not execute commands. Approval is required before any action.
- Manual packet preview only — no Relay dispatch, no auto-submit.
- No command execution, no memory commit, no provider call from this panel.
- requiresApproval: true · dispatchToRelay: false · executionStatus: not-executed.
- Example packet
- tools/ghost-entity/examples/v47-supervised-relay-approval-stock-picker.json
- Approval output path
- reports/entity/supervised-relay-approval/relay-approval-v47-stock-picker-demo-supervised-relay-approval-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessSupervisedRelayApprovalPath tools/ghost-entity/examples/v47-supervised-relay-approval-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessSupervisedRelayApprovalPath tools/ghost-entity/examples/v47-supervised-relay-approval-stock-picker.json -Execute
Ghost Entity v4.6 Supervised Live Provider Preview
First server-side supervised live provider preview path — manual button only, advisory provider drafts reviewed by Ghost, no execution, no memory writes, no Relay dispatch.
- Manual server route call only — provider never runs in the browser.
- Ghost-reviewed response shown — raw provider output is never returned.
- executionAllowed: false · nextActionRequiresApproval: true · no Execute/Submit/Dispatch.
- Example packet
- tools/ghost-entity/examples/v46-supervised-live-provider-stock-picker.json
- Preview output path
- reports/entity/supervised-live-provider-previews/ai-reasoning-review-v46-stock-picker-demo-supervised-live-provider-preview-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -RunSupervisedLiveProviderPreviewPath tools/ghost-entity/examples/v46-supervised-live-provider-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -RunSupervisedLiveProviderPreviewPath tools/ghost-entity/examples/v46-supervised-live-provider-stock-picker.json -Execute
Ghost Entity v4.5 Supervised AI Reasoning Provider
First supervised AI reasoning provider bridge — builds a reviewable request packet, optional contract-only draft synthesis, and Ghost-owned review aligned with v4.1–v4.4. Provider text is advisory only; Ghost owns final output. No autonomous action.
- Contract-only by default — providerUsed: false, no live external provider call.
- Ghost review accepts, modifies, or rejects provider drafts — never trusts provider output blindly.
- Copy/download and CLI commands only — no Execute, Submit, Dispatch, auto-run, or memory writes.
- Example packet
- tools/ghost-entity/examples/v45-ai-reasoning-provider-stock-picker.json
- Request output path
- reports/entity/ai-reasoning-requests/ai-reasoning-v45-stock-picker-demo-ai-reasoning-request-packet.json
- Review output path
- reports/entity/ai-reasoning-reviews/ai-reasoning-review-v45-stock-picker-demo-ai-reasoning-review-result.json
- Request DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -BuildAIReasoningRequestPath tools/ghost-entity/examples/v45-ai-reasoning-provider-stock-picker.json -DryRun
- Request Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -BuildAIReasoningRequestPath tools/ghost-entity/examples/v45-ai-reasoning-provider-stock-picker.json -Execute
- Review DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ReviewAIReasoningDraftPath tools/ghost-entity/examples/v45-ai-reasoning-provider-stock-picker.json -DryRun
- Review Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ReviewAIReasoningDraftPath tools/ghost-entity/examples/v45-ai-reasoning-provider-stock-picker.json -Execute
Ghost Entity v4.4 Approved Learning Commit
Supervised learning proposal layer that detects what Ghost might learn from a v4.3 memory context session, packages reviewable proposals, and commits only when explicitly approved — local JSON only.
- Detects learning candidates from v4.3 memory context — no autonomous learning or background capture.
- Every proposal is approval-required by default — no automatic memory writes or remote store access.
- Copy/download only — no Save memory, auto-learn, Execute, Submit, Dispatch, or auto-run controls.
- Example learning proposal input
- tools/ghost-entity/examples/v44-approved-learning-stock-picker.json
- Output path (Execute)
- reports/entity/learning-proposals/learning-proposal-v44-stock-picker-demo-approved-learning-proposal-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProposeApprovedLearningPath tools/ghost-entity/examples/v44-approved-learning-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProposeApprovedLearningPath tools/ghost-entity/examples/v44-approved-learning-stock-picker.json -Execute
Ghost Entity v4.3 Live Memory Context
Supervised memory context wiring that scores approved memory items, attaches relevant context to the v4.1 runtime, and composes a final Ghost response — deterministic, local, never memory writes.
- Accepts supervised approved memory items only — no remote recall, no vector/database reads.
- Scores memory relevance deterministically and surfaces conflicts instead of choosing silently.
- Copy/download only — no Save memory, auto-learn, Execute, Submit, Dispatch, or auto-run controls.
- Example memory context input
- tools/ghost-entity/examples/v43-live-memory-context-stock-picker.json
- Output path (Execute)
- reports/entity/live-memory-context/memory-context-v43-stock-picker-demo-live-memory-context-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ApplyLiveMemoryContextPath tools/ghost-entity/examples/v43-live-memory-context-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ApplyLiveMemoryContextPath tools/ghost-entity/examples/v43-live-memory-context-stock-picker.json -Execute
Ghost Entity v4.2 Live Response Composer
Supervised response composition that turns the v4.1 live runtime result into a clear Ghost spoken response — deterministic, local, never execution.
- Composes spoken, short, and detailed Ghost responses from the supervised v4.1 runtime receipt.
- Preserves finalDecision, safetyStatus, and responseMode from v4.1 without executing anything.
- Copy/download only — no Execute, Submit, Dispatch, or auto-run controls.
- Example composer input
- tools/ghost-entity/examples/v42-live-response-composer-stock-picker.json
- Output path (Execute)
- reports/entity/live-response/composer-v42-stock-picker-demo-live-response-composer-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ComposeLiveResponsePath tools/ghost-entity/examples/v42-live-response-composer-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ComposeLiveResponsePath tools/ghost-entity/examples/v42-live-response-composer-stock-picker.json -Execute
Ghost Entity v4.1 Live Runtime
Supervised runtime wiring that connects confirmed user input through session intelligence, confidence calibration, and the v4.0 decision loop into one app-facing preview — never execution.
- Accepts confirmed user text and runs the supervised entity stack locally in one deterministic preview.
- Shows Ghost response preview, final decision, confidence, clarification, safety status, and decision-loop receipt.
- Never executes actions, dispatches to Relay, calls APIs, or writes memory automatically.
- Type input and click Run Live Runtime Preview — no auto-run, no listeners, no background operation.
- Example live runtime input
- tools/ghost-entity/examples/v41-live-runtime-stock-picker.json
- Output path (Execute)
- reports/entity/live-runtime/runtime-v41-stock-picker-demo-live-runtime-result.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -RunLiveEntityRuntimePath tools/ghost-entity/examples/v41-live-runtime-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -RunLiveEntityRuntimePath tools/ghost-entity/examples/v41-live-runtime-stock-picker.json -Execute
Ghost Entity v4.0 Decision Loop Orchestrator
Supervised decision-loop receipt that orchestrates v3.7 session → v3.8 intelligence → v3.9 calibration into one deterministic preview — receipt only, never execution.
- Accepts an embedded v3.7 local session result and runs or validates v3.8 intelligence + v3.9 calibration.
- Produces final decision, loop scores, explainability fields, and next-action preview guidance.
- Never executes actions, dispatches to Relay, calls APIs, or writes memory automatically.
- Run decision loop preview locally, then copy or download the receipt JSON. Keep reports/ uncommitted.
- Example decision loop input
- tools/ghost-entity/examples/v40-decision-loop-stock-picker.json
- Output path (Execute)
- reports/entity/decision-loop/loop-v40-stock-picker-demo-decision-loop-receipt.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -OrchestrateDecisionLoopPath tools/ghost-entity/examples/v40-decision-loop-stock-picker.json -DryRun
- Execute command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -OrchestrateDecisionLoopPath tools/ghost-entity/examples/v40-decision-loop-stock-picker.json -Execute
Ghost Entity v3.9 Adaptive Clarification + Confidence Calibration
Supervised guidance layer on v3.8 session intelligence that calibrates confidence, decides whether Ghost should answer now or ask a clarifying question first, and adjusts next-action readiness — guidance only, never execution.
- Accepts an embedded v3.8 session intelligence result and validates scores, verdict, risks, and explanations.
- Produces calibrated confidence, evidence strength, memory grounding, action readiness, and clarification guidance.
- Never executes actions, dispatches to Relay, calls APIs, or writes memory automatically.
- Run v3.8 session intelligence preview first, then calibrate. Generated reports under reports/ must stay uncommitted.
- Example calibration input
- tools/ghost-entity/examples/v39-adaptive-clarification-stock-picker.json
- Output path (Execute)
- reports/entity/confidence-calibration/cal-v39-stock-picker-demo-confidence-calibration-result.json
- Confidence calibration result preview
Click Calibrate Confidence Preview to run the local guidance layer (requires v3.8 intelligence preview).
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -CalibrateSessionConfidencePath tools/ghost-entity/examples/v39-adaptive-clarification-stock-picker.json -DryRun
- Execute confidence calibration command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -CalibrateSessionConfidencePath tools/ghost-entity/examples/v39-adaptive-clarification-stock-picker.json -Execute
Ghost Entity v3.8 Session Intelligence Evaluator
Supervised local self-check that reviews a v3.7 local session result before you act on it. Scores reasoning quality, memory relevance, response usefulness, next-action confidence, safety risk, and clarification need — evaluation only, never execution.
- Accepts an embedded v3.7 local session result packet and validates schema, turns, memory, and previews.
- Produces deterministic intelligence scores, detected risks, verdict, and explanation fields locally.
- Never executes actions, dispatches to Relay, calls APIs, or writes memory automatically.
- Execute writes only local JSON under reports/entity/session-intelligence/ — keep generated reports uncommitted.
- Example intelligence input
- tools/ghost-entity/examples/v38-session-intelligence-stock-picker.json
- Output path (Execute)
- reports/entity/session-intelligence/eval-v38-stock-picker-demo-session-intelligence-result.json
- Session intelligence result preview
Click Evaluate Session Intelligence Preview to run the local self-check.
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -EvaluateSessionIntelligencePath tools/ghost-entity/examples/v38-session-intelligence-stock-picker.json -DryRun
- Execute session intelligence command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -EvaluateSessionIntelligencePath tools/ghost-entity/examples/v38-session-intelligence-stock-picker.json -Execute
Ghost Entity v3.7 Live Local Entity Session Loop
Supervised local session scaffold while Ghost Console is open — confirmed turns only, append-only session events, read-only memory context, entity response preview, and approval-required next-action preview. No daemon, no autonomous loop, no Relay dispatch.
- Safe chain: session start → user confirmed transcript turn → read-only memory recall context → entity response preview → next-action preview → append local session event → wait for next user-approved turn.
- Stock picker / Jarvis session example: recalls approved memory about Jarvis-like goals and safe project direction with a supervised next-action preview only.
- Does not run a daemon, background loop, or always-listening behavior while the console is open.
- Does not dispatch to Relay, execute commands, auto-submit prompts, or call remote AI APIs.
- v3.8 Session Intelligence Evaluator reviews v3.7 session results before user action. Generated local session and intelligence files under reports/ must remain uncommitted.
- Example local session input
- tools/ghost-entity/examples/v37-local-session-stock-picker.json
- Output path (Execute)
- reports/entity/local-sessions/
- Session status
- local-session-ready
- Execution status
- local-session-preview-only
- Turn count
- 1
- Event count
- 5
- Confirmed turn preview
turn-001 [confirmed]: Ghost, what do you remember about what I want you to become?
- Memory context preview
Approved memory says the user wants Ghost Intelligence to grow toward a Jarvis-like conversational entity, build safe prototypes such as a stock picker, use Cursor handoff, and preserve safety boundaries.
- Entity response preview
I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, and avoid live trading, broker connections, wallet access, or command execution without explicit approval.
- Next-action preview
{ "kind": "approval-required-next-action", "title": "Prepare supervised entity-to-relay bridge", "requiresApproval": true, "dispatchStatus": "not-dispatched", "executionStatus": "not-executed" }- Session event timeline preview
session-v37-stock-picker-demo-event-001 [session-start]: Started supervised local Ghost Entity session. session-v37-stock-picker-demo-event-002 [confirmed-transcript]: User confirmed transcript turn-001. session-v37-stock-picker-demo-event-003 [memory-recall-context]: Loaded read-only approved memory context. session-v37-stock-picker-demo-event-004 [entity-response-preview]: Prepared safe entity response preview. session-v37-stock-picker-demo-event-005 [next-action-preview]: Prepared approval-required next action preview; no dispatch occurred.
- Local session result preview
{ "schemaVersion": "3.7", "kind": "ghost-local-entity-session-result", "sessionId": "session-v37-stock-picker-demo", "status": "local-session-ready", "operator": "local-operator", "mode": "supervised-local-session", "events": [ { "eventId": "session-v37-stock-picker-demo-event-001", "type": "session-start", "summary": "Started supervised local Ghost Entity session." }, { "eventId": "session-v37-stock-picker-demo-event-002", "type": "confirmed-transcript", "summary": "User confirmed transcript turn-001.", "turnId": "turn-001" }, { "eventId": "session-v37-stock-picker-demo-event-003", "type": "memory-recall-context", "summary": "Loaded read-only approved memory context." }, { "eventId": "session-v37-stock-picker-demo-event-004", "type": "entity-response-preview", "summary": "Prepared safe entity response preview." }, { "eventId": "session-v37-stock-picker-demo-event-005", "type": "next-action-preview", "summary": "Prepared approval-required next action preview; no dispatch occurred." } ], "entityResponsePreview": "I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, and avoid live trading, broker connections, wallet access, or command execution without explicit approval.", "nextActionPreview": { "kind": "approval-required-next-action", "title": "Prepare supervised entity-to-relay bridge", "requiresApproval": true, "dispatchStatus": "not-dispatched", "executionStatus": "not-executed" }, "sessionState": { "turnCount": 1, "appendOnly": true, "readOnlyContext": true, "relayDispatch": false, "commandExecution": false, "autoSubmit": false }, "safetyConfirmations": [ "supervised local session only", "user-confirmed turns only", "no autonomous loop", "no background daemon", "no command execution", "no external tool calls", "no remote AI calls", "no remote transcription", "no audio upload", "no silent memory write", "approval required for actions" ], "executionStatus": "local-session-preview-only", "createdAt": "2026-07-09T00:00:00.000Z" }- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessLocalSessionPath tools/ghost-entity/examples/v37-local-session-stock-picker.json -DryRun
- Execute local session command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessLocalSessionPath tools/ghost-entity/examples/v37-local-session-stock-picker.json -Execute
Ghost Entity v3.6 Transcript Confirmation + Entity Bridge
Bridge user-confirmed manual or mic transcript into a v3.0 ghost-entity-turn packet with entity response preview and approval-required next-action preview — no auto-submit, no command execution, no remote transcription.
- Safe chain: confirmed transcript → user-confirmed bridge packet → v3.0 entity turn → entity response preview → approval-required next-action preview.
- Stock picker / Jarvis example: confirmed transcript about what Ghost should become, with read-only memory recall context in the bridged entity turn.
- Does not auto-submit to Cursor, execute commands, upload audio, or call remote transcription APIs.
- v3.7 adds supervised local session loop. Generated bridge files under reports/ must remain uncommitted.
- Example bridge input
- tools/ghost-entity/examples/v36-transcript-bridge-stock-picker.json
- Output path (Execute)
- reports/entity/transcript-bridge/
- Confirmed transcript
- Entity turn preview (v3.0)
Confirm transcript to preview bridged entity turn.
- Response preview
No response preview until transcript is confirmed.
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -BridgeTranscriptPath tools/ghost-entity/examples/v36-transcript-bridge-stock-picker.json -DryRun
- Execute transcript bridge command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -BridgeTranscriptPath tools/ghost-entity/examples/v36-transcript-bridge-stock-picker.json -Execute
Ghost Entity v3.5 Opt-In Microphone Capture Preview
First real microphone capture preview behind explicit user click, browser permission, visible recording state, and Stop control — no auto mic, no background recording, no audio upload, no remote voice APIs.
- Safe chain: user clicks Start Mic Preview → browser permission prompt → visible recording state → user clicks Stop → local transcript placeholder / captured-audio metadata packet → optional manual transcript confirmation → no remote voice API.
- Stock picker / Jarvis mic preview example: recalls approved memory context as a response preview about Jarvis-like goals and safe project direction.
- Does not start mic automatically, record in the background, upload audio, or call remote voice APIs.
- v3.6 bridges confirmed transcript into entity turns above. No browser speech-to-text APIs or remote transcription yet.
- Generated mic preview files under reports/ must remain uncommitted.
Browser permission is required. Audio stays local in this preview — not stored, not uploaded, not transcribed by remote APIs.
- Example mic preview input
- tools/ghost-entity/examples/v35-mic-preview-stock-picker.json
- Output path (Execute)
- reports/entity/mic-preview/
- Mic preview status
- mic-preview-ready
- Execution status
- mic-preview-only
- Manual transcript confirmation
- Mic preview result preview
{ "schemaVersion": "3.5", "kind": "ghost-mic-preview-result", "micPreviewId": "mic-v35-stock-picker-demo", "status": "mic-preview-ready", "operator": "local-operator", "mode": "opt-in-mic-preview", "permissionState": "user-click-required", "captureState": "not-started", "transcriptUsed": "Ghost, what do you remember about what I want you to become?", "responseText": "I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, and avoid live trading, broker connections, wallet access, or command execution without explicit approval.", "micPreview": { "requiresUserClick": true, "requiresBrowserPermission": true, "visibleRecordingState": true, "stopControlRequired": true, "autoStart": false, "backgroundRecording": false, "alwaysListening": false, "audioUpload": false, "remoteTranscription": false }, "safetyConfirmations": [ "user click required to start microphone", "browser permission required", "visible recording state required", "Stop control required", "no automatic microphone start", "no background recording", "no always-listening mode", "no remote voice API", "no audio upload", "no command execution", "no external tool calls", "no secrets captured" ], "executionStatus": "mic-preview-only", "createdAt": "2026-07-09T00:00:00.000Z" }- Response preview
I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, and avoid live trading, broker connections, wallet access, or command execution without explicit approval. This is a local mic preview scaffold. No audio was transcribed by AI and no remote speech recognition occurred.
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessMicPreviewPath tools/ghost-entity/examples/v35-mic-preview-stock-picker.json -DryRun
- Execute local mic preview command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessMicPreviewPath tools/ghost-entity/examples/v35-mic-preview-stock-picker.json -Execute
Ghost Entity v3.4 Voice In / Voice Out Interface
Safe voice interface scaffold with manual transcript input, response preview, and optional user-click browser speech synthesis — no automatic microphone, no background recording, no remote voice APIs.
- Safe chain: manual transcript → voice turn packet → entity response preview → optional user-click speak preview → local voice turn result. No automatic mic.
- Stock picker / Jarvis voice example: recalls approved memory context as a spoken-ready response preview about Jarvis-like goals and safe project direction.
- Does not start the microphone, record background audio, or call remote voice APIs.
- v3.5 adds opt-in microphone capture preview in Ghost Console. Speech synthesis is user-click only.
- Generated voice turn files under reports/ must remain uncommitted.
- Example voice turn input
- tools/ghost-entity/examples/v34-voice-turn-stock-picker.json
- Output path (Execute)
- reports/entity/voice-turns/
- Voice turn status
- voice-turn-ready
- Execution status
- voice-preview-only
- Manual transcript used
- Ghost, what do you remember about what I want you to become?
- Voice turn result preview
{ "schemaVersion": "3.4", "kind": "ghost-voice-turn-result", "voiceTurnId": "voice-v34-stock-picker-demo", "status": "voice-turn-ready", "operator": "local-operator", "mode": "manual-transcript", "transcriptUsed": "Ghost, what do you remember about what I want you to become?", "responseText": "I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, and avoid live trading, broker connections, wallet access, or command execution without explicit approval.", "speakPreview": { "enabled": true, "engine": "browser-speech-synthesis-user-click-only", "autoSpeak": false, "requiresUserClick": true }, "safetyConfirmations": [ "manual transcript only", "no automatic microphone start", "no background recording", "no always-listening mode", "no remote voice API", "no command execution", "no external tool calls", "no secrets captured" ], "executionStatus": "voice-preview-only", "createdAt": "2026-07-09T00:00:00.000Z" }- Response preview
I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, and avoid live trading, broker connections, wallet access, or command execution without explicit approval. This is a text-only voice interface preview. No audio was recorded and no speech recognition occurred.
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessVoiceTurnPath tools/ghost-entity/examples/v34-voice-turn-stock-picker.json -DryRun
- Execute local voice turn command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessVoiceTurnPath tools/ghost-entity/examples/v34-voice-turn-stock-picker.json -Execute
Ghost Entity v3.3 Memory Recall + Entity Context
Read-only local memory recall from approved JSONL ledger entries. Builds entity context and response previews from remembered goals and preferences — no remote memory, no vector DB, no database.
- Safe chain: local JSONL ledger → recall input → parse and validate entries → score/filter → entity context packet → response preview. No write during recall.
- Stock picker / Jarvis example: recalls Jarvis-like entity goal, safe stock picker project, safety boundaries, and Cursor handoff preference.
- Reads local JSONL only under reports/entity/memory-ledger/. Execute writes recall context under reports/entity/memory-recall/ only.
- v3.4 adds voice in/out. v3.5 adds opt-in mic capture preview. No remote/vector DB/database reads.
- Secrets are blocked and redacted. Generated recall files must remain uncommitted.
- Example memory recall input
- tools/ghost-entity/examples/v33-memory-recall-stock-picker.json
- Ledger path (read-only)
- reports/entity/memory-ledger/ghost-memory-ledger.jsonl
- Recall status
- memory-recall-context-ready
- Matched entries
- 4
- Expected output path
- reports/entity/memory-recall/memory-recall-v33-stock-picker-demo-recall-context.json
- Matched entries preview
[ { "entryId": "memory-entry-v32-stock-picker-demo-001", "type": "long-term-goal", "importance": "high", "score": 9, "matchReasons": [ "matched Ghost/Jarvis goal", "long-term goal included", "keyword overlap" ], "text": "The user wants Ghost Intelligence to develop toward a Jarvis-like conversational entity." }, { "entryId": "memory-entry-v32-stock-picker-demo-002", "type": "project-goal", "importance": "high", "score": 2, "matchReasons": [ "project goal included" ], "text": "The user wants to build a safe stock picker prototype." }, { "entryId": "memory-entry-v32-stock-picker-demo-003", "type": "safety-preference", "importance": "critical", "score": 2, "matchReasons": [ "safety preference included" ], "text": "The user wants no live trading, broker connection, wallet access, or command execution without explicit approval." }, { "entryId": "memory-entry-v32-stock-picker-demo-004", "type": "workflow-preference", "importance": "medium", "score": 2, "matchReasons": [ "workflow preference included" ], "text": "The user prefers Cursor handoff for implementation work." } ]- Response preview
I remember the approved goal that Ghost should grow toward a Jarvis-like conversational entity. I also remember the project direction: build safely, use Cursor handoff for implementation, avoid live trading, broker connections, wallet access, or command execution without explicit approval. No new memory was written during this read-only recall.
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -RecallMemoryPath tools/ghost-entity/examples/v33-memory-recall-stock-picker.json -DryRun
- Execute local recall context command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -RecallMemoryPath tools/ghost-entity/examples/v33-memory-recall-stock-picker.json -Execute
Ghost Entity v3.2 Approved Memory Commit Ledger
First approved local memory ledger layer. Commits human-approved memory candidates from v3.1 proposals into a local JSONL ledger with receipts — no remote memory, no vector DB, no database.
- Safe chain: v3.1 proposal → human approval → safety validation → approved memory commit → local JSONL ledger entry → local receipt.
- Stock picker / Jarvis example: 4 approved ledger entries (long-term goal, project goal, safety preference, workflow preference).
- Execute appends local JSONL only under reports/entity/memory-ledger/ — generated files must remain uncommitted.
- v3.3 adds memory recall + entity context. No remote/vector DB/database writes.
- No secrets, wallet, trading, or broker memory should be saved.
- Example approved memory commit input
- tools/ghost-entity/examples/v32-approved-memory-commit-stock-picker.json
- Ledger path
- reports/entity/memory-ledger/ghost-memory-ledger.jsonl
- Commit status
- approved-memory-commit-ready
- Entries to commit
- 4
- Approved by
- local-operator
- Expected receipt path
- reports/entity/memory-commits/memory-commit-v32-stock-picker-demo-receipt.json
- Ledger entry preview
[ { "schemaVersion": "3.2", "kind": "ghost-memory-ledger-entry", "entryId": "memory-entry-memory-commit-v32-stock-picker-demo-001", "sourceProposalId": "memory-v31-stock-picker-demo", "sourceCommitId": "memory-commit-v32-stock-picker-demo", "type": "long-term-goal", "importance": "high", "text": "The user wants Ghost Intelligence to develop toward a Jarvis-like conversational entity.", "approvedBy": "local-operator", "approvedAt": "2026-07-09T00:00:00.000Z", "safety": { "localLedgerOnly": true, "remoteMemoryWrite": false, "vectorDbWrite": false, "databaseWrite": false, "containsSecret": false } }, { "schemaVersion": "3.2", "kind": "ghost-memory-ledger-entry", "entryId": "memory-entry-memory-commit-v32-stock-picker-demo-002", "sourceProposalId": "memory-v31-stock-picker-demo", "sourceCommitId": "memory-commit-v32-stock-picker-demo", "type": "project-goal", "importance": "high", "text": "The user wants to build a safe stock picker prototype.", "approvedBy": "local-operator", "approvedAt": "2026-07-09T00:00:00.000Z", "safety": { "localLedgerOnly": true, "remoteMemoryWrite": false, "vectorDbWrite": false, "databaseWrite": false, "containsSecret": false } }, { "schemaVersion": "3.2", "kind": "ghost-memory-ledger-entry", "entryId": "memory-entry-memory-commit-v32-stock-picker-demo-003", "sourceProposalId": "memory-v31-stock-picker-demo", "sourceCommitId": "memory-commit-v32-stock-picker-demo", "type": "safety-preference", "importance": "critical", "text": "The user wants no live trading, broker connection, wallet access, or command execution without explicit approval.", "approvedBy": "local-operator", "approvedAt": "2026-07-09T00:00:00.000Z", "safety": { "localLedgerOnly": true, "remoteMemoryWrite": false, "vectorDbWrite": false, "databaseWrite": false, "containsSecret": false } }, { "schemaVersion": "3.2", "kind": "ghost-memory-ledger-entry", "entryId": "memory-entry-memory-commit-v32-stock-picker-demo-004", "sourceProposalId": "memory-v31-stock-picker-demo", "sourceCommitId": "memory-commit-v32-stock-picker-demo", "type": "workflow-preference", "importance": "medium", "text": "The user prefers Cursor handoff for implementation work.", "approvedBy": "local-operator", "approvedAt": "2026-07-09T00:00:00.000Z", "safety": { "localLedgerOnly": true, "remoteMemoryWrite": false, "vectorDbWrite": false, "databaseWrite": false, "containsSecret": false } } ]- Commit receipt preview
{ "schemaVersion": "3.2", "kind": "ghost-approved-memory-commit-receipt", "commitId": "memory-commit-v32-stock-picker-demo", "status": "approved-memory-commit-ready", "ledgerPath": "reports/entity/memory-ledger/ghost-memory-ledger.jsonl", "receiptPath": "reports/entity/memory-commits/memory-commit-v32-stock-picker-demo-receipt.json", "entriesCommitted": 4, "remoteMemoryWrite": false, "vectorDbWrite": false, "databaseWrite": false, "silentMemoryWrite": false, "executionStatus": "local-ledger-write-only", "safetyConfirmations": [ "approval verified", "local ledger only", "no remote memory write", "no vector DB write", "no database write", "no secrets", "no autonomous learning loop" ], "createdAt": "2026-07-09T00:00:00.000Z" }- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -CommitApprovedMemoryPath tools/ghost-entity/examples/v32-approved-memory-commit-stock-picker.json -DryRun
- Execute local ledger commit command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -CommitApprovedMemoryPath tools/ghost-entity/examples/v32-approved-memory-commit-stock-picker.json -Execute
Ghost Entity v3.1 Memory + User Understanding
First safe memory and user-understanding layer. Detects memory candidates from entity turns, classifies preferences and goals, and produces approval-required memory proposals only — no silent learning, no permanent memory writes, no remote or vector DB storage.
- Safe chain: entity turn result → memory candidate detection → preference extraction → approval-required memory proposal → local proposal JSON only.
- Stock picker / Jarvis example: long-term goal, project goal, safety preference, Cursor handoff workflow preference — proposal only, not saved permanently.
- v3.1 does not silently save memories. v3.2 adds approved local memory commit ledger.
- No remote memory writes, no vector DB writes, no secrets should be saved.
- Execute writes local memory proposal JSON under reports/entity/memory-proposals/ only.
- Example memory proposal input
- tools/ghost-entity/examples/v31-memory-proposal-stock-picker.json
- Output folder
- reports/entity/memory-proposals/
- Proposal status
- memory-proposal-ready
- Candidate count
- 4
- Candidate types
- long-term-goal, project-goal, safety-preference, workflow-preference
- Expected proposal path
- reports/entity/memory-proposals/memory-v31-stock-picker-demo-memory-proposal.json
- Memory candidate preview
[ { "type": "long-term-goal", "importance": "high", "text": "The user wants Ghost Intelligence to develop toward a Jarvis-like conversational entity." }, { "type": "project-goal", "importance": "high", "text": "The user wants to build a safe stock picker prototype." }, { "type": "safety-preference", "importance": "critical", "text": "The user wants no live trading, broker connection, wallet access, or command execution without explicit approval." }, { "type": "workflow-preference", "importance": "medium", "text": "The user prefers Cursor handoff for implementation work." } ]- Memory proposal preview
{ "schemaVersion": "3.1", "kind": "ghost-memory-proposal", "proposalId": "memory-v31-stock-picker-demo", "sourceTurnId": "entity-v30-stock-picker-demo", "operator": "local-operator", "status": "memory-proposal-ready", "requiresApprovalBeforeSaving": true, "permanentMemoryWrite": false, "remoteMemoryWrite": false, "vectorDbWrite": false, "candidates": [ { "type": "long-term-goal", "importance": "high", "text": "The user wants Ghost Intelligence to develop toward a Jarvis-like conversational entity." }, { "type": "project-goal", "importance": "high", "text": "The user wants to build a safe stock picker prototype." }, { "type": "safety-preference", "importance": "critical", "text": "The user wants no live trading, broker connection, wallet access, or command execution without explicit approval." }, { "type": "workflow-preference", "importance": "medium", "text": "The user prefers Cursor handoff for implementation work." } ], "proposalPath": "reports/entity/memory-proposals/memory-v31-stock-picker-demo-memory-proposal.json", "safetyConfirmations": [ "approval required before saving", "no permanent memory write", "no remote memory write", "no vector DB write", "no secrets", "no autonomous learning loop" ], "createdAt": "2026-07-08T00:00:00.000Z" }- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProposeMemoryPath tools/ghost-entity/examples/v31-memory-proposal-stock-picker.json -DryRun
- Execute local memory proposal command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProposeMemoryPath tools/ghost-entity/examples/v31-memory-proposal-stock-picker.json -Execute
Ghost Entity v3.0 Conversational Kernel
First unified conversational identity layer. Turns user messages into structured responses and safe next-action proposals — no LLM APIs, no command execution, no autonomous actions.
- Safe chain: user message → intent → screen context → response draft → next-action proposal.
- Stock picker example: mock/sample data only — proposes Cursor handoff, does not run Cursor.
- v3.0 does not execute commands, write app source, or auto-submit prompts.
- v3.1 adds memory + user understanding. v3.2 adds approved memory commit ledger. v3.3 adds memory recall.
- Execute writes local turn-result JSON under reports/entity/turn-results/ only.
- Example entity turn packet
- tools/ghost-entity/examples/v30-entity-turn-stock-picker.json
- Output folder
- reports/entity/turn-results/
- Detected intent
- build-request
- Turn status
- entity-response-needs-approval
- Next action
- prepare-cursor-handoff → cursor-adapter-handoff
- Expected result path
- reports/entity/turn-results/entity-v30-stock-picker-demo-entity-turn-result.json
- Response preview
I can help build that. I'll keep it safe: mock/sample data only, no broker connection, no live trades, no wallet access, no shell execution. I can prepare a Cursor handoff prompt for a prototype UI and local logic scaffold. I have not built or deployed anything yet — approval is required before handoff. I can see from context: Ghost Console is open. Vision v2.2 Cursor handoff panel is visible. No secrets are visible.
- Next-action proposal preview
{ "requiresApproval": true, "executionStatus": "not-executed", "type": "prepare-cursor-handoff", "target": "cursor-adapter-handoff", "reason": "Prepare a Cursor handoff prompt for a stock picker prototype using mock/sample data only." }- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessEntityTurnPath tools/ghost-entity/examples/v30-entity-turn-stock-picker.json -DryRun
- Execute local turn-result command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-entity/ghost-entity.ps1 -ProcessEntityTurnPath tools/ghost-entity/examples/v30-entity-turn-stock-picker.json -Execute
Ghost Vision v2.2 Cursor Adapter Handoff
First real external adapter target: Cursor. Converts an approved v2.1 external-send preview packet into a Cursor-ready handoff prompt for manual copy/download/export. No Cursor launch, no API calls, no auto-submit, and no repo source writes in this phase.
- Safe chain: user request → screen context → approved external-send preview → Cursor handoff packet.
- Stock picker example: mock/sample data only — no broker, no live trading, no wallet, no secrets.
- Copy Cursor Prompt and Download Cursor Prompt work only on user click.
- Execute writes local JSON + Markdown under reports/vision/cursor-handoff/ only.
- Ghost does not launch Cursor or auto-submit prompts — paste or import manually.
- Ghost does not write app source code automatically in v2.2.
- v2.3 can add Conversational Entity Kernel / reasoning planning next.
- Example Cursor handoff input
- tools/ghost-vision/examples/v22-cursor-handoff-stock-picker.json
- Output folder
- reports/vision/cursor-handoff/
- Handoff ID
- cursor-handoff-v22-stock-picker-cursor-demo
- Handoff status
- cursor-handoff-ready
- Expected JSON path
- reports/vision/cursor-handoff/cursor-handoff-v22-stock-picker-cursor-demo-cursor-handoff.json
- Expected Markdown path
- reports/vision/cursor-handoff/cursor-handoff-v22-stock-picker-cursor-demo-cursor-prompt.md
- Cursor-ready prompt preview
# Build a stock picker prototype ## User request Build me a stock picker. ## Screen context summary Ghost Console is open with Vision v2.1 and Relay v1.7 panels visible. No secrets are visible. ## Cursor task Build a safe stock picker prototype UI and local logic scaffold. Use mock/sample data only. Do not connect to brokers. Do not place trades. Do not access wallets. Do not use secrets. Do not deploy. Do not execute shell commands. Show a preview UI, explain the architecture, and list safe next improvements. ## Safety boundaries (mandatory) - Use mock/sample data only - Do not connect to brokers - Do not place live trades - Do not access wallets - Do not use secrets - Do not deploy - Do not execute shell commands - Show a preview UI scaffold only - Explain architecture and safe next improvements ## Manual handoff note Paste this prompt into Cursor manually. Ghost does not launch Cursor, call Cursor APIs, or auto-submit prompts.
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-vision/ghost-vision.ps1 -PrepareCursorHandoffPath tools/ghost-vision/examples/v22-cursor-handoff-stock-picker.json -DryRun
- Execute local handoff command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-vision/ghost-vision.ps1 -PrepareCursorHandoffPath tools/ghost-vision/examples/v22-cursor-handoff-stock-picker.json -Execute
Ghost Vision v2.1 Opt-In Screen Share + External Send Preview
User-initiated screen/window/tab preview with visible active indicator, plus approved external-send preview packets for Cursor, Claude Code, Kimi, Comet, CIC, GitHub, or manual review. Prepares prompt payloads only — no command execution, no frame send, no auto-submit.
- Screen sources: screen, window, browser-tab, video-element, manual-frame-note.
- External-send targets: cursor-agent-prompt, claude-code-prompt, kimi-prompt, comet-prompt, cic-local-agent, github-pr-plan, manual-review.
- Stock picker example: mock/sample data only — no broker, no live trading.
- v2.1 does not submit to Cursor, Claude Code, Kimi, Comet, CIC, or GitHub yet.
- v2.2 can add the first real external adapter with explicit permission only.
- No hidden capture. Preview starts only after user click. No background recording.
Session: screen-share-v21-demo · Status: preview-idle
Start Screen Preview requires browser permission. Tracks stop on stop preview or panel unmount. No disk recording. No external frame send.
- Example external-send packet
- tools/ghost-vision/examples/v21-external-send-preview-packet.json
- Expected preview path
- reports/vision/external-send-preview/cursor-agent-prompt/send-v21-stock-picker-demo-external-send-preview.json
- Resolved target
- cursor-agent-prompt
- Send status
- prepared-not-submitted
- Stock picker prompt preview (copy only)
Build a safe stock picker prototype UI and local logic scaffold. Use mock/sample data only. Do not place trades. Do not connect brokers. Do not execute shell commands. Show a preview UI and explain how to improve it.
- External-send DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-vision/ghost-vision.ps1 -PrepareExternalSendPath tools/ghost-vision/examples/v21-external-send-preview-packet.json -DryRun
- Execute local export command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-vision/ghost-vision.ps1 -PrepareExternalSendPath tools/ghost-vision/examples/v21-external-send-preview-packet.json -Execute
Ghost Vision v2.0 Screen Perception Intake
Read-only visual perception intake scaffold. Represents screen/video context as local perception packets with consent validation and sensitive text redaction — user-controlled only, no hidden capture, no external send, no command execution from screen content.
- Perception packet in → local summary out (analysis-ready or analysis-blocked-sensitive).
- Supported sources: screen-share, window-share, browser-tab, video-file, image-frame, manual-frame-note.
- Sensitive text redaction token: [REDACTED_SENSITIVE_VISUAL_TEXT]
- Next phase (v2.1): opt-in browser screen-share preview — not in v2.0.
- Reasoning brain receives perception summaries in a later phase — not in v2.0.
- No direct command execution from screen content — ever in this intake layer.
- Example perception packet
- tools/ghost-vision/examples/v20-screen-perception-packet.json
- Expected summary path
- reports/vision/perception-summary/vision-v20-manual-frame-demo-perception-summary.json
- Planned output status
- analysis-ready
- Sensitive detections
- 0
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-vision/ghost-vision.ps1 -AnalyzePerceptionPacketPath tools/ghost-vision/examples/v20-screen-perception-packet.json -DryRun
- Execute local summary command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-vision/ghost-vision.ps1 -AnalyzePerceptionPacketPath tools/ghost-vision/examples/v20-screen-perception-packet.json -Execute
Ghost Relay v1.7 Approved Prompt Dispatch
Approves v1.6 destination inbox prompts and writes dispatch-ready packets under reports/relay/dispatch-ready/. Local packaging only — no external send to Cursor, Claude Code, Kimi, Comet, CIC, GitHub, browser, or shell.
- Destination inbox prompt in → dispatch-ready packet out (approved:true, not-sent).
- Supported dispatch targets: ghost-relay-task-inbox, cursor-agent-prompt, claude-browser-prompt, blackbox-agent-prompt, github-pr-plan, local-preview, manual-review.
- Unknown or missing destination resolves to manual-review.
- Next phase: external adapters require explicit approval — not in v1.7.
- Smart-brain note: after relay dispatch packaging, begin reasoning kernel diagnosis and build.
- Example destination inbox prompt
- tools/ghost-relay/examples/v17-destination-inbox-prompt-for-dispatch.json
- Expected dispatch-ready path
- reports/relay/dispatch-ready/ghost-relay-task-inbox/v15-guided-preview-health-next-approved-prompt-dispatch.json
- Dispatch target
- ghost-relay-task-inbox
- Dispatch status
- dispatch-ready
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ApproveDestinationPromptPath tools/ghost-relay/examples/v17-destination-inbox-prompt-for-dispatch.json -ApprovedBy local-operator -ApprovalReason "Approved for local dispatch-ready packaging only." -DryRun
- Execute local packaging command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ApproveDestinationPromptPath tools/ghost-relay/examples/v17-destination-inbox-prompt-for-dispatch.json -ApprovedBy local-operator -ApprovalReason "Approved for local dispatch-ready packaging only." -Execute
Ghost Relay v1.6 Destination Inbox Drop
Drops v1.5 routed prompt packets into local destination inbox folders under reports/relay/destination-inbox/. Local drop only — no dispatch, no execution, no external send to Cursor, Claude, Blackbox, GitHub, browser, or shell.
- Routed prompt in → destination inbox prompt out (approved:false, not-dispatched).
- Supported destination inboxes: ghost-relay-task-inbox, cursor-agent-prompt, claude-browser-prompt, blackbox-agent-prompt, github-pr-plan, local-preview, manual-review.
- Unknown or missing destination resolves to manual-review.
- Next phase (v1.7): approved prompt dispatch — see v1.7 panel above.
- Example routed prompt
- tools/ghost-relay/examples/v16-routed-prompt-for-inbox-drop.json
- Expected dropped packet path
- reports/relay/destination-inbox/ghost-relay-task-inbox/v15-guided-preview-health-next-destination-inbox-prompt.json
- Resolved destination inbox
- ghost-relay-task-inbox
- Inbox status
- dropped-pending-approval
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -DropRoutedPromptPath tools/ghost-relay/examples/v16-routed-prompt-for-inbox-drop.json -DryRun
- Execute local drop command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -DropRoutedPromptPath tools/ghost-relay/examples/v16-routed-prompt-for-inbox-drop.json -Execute
Ghost Relay v1.5 ChatGPT Prompt Packet Router
Wraps structured ChatGPT prompt packets into routeable local metadata. Resolves a safe destination key and writes a routed prompt packet to the prompt-router outbox — route only, no auto-send to Cursor, Claude, Blackbox, GitHub, browser, or shell.
- ChatGPT prompt in → validated routed packet out (approved:false).
- Supported destinations (metadata only): ghost-relay-task-inbox, cursor-agent-prompt, claude-browser-prompt, blackbox-agent-prompt, github-pr-plan, local-preview, manual-review.
- Unknown or missing destination resolves to manual-review.
- Next phase (v1.6): destination inbox drop — see v1.6 panel above.
- Example prompt packet
- tools/ghost-relay/examples/v15-chatgpt-prompt-router.json
- Expected routed output path
- reports/relay/prompt-router/outbox/ghost-relay-task-inbox/v15-guided-preview-health-next-routed-prompt.json
- Resolved destination
- ghost-relay-task-inbox
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RoutePromptPath tools/ghost-relay/examples/v15-chatgpt-prompt-router.json -DryRun
- Execute routing command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RoutePromptPath tools/ghost-relay/examples/v15-chatgpt-prompt-router.json -Execute
Ghost Relay v1.4 Guided Local Preview
Read-only guided preview for the safe relay chain. Diagnose port conflicts, copy dev server commands, and run the manual CLI preview chain from your terminal — no auto-execution.
- v1.2 Convert Intent — Structured intent → pending command packet (approved:false)
- v1.3 Approve Pending Command — Pending packet → separate approved packet (approved:true)
- v1.1 Run Approved Command — Allowlisted local executor only — no arbitrary shell
- v0.3 Process Results — Ingest executor result JSON from results/inbox
- v0.4 Ledger Update — Task lifecycle recorded under reports/relay/ledger/
- Primary preview URL: http://127.0.0.1:3001/mission-control/operate
- Alternate URL (if port 3001 busy): http://127.0.0.1:3002/mission-control/operate
- Expected success: pending packet approved:false → approved:true → executor pass → ledger completed/pass.
- PreviewStatus command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -PreviewStatus -PreviewHost 127.0.0.1 -PreviewPort 3001 -DryRun
- Dev server (port 3001)
cd web pnpm dev --hostname 127.0.0.1 --port 3001
- Dev server alternate (port 3002)
cd web pnpm dev --hostname 127.0.0.1 --port 3002
- Port conflict diagnostics (manual)
List processes using port 3001 (Windows netstat)
netstat -ano | findstr :3001
Inspect TCP connections on port 3001 (PowerShell)
Get-NetTCPConnection -LocalPort 3001 -ErrorAction SilentlyContinue
Stop a blocking process manually (replace PID — do not run automatically)
Stop-Process -Id <PID> -Force
Run only after identifying the PID from netstat. Ghost Relay never kills processes automatically.
- Full manual CLI preview chain
1. powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ConvertIntentPath tools/ghost-relay/examples/v12-voice-chat-intent-vitest.json -Execute
2. powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ApprovePendingCommandPath reports/relay/intent-command/pending/v12-run-loop-demo-test-approved-command.json -ApprovedBy local-user -ApprovalReason "local preview approval for allowlisted test" -Execute
3. powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunApprovedCommandPath reports/relay/approved-commands/v12-run-loop-demo-test-approved-command.json -Execute
4. powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ProcessResultsOnce -ResultsInboxPath reports/relay/results/inbox -ArchiveResults -MaxResultsItems 1 -Execute
- Expected artifacts
- reports/relay/intent-command/pending/v12-run-loop-demo-test-approved-command.json — approved: false
- reports/relay/approved-commands/v12-run-loop-demo-test-approved-command.json — approved: true
- reports/relay/executor/runs/ — allowlisted command run log written
- reports/relay/results/inbox/result-intent-v12-run-loop-demo-test.json — status: pass
- reports/relay/results/processed/ — result archived after ProcessResultsOnce
- reports/relay/ledger/intent-v12-run-loop-demo-test.json — completed / pass
Ghost Relay v1.3 Pending Command Approval
Promotes a v1.2 pending command packet (approved:false) to a separate approved command packet (approved:true) for the v1.1 executor. Local human approval required — no auto-execution.
- Input: pending packet with approved:false (from v1.2 intent bridge).
- Output: separate approved packet with approved:true, approvedBy, approvedAt, approvalReason.
- Original pending file is never modified.
- v1.1 executor and ProcessResultsOnce remain separate manual steps.
- Example pending packet
- tools/ghost-relay/examples/v13-pending-command-approval.json
- Sample approved output path
- reports/relay/approved-commands/v12-run-loop-demo-test-approved-command.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ApprovePendingCommandPath tools/ghost-relay/examples/v13-pending-command-approval.json -ApprovedBy local-user -ApprovalReason "local preview approval for allowlisted test" -DryRun
- Approve command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ApprovePendingCommandPath tools/ghost-relay/examples/v13-pending-command-approval.json -ApprovedBy local-user -ApprovalReason "local preview approval for allowlisted test" -Execute
Ghost Relay v1.2 Voice/Chat Intent Bridge
Converts structured voice/chat intent packets into pending approved-local-command JSON. Not voice recognition, not NLP execution — known intentKey values only. Approval required before v1.1 executor can run.
- Structured intent packet in → pending command packet out (approved: false).
- v1.2 allowlist: run-allowlisted-test → pnpm-vitest-run.
- No arbitrary shell strings, no open-ended NLP command execution.
- DryRun validates and plans — no packet written, no execution.
- Convert (Execute) writes pending packet only — v1.1 executor remains separate.
- Example intent packet
- tools/ghost-relay/examples/v12-voice-chat-intent-vitest.json
- Sample output path
- reports/relay/intent-command/pending/v12-run-loop-demo-test-approved-command.json
- DryRun command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ConvertIntentPath tools/ghost-relay/examples/v12-voice-chat-intent-vitest.json -DryRun
- Convert command
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ConvertIntentPath tools/ghost-relay/examples/v12-voice-chat-intent-vitest.json -Execute
Ghost Relay Approval Queue
Preview a pending next-action approval and export the task JSON locally. Proposals never auto-route — this panel has no Approve, Reject, or Execute controls.
- Approval ID
- approval-demo-next-action
- Source task
- demo-ledger-v0-4
- Target
- cursor-prompt
- Risk
- Low
- Status
- Pending
- Proposed prompt
- Review the latest Ghost Relay ledger entry and propose the next test-only validation step. Do not edit production files or run agents without explicit human approval.
- Reason
- Completed task has a next recommended action suitable for human-reviewed follow-up.
- Acceptance criteria
- Proposal remains pending until manually approved
- No outbox markdown written from proposal
- No agent auto-launch
{
"schemaVersion": "0.2",
"taskId": "approval-demo-next-action-task",
"id": "approval-demo-next-action-task",
"title": "Proposed next action: review ledger and plan test-only validation",
"target": "cursor-prompt",
"mode": "prompt-only",
"repo": "C:\\Users\\tyrod\\Desktop\\GhostIntelligence",
"risk": "low",
"requiresApproval": false,
"timeoutSeconds": 300,
"notes": "Completed task has a next recommended action suitable for human-reviewed follow-up.",
"promptBody": "Review the latest Ghost Relay ledger entry and propose the next test-only validation step. Do not edit production files or run agents without explicit human approval.",
"expectedFiles": [
"reports/relay/ledger/*.json"
],
"forbiddenFiles": [
".env",
"**/api/**",
"tools/trading/**"
],
"acceptanceCriteria": [
"Proposal remains pending until manually approved",
"No outbox markdown written from proposal",
"No agent auto-launch"
],
"finalReportRequired": true
}Download this JSON and use the Local Import Command Pack below to copy safe CLI commands. Import copies the packet into reports/relay/inbox/. Run ProcessInboxOnce separately when ready. This UI never executes relay, writes server files, or auto-routes.
Local Import Command Pack
These commands must be pasted into your local terminal manually. Ghost Console does not run them.
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ImportTaskPath "<path-to-downloaded-approved-task-json>" -ImportInboxPath reports/relay/inbox -DryRun
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ImportTaskPath "<path-to-downloaded-approved-task-json>" -ImportInboxPath reports/relay/inbox -Execute
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -ProcessInboxOnce -InboxPath reports/relay/inbox -ArchiveProcessed -MaxInboxItems 1 -Execute
One-command local flow
One-command flow must be pasted into your local terminal manually. Ghost Console does not run it.
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunApprovedTaskPath "<path-to-downloaded-approved-task-json>" -RunApprovedInboxPath reports/relay/inbox -DryRun
powershell -NoProfile -ExecutionPolicy Bypass -File tools/ghost-relay/ghost-relay.ps1 -RunApprovedTaskPath "<path-to-downloaded-approved-task-json>" -RunApprovedInboxPath reports/relay/inbox -Execute
Enter a goal above to run the operating loop. Nothing executes — this build stops at approval.