NeuralForge Builds NeuralForge

Home

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Open command bar

Spaces Dashboard

neural-forge-build

Source-backed space for the NeuralForge self-build loop.

Sources: 3
Actions: 3
Panels: 16

External actionsapproval gated
Generated widgetssandboxed

Project Overview

layer contract

NeuralForge keeps verification, execution, memory, contracts, and generated workspace surfaces separated until every risky boundary is approved.

RLA Forgeverification brain
OpenClawexecution body in dry-run
Hermesprocedural skill review
Space Agentliving browser workspace
Garnetsemantic contract layer
Memory Coreworking, episodic, semantic, procedural continuity

Workspace

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Inspect widget sandbox

execution disabled

Generated widget execution is disabled until metadata, approvals, dependencies, touched files, and rollback point are reviewed.

Purpose: Show the full product skeleton without enabling live generated-code execution
Permission: read_only
Network: disabled
Rollback: neural-forge-build@2026-04-29T18:30:00.000Z

Procedural Skills Review

Hermes pattern

Run NeuralForge build skeleton loop review_required
Compile Garnet workspace spec -> Verify architecture claims through RLA adapter -> Plan OpenClaw actions in dry-run mode -> Write Memory Core records -> Render Living Workspace panels -> Snapshot state for rollback

RLA Explorer

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Route a fixture query

Natural-Language Route Result

screen 26 | offline router

Fixture queries from specs/rla-taxonomy/taxonomy-canon-v1.json route offline against locked canon v1-provisional-2026-07-14 (provisional). Every route names all five subsystem destinations and unmatched input fails closed to human review.

Verify this healthcare clinic expansion proposal ready
Healthcare Innovation Lab (domain 7, pro tier, 78% confidence)
matched terms: healthcare, clinic
Example
nl-route-healthcare-clinic

Evidence refs
taxonomy-canon:v1-provisional-2026-07-14, domain:7, term:healthcare, term:clinic
- rla-adapter.createOfflineRlaVerification ready Offline verification staged for domain 7 (Healthcare Innovation Lab)
- garnet-bridge.compileGarnetSource ready Manifest translation staged; Garnet Bridge translates, it never compiles upstream tooling
- openclaw-adapter.planOpenClawDryRun ready Dry-run-first plan staged behind capability manifests and approval policy
- memory-core.remember ready Continuity metadata records the routed query and its evidence refs
- reports.session-route-log ready Route result is appended to the local session report evidence
Route a real estate property acquisition memo ready
Real Estate Empire Building (domain 18, pro tier, 70% confidence)
matched terms: real estate, property
Example
nl-route-real-estate-memo

Evidence refs
taxonomy-canon:v1-provisional-2026-07-14, domain:18, term:real estate, term:property
- rla-adapter.createOfflineRlaVerification ready Offline verification staged for domain 18 (Real Estate Empire Building)
- garnet-bridge.compileGarnetSource ready Manifest translation staged; Garnet Bridge translates, it never compiles upstream tooling
- openclaw-adapter.planOpenClawDryRun ready Dry-run-first plan staged behind capability manifests and approval policy
- memory-core.remember ready Continuity metadata records the routed query and its evidence refs
- reports.session-route-log ready Route result is appended to the local session report evidence
Ship production software code for the execution gateway ready
Elite Software Craft (domain 6, free tier, 72% confidence)
matched terms: software, execution, code
Example
nl-route-software-execution

Evidence refs
taxonomy-canon:v1-provisional-2026-07-14, domain:6, term:software, term:execution, term:code
- rla-adapter.createOfflineRlaVerification ready Offline verification staged for domain 6 (Elite Software Craft)
- garnet-bridge.compileGarnetSource ready Manifest translation staged; Garnet Bridge translates, it never compiles upstream tooling
- openclaw-adapter.planOpenClawDryRun ready Dry-run-first plan staged behind capability manifests and approval policy
- memory-core.remember ready Continuity metadata records the routed query and its evidence refs
- reports.session-route-log ready Route result is appended to the local session report evidence
Backtest a quantitative trading research idea ready
Quantitative Trading Systems (domain 17, pro tier, 74% confidence)
matched terms: quantitative, trading, backtest
Example
nl-route-quant-backtest

Evidence refs
taxonomy-canon:v1-provisional-2026-07-14, domain:17, term:quantitative, term:trading, term:backtest
- rla-adapter.createOfflineRlaVerification ready Offline verification staged for domain 17 (Quantitative Trading Systems)
- garnet-bridge.compileGarnetSource ready Manifest translation staged; Garnet Bridge translates, it never compiles upstream tooling
- openclaw-adapter.planOpenClawDryRun ready Dry-run-first plan staged behind capability manifests and approval policy
- memory-core.remember ready Continuity metadata records the routed query and its evidence refs
- reports.session-route-log ready Route result is appended to the local session report evidence
Draft a mars colonization habitat brief ready
Space Exploration & Colonization (domain 41, apex tier, 64% confidence)
matched terms: mars, colonization
Example
nl-route-mars-colonization

Evidence refs
taxonomy-canon:v1-provisional-2026-07-14, domain:41, term:mars, term:colonization
- rla-adapter.createOfflineRlaVerification ready Offline verification staged for domain 41 (Space Exploration & Colonization)
- garnet-bridge.compileGarnetSource ready Manifest translation staged; Garnet Bridge translates, it never compiles upstream tooling
- openclaw-adapter.planOpenClawDryRun ready Dry-run-first plan staged behind capability manifests and approval policy
- memory-core.remember ready Continuity metadata records the routed query and its evidence refs
- reports.session-route-log ready Route result is appended to the local session report evidence
zxqv flurble wibble needs_review
Truth-Seeking & Inquiry (domain 79, apex tier, 45% confidence)
matched terms: none — fail-closed fallback
Example
nl-route-nonsense-fallback

Evidence refs
taxonomy-canon:v1-provisional-2026-07-14, domain:79, fallback:no-matching-terms
- rla-adapter.createOfflineRlaVerification blocked Fail-closed fallback: no canon routing terms matched, human review required before verification
- garnet-bridge.compileGarnetSource blocked Fail-closed fallback: no canon routing terms matched, human review required before verification
- openclaw-adapter.planOpenClawDryRun blocked Fail-closed fallback: no canon routing terms matched, human review required before verification
- memory-core.remember ready Continuity metadata records the routed query and its evidence refs
- reports.session-route-log ready Route result is appended to the local session report evidence

Forge Score

passed

86% minimum 86%

Verification threshold met in offline RLA skeleton

Forge Score Explanation

weighted factors

Score: 86%
Threshold: 86%
Pass/block state: pass

claim passed weight 20% | Offline RLA mock completed claim layer
source passed weight 25% | Offline RLA mock completed source layer
model_jury passed weight 20% | Offline RLA mock completed model_jury layer
counterargument passed weight 15% | Offline RLA mock completed counterargument layer
score passed weight 10% | Offline RLA mock completed score layer
execution_check passed weight 10% | Offline RLA mock completed execution_check layer

Factor weights are provisional NeuralForge projections for the offline RLA skeleton, not upstream RLAForge semantics.

Verification Trace

trace-neural-forge-build

claim passed
Offline RLA mock completed claim layer
source passed
Offline RLA mock completed source layer
model_jury passed
Offline RLA mock completed model_jury layer
counterargument passed
Offline RLA mock completed counterargument layer
score passed
Offline RLA mock completed score layer
execution_check passed
Offline RLA mock completed execution_check layer

Domain Taxonomy

RLA route

Elite Software Craft72%
Truth-Seeking & Inquiry45%

Offline routing keeps domain scoring separate from execution.

Actions

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Review dry-run plans

Actions

OpenClaw firewall

recall-source-corpus dry_run
Recall source corpus and current project context
dry-run preview | memory.recall
draft-product-skeleton dry_run
Draft local product skeleton artifacts from verified manifests
dry-run preview | filesystem.write
publish-progress-update blocked
Publish a progress update after human approval
approval required | channels.external_message

OpenClaw Dry-Run Console

dry-run-first

Plans: 3
Approval-blocked: 1
External default: requireApproval

recall-source-corpus dry_run internal_read | memory.recall | dry-run allowed | dry run supported
draft-product-skeleton dry_run local_file_write | filesystem.write | dry-run allowed | dry run supported
publish-progress-update blocked external_message | channels.external_message | approval required | dry run supported

Contracts

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Browse source contracts

File Source Browsersource-backed

        garnet-workspace-spec
        specs/garnet/examples/neuralforge-build.garnet
        schema | Limited NeuralForge Garnet workspace fixture
      

        phase4a-source-lock
        docs/architecture/008-phase-4a-context-lock.md
        architecture-doc | Phase 4A source verification and repo selection
      

        adr-0002
        docs/decisions/ADR-0002-openclaw-hermes-space-agent-fusion.md
        architecture-doc | Fusion boundary for OpenClaw, Hermes, and Space Agent
      

Memory

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Inspect rollback snapshots

Memory

snapshot-neural-forge-build-1

Events: 10
Links: 0
Skills: 1

Working session
Episodic project
Semantic source-backed
Procedural review queue
Snapshot snapshot-neural-forge-build-1

Time Travel Snapshots

rollback required

Latest rollback point: neural-forge-build@2026-04-29T18:30:00.000Z

Strategy: snapshot
Snapshot scope: project
Snapshot id: snapshot-neural-forge-build-1

Reports

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Open artifact explorer

Session Artifact Explorer

read-only artifact loader

Persisted session artifacts are review surfaces for the gateway pipeline; they do not load code, open network connections, or mutate state.

Session: neural-forge-build@snapshot-neural-forge-build-1
Artifacts: 12
Mode: read_only

Session summary indexed
neural-forge-build verification, action, memory, and workspace summary
summary | phase8-demo-summary.json
Workspace state indexed
16 rendered panels with read-only controls
workspace | phase8-session-workspace-state.json
Verification trace passed
RLA trace trace-neural-forge-build with score 0.86
verification | phase8-session-verification-trace.json
OpenClaw action ledger blocked
3 planned actions, 1 approval-blocked
actions | phase8-session-action-ledger.json
Memory snapshot indexed
snapshot-neural-forge-build-1
memory | phase8-session-memory-snapshot.json
Procedural skill candidate review_required
Run NeuralForge build skeleton loop
procedural | phase8-session-procedural-skill.json
Browser QA harness needs_review
Desktop, mobile, disclosure, and safety assertions for the static Living Workspace shell
browser_qa | phase8-session-browser-qa.json
Artifact id
browser-qa-harness

Source
Living Workspace browser QA

Path
phase8-session-browser-qa.json

Source refs
garnet-workspace-spec, phase4a-source-lock, adr-0002

Mutation
disabled

Network
disabled
Browser QA checklist
- Desktop layout smoke desktop-layout #session-artifact-explorer Session Artifact Explorer is present without horizontal overflow
- Action review disclosure action-disclosure [data-open-review="action-review-recall-source-corpus"] Review detail opens while external execution remains disabled
- Widget sandbox disclosure sandbox-disclosure [data-open-review="widget-sandbox-review"] Sandbox metadata opens while generated widget execution remains disabled
- Artifact review disclosure artifact-disclosure [data-open-review="session-artifact-browser-qa-harness"] Browser QA checklist detail opens as read-only metadata
- Mobile navigation reachability mobile-nav nav a[href="#session-artifact-explorer"] Explorer link remains reachable at mobile width
- Disabled safety controls safety-disabled-controls button[disabled] Execution and mutation controls stay disabled
Guardrails
- no console warnings or errorsrequired
- no page-level horizontal overflowrequired
- no remote scripts or stylesheetsrequired
- no browser network loadingrequired
- no dynamic code executionrequired
- no artifact mutationrequired
Evidence slots
- Desktop explorer screenshotmanual_required
- Expanded disclosure screenshotmanual_required
- Mobile explorer reachability screenshotmanual_required
- Console warning/error logmanual_required
Compound Engineering loop source_mapped
Strategy, requirements, plan, work, review, and compound stages mapped into the NeuralForge phase loop
compound_engineering | phase8-session-compound-loop.json
Artifact id
compound-engineering-loop

Source
EveryInc/compound-engineering-plugin

Path
phase8-session-compound-loop.json

Source refs
garnet-workspace-spec, phase4a-source-lock, adr-0002

Mutation
disabled

Network
disabled
CE phase loop
- Strategy anchor /ce-strategy phase architecture objective | target problem, persona, metrics, and tracks
- Requirements brainstorm /ce-brainstorm phase acceptance criteria | right-sized requirements before implementation
- Implementation plan /ce-plan sequenced plan and verification gates | bounded plan with confidence checks
- Work execution /ce-work branch execution against current AGENTS.md contracts | small reviewed changes with task tracking
- Review panel /ce-code-review and /ce-doc-review local checks, browser QA, and GitHub PR checks | persona-reviewed findings before merge
- Compound learning /ce-compound architecture docs, source summaries, and phase notes | reusable learning that makes the next phase easier
Guardrails
- no CE plugin install required for repository checksrequired
- no global agent or plugin mutationrequired
- no hook installationrequired
- no live external executionrequired
- no generated widget executionrequired
- no artifact mutationrequired
Evidence slots
- Phase strategy notesource_mapped
- Requirements and acceptance notessource_mapped
- Review and verification evidencesource_mapped
- Reusable learning capturesource_mapped
ECC readiness gate source_mapped
Codex plugin, quality, security, orchestration, memory, and hook surfaces mapped to NeuralForge approval gates
ecc_readiness | phase8-session-ecc-readiness.json
Artifact id
ecc-readiness-gate

Source
affaan-m/ECC

Path
phase8-session-ecc-readiness.json

Source refs
garnet-workspace-spec, phase4a-source-lock, adr-0002

Mutation
disabled

Network
disabled
ECC readiness checks
- Codex plugin surface source_mapped .codex-plugin/plugin.json Garnet contract review | plugin slug, skills path, MCP policy, and install fragility noted before activation
- Workflow quality source_mapped workflow-quality RLA Forge verification evidence | TDD, e2e, eval harness, and verification-loop vocabulary mapped into checks
- Security and AgentShield source_mapped security approval and threat-boundary review | security-review, security-scan, hook, MCP, permission, and secret-surface review before install
- Loop orchestration source_mapped orchestration and agentic-patterns OpenClaw dry-run adapter boundary | loop and worktree/tmux orchestration kept as metadata until execution approval
- Memory continuity source_mapped continuous-learning and memory hooks Memory Core review queue | learned skill and session persistence paths reviewed before any memory write bridge
- Hook runtime blocked hooks-runtime disabled until explicit runtime approval | no hook installation, duplication, or user-level config mutation in this phase
Guardrails
- no ECC plugin install during repository checksrequired
- no global Codex, Claude, Cursor, or OpenCode state mutationrequired
- no hook runtime activationrequired
- no MCP credential or environment variable importrequired
- no live external executionrequired
- no generated widget executionrequired
- no artifact mutationrequired
Evidence slots
- ECC source provenancesource_mapped
- Codex plugin compatibility notesource_mapped
- Quality and security gate mapsource_mapped
- Runtime activation approval recordblocked
Dogfood sandbox readiness needs_review
Evidence-backed readiness matrix and first approved-sandbox design while execution remains disabled
dogfood_sandbox | phase8-session-dogfood-sandbox.json
Artifact id
dogfood-sandbox-readiness

Source
dogfood-readiness Phase 15 audit

Path
phase8-session-dogfood-sandbox.json

Source refs
garnet-workspace-spec, phase4a-source-lock, adr-0002

Mutation
disabled

Network
disabled
Dogfood readiness matrix

Score
78

Decision
conditionally ready after listed fixes

Formula
100 - medium*5 - blocked_core_cases*3 - unverified_core_claims*3 - security_coverage_gaps*3
- Repo provenance passed source truth git status, remotes, HEAD, and GitHub main workflow verified after Phase 14 merge
- Baseline verification passed release readiness npm run check, CLI demo, git diff --check, and GitHub Check workflow passed
- Browser dogfood passed Living Workspace usability in-app browser desktop and 390x844 mobile smoke found no overflow or console errors
- Security boundary partial sandbox trust static checks block remote scripts, inline handlers, eval, fetch, iframe, and secret patterns
- Runtime sandbox blocked approved execution no local-only OpenClaw runner or generated-widget execution sandbox exists yet
- External actions blocked approval policy publish-progress-update remains approval-blocked and liveActionsExecuted remains false
Approved sandbox candidates
- Local artifact export sandbox design_only OpenClaw local dry-run adapter write deterministic artifacts only under caller-provided temp output directories | per-run approval required before moving beyond existing CLI writer | rollback: delete temp output directory and restore Memory Core snapshot
- Generated widget preview sandbox blocked Living Workspace renderer render reviewed metadata and static markup with network and dynamic code disabled | blocked until DOM isolation and rollback checks are implemented | rollback: discard preview state and keep persisted artifact unchanged
- GitHub metadata review sandbox read_only read-only repo evidence lane read PR/check metadata for readiness evidence only | write/merge actions remain explicit external repo operations | rollback: no repo mutation from metadata reads
Report refs
- reports/dogfood-readiness-phase15-2026-06-20.mdevidence
- reports/dogfood-readiness-phase15-2026-06-20.jsonevidence
- reports/dogfood-readiness-phase15-2026-06-20-matrix.mdevidence
- reports/dogfood-readiness-phase15-2026-06-20-mutations.mdevidence
- reports/dogfood-readiness-phase15-2026-06-20.htmlevidence
Guardrails
- no sandbox execution yetrequired
- no live external executionrequired
- no generated widget executionrequired
- no hook runtime activationrequired
- no plugin installationrequired
- no MCP credential or environment importrequired
- no artifact mutation outside caller-selected output directoryrequired
Evidence slots
- Dogfood readiness reportrecorded
- Coverage and trust-boundary matrixrecorded
- Falsification and mutation logrecorded
- Sandbox runtime approval recordblocked
OpenClaw local sandbox sandbox_available
Local file-write sandbox runner for approval-allowed actions under caller-selected output directories
local_sandbox | phase8-session-local-sandbox-runner.json
Artifact id
openclaw-local-sandbox

Source
OpenClaw adapter local sandbox

Path
phase8-session-local-sandbox-runner.json

Source refs
garnet-workspace-spec, phase4a-source-lock, adr-0002

Mutation
disabled

Network
disabled
Local sandbox runner
- Allowed action type implemented OpenClaw approval policy Only approval-allowed local_file_write actions can enter the sandbox
- Required sandbox directory implemented caller-selected output runOpenClawLocalSandbox blocks when sandboxDir is missing
- Deterministic artifact write implemented local artifact export runner writes <action-id>.sandbox.json with liveExternalActionExecuted false
- External actions blocked implemented external action firewall approval-required external_message actions stop at dry-run approval gate
- Rollback contract implemented time travel and cleanup artifact records rollback strategy delete_sandbox_dir
Sandboxed action candidates
- Draft product skeleton implemented runOpenClawLocalSandbox local_file_write action metadata exported under caller-selected sandboxDir | allowed only when manifest approval is not_required and dry-run planning passes | rollback: delete_sandbox_dir
Guardrails
- no default demo sandbox executionrequired
- no external action executionrequired
- no shell command executionrequired
- no browser action executionrequired
- no generated widget executionrequired
- no hook runtime activationrequired
- no writes outside caller-selected sandbox directoryrequired
Evidence slots
- OpenClaw local sandbox unit testspassed
- CLI demo remains read-only by defaultpending_full_gate
- Living Workspace artifact browser reviewpending_browser_smoke
NL route flow ready
Offline NL flow routed Verify this healthcare clinic expansion proposal to Healthcare Innovation Lab (domain 7) with a pass Forge Score and 1 approval-blocked of 3 dry-run plans
nl_route_flow | phase8-session-nl-route.json
Artifact id
nl-route-flow

Source
gateway NL flow pipeline

Path
phase8-session-nl-route.json

Source refs
garnet-workspace-spec, phase4a-source-lock, adr-0002

Mutation
disabled

Network
disabled
NL flow summary

Query
Verify this healthcare clinic expansion proposal

Domain
Healthcare Innovation Lab (domain 7)

Confidence
78%

Forge Score state
pass

Dry-run plans
2 allowed | 1 approval-blocked

Memory record
nl-route-verify-this-healthcare-clinic-expansion-proposal

Live actions
not executed

Flow checks
- Route result panel route-result-panel #nl-route-result Screen 26 route result panel stays rendered read-only
- Offline command bar command-bar-offline #workspace-command-bar Command bar routes fixture queries offline without network calls
- Dry-run console dry-run-console-panel #openclaw-dry-run Dry-run console keeps the approval-blocked publish plan visible
- Route map disclosure route-map-disclosure [data-open-review="nl-route-review-nl-route-healthcare-clinic"] Fixture route map detail opens as read-only metadata
Guardrails
- no live routing endpointrequired
- no live external executionrequired
- no generated widget executionrequired
- no artifact mutationrequired
- liveActionsExecuted stays falserequired
Evidence slots
- NL route flow recordrecorded
- Memory Core session recordrecorded
- Approval-blocked publish planblocked

Settings

RLA verified
Approval gate
Rollback available
Runtime: no live action enabled

Safety and Approval Settings

fail closed

Approval policy stays fail-closed: external actions default to requireApproval and destructive actions default to deny. Runtime toggles remain disabled until an approved later phase.

External actions defaultrequireApproval
Destructive actions defaultdeny
Allow listread_local_sources, draft_documents, run_local_validation
Approval listexternal_message, shell_destructive, publish_repo
Deny listdelete_source_corpus, bypass_widget_sandbox

Home

Spaces Dashboard

Project Overview

Workspace

Generated Widget Sandbox

Procedural Skills Review

RLA Explorer

Natural-Language Route Result

Forge Score

Forge Score Explanation

Verification Trace

Domain Taxonomy

Actions

Actions

OpenClaw Dry-Run Console

Contracts

File Source Browser

Memory

Memory

Time Travel Snapshots

Reports

Session Artifact Explorer

Browser QA checklist

Guardrails

Evidence slots

CE phase loop

Guardrails

Evidence slots

ECC readiness checks

Guardrails

Evidence slots

Dogfood readiness matrix

Approved sandbox candidates

Report refs

Guardrails

Evidence slots

Local sandbox runner

Sandboxed action candidates

Guardrails

Evidence slots

NL flow summary

Flow checks

Guardrails

Evidence slots

Settings

Safety and Approval Settings