Sol documentation index
August 2, 2026 · View on GitHub
- Class: index
- Status: current
- Dispatch: no. Follow the authority order below.
- Owner: Sol roadmap
Sol owns the OpenAgents roadmap that uses implementation results. Omega, a tracked Zed fork, is the primary Desktop and IDE destination. Omega is also the primary company workroom destination. It implements useful Buzz outcomes as native GPUI panes. Existing configured agents attach through explicit adapters and keep their custody boundaries.
The separate Buzz installation and standalone relay and forge program are canceled. The current Electron application remains the supported release and rollback source until the Omega cutover gates pass.
The accepted Desktop Codex Workroom baseline retains its proof records and
owner-approved packet ledgers. The 2026-08-02 live snapshot has no active
roadmap:sol product issue. Current work must come from another exact live
issue or owner-accepted packet.
The three 2026-07-17 surface ProductSpecs are results that agree with the roadmap.
Exact issues and packets continue to control dispatch.
Managed-sandbox epic #9023 and its projected leaves are closed. Portable movement and cross-machine Full Auto admission remain closed. Autonomous provider policy, fleet policy, voice, and settlement also remain closed. An owner must separately approve these expansions. This index points to the authorities. It does not store the roadmap revision or issue count.
Controlled language status
The 17 active Sol records use ASD-STE100 and the controlled agent compact profile. The compact profile permits reviewed density only when it improves fast and precise agent use. It does not weaken authority, safety, evidence, or ambiguity controls.
The Sol document manifest classifies current, historical, evidence, redirect, and tombstone records.
The P5 receipt in docs/ste records the conversion set.
Dispatch-safe reading order
When documents disagree, use this order:
AGENTS.mdandINVARIANTS.mdfor repository law.- Owning schemas, behavior contracts, product promises, and runtime policy for enforceable behavior.
MASTER_ROADMAP.mdfor current owner decisions, priority, gates, issue disposition, and execution order.- Live
roadmap:solissues andCLAIM_PROTOCOL.mdfor current work, ownership, and collision avoidance. - Current code, tests, deployments, release records, and indexed receipts for factual proof state.
- Dated analyses and teardowns only as pinned historical evidence.
Never dispatch from a revision number, issue count, current-state paragraph, or copy/paste prompt in a dated analysis. Refresh the authorities above.
Current authorities and contracts
../../AUTHORITY.md— the current owner action profile. It specifies programs, grants, budgets, independence, blocker routes, reserved actions, and receipt policy.2026-07-23-omega-zed-primary-surface-accepted-plan.md— active owner direction for Omega as the primary Desktop and IDE destination. It owns the Rust application core, packaged Node/Effect companion, native workrooms, existing-agent adapters, bounded Nostr interoperability, migration, proof, and cutover packet order. It admits planning and contract work only.2026-07-18-owner-delegated-autonomy-accepted-plan.md— active owner direction. It removes generic owner and device waits. It also directs completion of Full Auto, root specs, and Fast Follow. After that work, pursue transcript-backed user and revenue results within the profile.2026-07-18-sarah-owner-orchestrator-reboot-accepted-plan.md— active owner direction and implementation limits for Sarah. Sarah is the authenticated persistent orchestrator in OpenAgents mobile. It uses the cited business estimates and current capability brokers. Public/sarahand the standalone app remain retired.2026-07-19-managed-agent-sandboxes-accepted-plan.md— retained owner direction, ProductSpec, and closed SBX-00..10 issue ledger. It specifies the proof limit for one GCPSandboxResourceand the pinned Box v1 facade. Its receipts remain evidence, not a new live queue.2026-07-18-sarah-terminal-history-harness-implementation.md— current implementation record for Sarah's owner-private terminal history. It covers the experience compiler and the limited conversation candidate. It also covers the independent evaluation, release gate, and immutable next-turn harness.2026-07-18-autonomous-release-feedback-and-changelog-accepted-plan.md— active revision-2 RC release authority, RC17–RC20 feedback-loop correction, impact-based delivery/OTA boundaries, immutable GitHub publication, bounded issue/Forum communications, and attributed/changelogclosure plan.MASTER_ROADMAP.md— canonical product direction, durable gates, non-goals, current issue set, and next-ready sequence.issues/README.md— classification and navigation for checked-in issue sources. Live GitHub state remains the operational authority.CLAIM_PROTOCOL.md— same-session and cross-session claims, hot-contract coordination, staleness audit, and release receipts.2026-07-16-fast-follow-expansion-accepted-plan.md— owner admission, ordered five-day program, packet ledger, and explicit non-authority boundary.2026-07-17-t3-code-ui-full-harvest-accepted-plan.md— owner admission and bounded packet ledger for the complete T3 Code UI harvest into the mounted OpenAgents Desktop React workbench.2026-07-17-t3-code-mobile-full-parity-accepted-plan.md— owner admission and bounded packet ledger for the complete T3 mobile workbench adaptation through physical-device/distribution closure.2026-07-18-full-auto-asap-effect-harness-sprint-recommendation.md— implemented architecture decision for #9000–#9002. The team repaired the current Effect-owned Desktop mission, thread, and provider seams. The native Codex and Claude lanes passed the owner-real matrix. The test did not invoke the Vercel Harness fallback.2026-07-18-full-auto-one-click-concurrent-runs-implementation.md— current implementation record for the one-prompt launcher, collapsed advanced policy, bounded concurrent runs, and exact-run operator monitor.2026-07-18-electron-ai-sdk-codex-claude-full-auto-rewrite-roadmap.md— owner-directed and owner-local experiment for a limited Codex-first Full Auto loop. The experiment uses one shared Vercel AI SDK Harness workspace. Keep it as the provider oracle and fallback. It is not the production Full Auto authority or the selected sprint path.../../specs/desktop/full-auto.product-spec.md— Full Auto rev-14 mission and lifecycle for epic #8967. It specifies routes, guardrails, liveness, evidence, UX, and handoff intent. Live child issues control the implementation sequence.CHALLENGE_LEDGER.md— accepted risks, falsifiers, tripwires, and review points.receipts/README.md— indexed proof snapshots, rungs, and final dispositions. Evidence never selects the current queue.OPERATING_MODEL.md— bounded implementation and proof discipline. Its durable rules apply. Check transient status against the master and live issues.SUBSYSTEM_IMPLEMENTATION_IMPLICATIONS.md— subsystem ownership and architecture implications. Until its scheduled cleanup, current product state comes from the master and code, not its dated status prose.2026-07-10-r1-r2-identity-sync-contract.md— frozen identity and Sync contract for the R1/R2 boundary.2026-07-14-node-pnpm-vite-plus-full-conversion-plan.md— owner-selected Node/pnpm/Vite Plus destination, T3-style migration order, proof gates, non-MVP payment decommission/deletion contract, and the bounded live issue program. Dispatch still requires a valid issue claim.2026-07-14-node-pnpm-vite-plus-vp0-baseline.md— reproducible classified cutover inventory, selected destination pins, clean-source performance/failure baseline, and VP-1 through VP-6 claim ledger. It is evidence, not a current dispatch queue.2026-07-14-node-pnpm-vite-plus-vp1-reconciliation-decommission-receipt.md— aggregate-only production reconciliation, immutable private-archive digests, service/Cloud SQL shutdown evidence, and the recovery boundary for the retired money, markets, and Sites graph.2026-07-14-node-pnpm-vite-plus-vp2-node-runtime-receipt.md— stock-Node runtime, retained-service, SQLite, and public executable proof.2026-07-14-node-pnpm-vite-plus-vp3-parity-receipt.md— complete Vite Plus test parity, Effect TSGo disposition, host exceptions, architecture-plugin proof, and bounded VP-4 debt.
Evidence and historical baselines
These documents explain decisions or preserve point-in-time evidence. They are not current queues:
-
2026-07-23-nostr-first-rust-effect-desktop-analysis.md— retained comparison analysis for a Nostr-first Desktop with a larger Rust mechanics layer. The Omega plan supersedes its shell-selection question. Its Rust, Buzz, Nostr, provenance, and authority evidence remains useful. -
2026-07-20-pylon-bip39-nostr-spark-identity-recovery-audit.md— audit of the shared BIP-39 root that produced Pylon Nostr and Spark keys. It lists all known local recovery candidates and public comparison sources. It selects a shared local identity service and platform secret stores for future custody. It does not read a live secret, activate a wallet, or authorize implementation. -
2026-07-18-atif-thread-model-audit.md— implemented audit for the OpenAgents thread and event model. OpenAgents threads and events are the canonical operational truth. ATIF is a versioned public-safe evidence projection. The audit applies to IDE/SBX event export, but not to sandbox lifecycle authority. -
2026-07-18-sarah-openai-voice-api-audit.md— audit of official OpenAI documents and current code. It covers request TTS, limited transcription, and Realtime transcription. It also covers speech-to-speech Realtime, custom voices, and data controls. It recommends a chained provider adapter for the owner-private Sarah and audio authorities. It does not recommend a second voice agent. -
receipts/2026-07-18-full-auto-real-owner-acceptance.md— owner-real development receipt for six of six visible Codex and Claude tests. It records durable restart and thread-pressure proof. It also records report and analyzer digests and one automatic Claude-to-Codex rotation. Signed package and independent assurance remain #8978/#8979. -
2026-07-18-vercel-ai-sdk-source-derived-effect-conversion-audit.md— new baseline audit of selected Apache-2.0 Vercel source. The audit examines the current Effect v4 AI modules. It rejects a duplicate generic Effect AI SDK. It keeps only a conditional Effect Harness seam that comes from the source after native proof. -
2026-07-18-openagents-desktop-vercel-ai-sdk-in-place-reset-audit.md— point-in-time technical and product audit of the active OpenAgents Desktop app. It examines a change to the hardened Electron and release shell. The proposed change replaces the runtime and renderer with a limited Vercel AI SDK experiment. The new baseline rejects that reset for the current sprint. It records the required contract retirements and does not authorize deletion. -
../fable/2026-07-17-surface-vision-gap-analysis-and-roadmap.md— detailed AC-by-AC Desktop, mobile, and web audit and seven-track proposal. MASTER_ROADMAP owns its reconciled sequence and current truth. -
../fable/2026-07-17-full-auto-implementation-audit.md— corrected overnight incident and run-model audit that anchors #8967. Later landings do not rewrite its point-in-time findings. -
2026-07-16-desktop-local-usage-opt-in-verification.md— default-off privacy, admission, outbox, opt-out, and owner-gated live-proof contract for Desktop local-turn usage ingestion under issue #8911. -
2026-07-16-openagents-desktop-full-auto-deep-dive.md— pinned revision-2 runtime and proof audit. Its first section agrees with the landed revision-9 criteria and current residuals. The original findings remain historical evidence, not a current dispatch queue. -
2026-07-16-full-auto-shared-mac-dogfood-runbook.md— reconciles the #8928 stop timeline, pins durable disable attribution, and defines exact-PID/shared-userData process ownership for safe Desktop dogfood. -
2026-07-15-khala-ui-components-workbench-receipt.md— deterministic implementation proof for the complete 30-capability, non-audio Khala UI workbench at/components/khala, its seven-package Effect Native vendor boundary, and its React/Electron-compatible lifecycle contract. -
2026-07-15-effect-native-storybook-workbench-receipt.md— implementation and release proof for the rendered 79-component, 108-story Effect Native workbench at/components/coreand its renderer, training, and visual-token family surfaces. -
2026-07-15-tanstack-start-unified-docs-migration-receipt.md— live closure evidence for the unified TanStack Start documentation reader. It covers same-origin agent artifacts and the monolith load-balancer cutover. It also records retirement of the Blume workspace andopenagents-docsservice graph. -
2026-07-15-tanstack-start-unified-docs-migration-audit.md— owner-directed consolidation audit that selected the now-completed public docs move into TanStack Start. The migration receipt supersedes its pre-implementation status. -
2026-07-15-arwes-first-principles-khala-ui-architecture-audit.md— pinned first-principles Arwes audit and ordered architecture. It specifies an owned Khala UI language in Effect Native. Implementation requires limited claimed issues. It does not authorize growth of the frozen/tanstackcomparison. -
2026-07-14-blume-docs-site-integration-audit.md— pinned analysis of Blume suitability and destination-toolchain compatibility. It also covers public content and the same-origin/docsservice. Implementation requires a new limited post-VP-4 issue and claim. -
2026-07-14-native-sdk-desktop-after-action-audit.md— closed after-action record for the removed Native SDK experiment. It explains why the hybrid did not meet the adoption criterion. It keeps only component and automation lessons. It establishes Electron as the sole Desktop host. -
2026-07-14-google-cloud-authority-cleanup-after-action.md— corrective authority record. Google Cloud is the sole production infrastructure. SHC was only a retired limited pilot. The team archived and removed the Cloudflare account residue. The team did not treat this residue as a migration target. -
2026-07-10-openagents-desktop-product-architecture.md— durable process/data/authority topology plus a dated delivery baseline. -
2026-07-10-opencode-khala-openagents-desktop-parity-audit.md— pinned competitor-capability and OpenAgents implementation baseline. Its score is historical, not current product status. -
../mvp/README.md— canonical first-MVP package: the exact ProductSpec plus its point-in-time OpenChamber/OpenCode/Codex audit. It is not a dispatch queue or a CUT-27 completion claim. -
2026-07-10-khala-code-mvp-to-openagents-mobile-port-plan.md— legacy capability disposition and remote-workroom design evidence. The master owns current acceptance and sequence. -
2026-07-10-episode-195-followup-analysis.md— dated product and compute-economics analysis. -
2026-07-11-openagents-coding-cutover-issue-plan.md— dated CUT-01–CUT-27 dependency and acceptance design. Live issue state and receipts override its mutable status prose. -
2026-07-11-remote-first-portable-coding-sessions-pathway.md— historical portable-session, target, broker, mobile, and voice design. Its open leaves have the closed not-planned state. It cannot dispatch work. -
2026-07-12-cut27-cutover-readiness-audit.md— chronological readiness evidence, not a current blocker table. -
../teardowns/README.md— point-in-time product and implementation evidence. Promoted requirements must live in an owning contract, roadmap gate, issue, or verifier.
Backroom keeps the July 9 Sol argument corpus as exact, non-dispatch evidence.
It has a two-way
archive manifest.
The July 10 execution diary remains pinned historical evidence. Neither is in
this safe dispatch path.
Proof and receipt rule
Preserve the distinction between:
- code landed.
- fixture or model proven.
- deployed or distributed.
- live proven.
- owner accepted.
- closed.
A later receipt can supersede an intermediate statement in an earlier receipt. This change does not make the earlier evidence false. Use the file's final disposition and the canonical roadmap. Do not promote an intermediate rung to current truth.
Documentation maintenance
Every new or changed Sol artifact should:
- declare its class, snapshot, dispatch status, and owner.
- separate durable contract from transient implementation state.
- link to live issue authority instead of an open count.
- avoid hard-coded master revisions in active documents.
- name the proof rung and final disposition for receipts.
- move authoritative conclusions out of historical analysis.
- preserve owner decisions, failures, contracts, and non-revival tombstones.
- use Backroom for fully obsolete narrative only after conclusion extraction, link migration, and a pushed provenance manifest.
- run the Sol documentation gate before you land a roadmap or proof document.
The offline test
scripts/check-sol-docs.test.tsandpnpm run check:sol-docsmust be green. The pre-pushcheck:fastprofile does not run this gate, so run it by hand before a docs-only or proof-document push. The gate checks theMASTER_ROADMAP.mdline budget, internal links, and the document manifest.
The target is one small dispatch surface surrounded by indexed evidence, not a freshened copy of every old plan.