core/ROADMAP.md

Armature — Roadmap

Current: v0.7.7 — API Tokens + Extension Permissions

Self-hosted extensible platform kernel. Auth, identity, packages, Starlark sandbox, storage, realtime, and ops are kernel primitives. Everything else is an extension.

Kernel capabilities: Auth (builtin/mTLS/OIDC) · Users/teams/groups/RBAC · Surfaces/extensions/libraries/workflows · Starlark sandbox (capability-gated) · Object storage (PVC/S3) + ext_data tables · WebSocket hub + realtime pub/sub · Audit log · Notifications · Scheduled tasks · Cluster registry + HA

Completed history: v0.2.x–v0.5.x fully documented in CHANGELOG.md. Highlights: RBAC + settings cascade, event bus + triggers, SDK stabilization, workflow engine (multi-stage, team roles, signoff gate, public entry, SLA), package distribution, Notes surface (CM6, folders, tags, backlinks, graph), realtime primitive, Chat surface (chat-core library + surface + polish), upgrade test harness, cluster registry + HA.

---

v0.6.x — Completed (MVP + Hardening)

All v0.6.x work is shipped and documented in CHANGELOG.md. Summary:

Version	Title	Key Deliverables
v0.6.0	Cluster Registry + HA	PG-backed node registry, heartbeat sweep, LISTEN/NOTIFY routing, self-eviction
v0.6.1	Backup/Restore + Docs	.swb archive format, server-side backups, docs surface + 5 guides
v0.6.2	Docs Polish + OpenAPI	Dark mode fix, topbar nav, api_schema manifest field, dynamic spec builder
v0.6.3	Dead Code Sweep	Registry install fix, dead Go/JS/HTML deletion, narrowed default bundle
v0.6.4	Admin Health/Metrics	Cluster dashboard merged into Admin tab, block renderer requires removed
v0.6.5	Renderer Pipeline	sw.renderers.register() kernel primitive, unified markdown, docs rewrite
v0.6.6	Final Hardening	Dependency auto-activation, ValidateManifest(), OIDC nonce, ICD/SDK update
v0.6.7	Native mTLS	TLS_MODE config, MTLSNativeProvider, node-to-node mTLS, armature-ca.sh
v0.6.8	Cookie Fix + UI Roadmap	Cookie SameSite fix, UI hardening roadmap published
v0.6.9	Session Lifetime Config	Admin-configurable TTLs, idle timeout, "keep me logged in"
v0.6.10	Viewport Foundation	Single layout model, CSS zoom, 100dvh, dead shell deprecated
v0.6.11	CSS Deduplication	Old primitive system retired, one class per concept
v0.6.12	Extension CSS Isolation	Prefix enforcement via linter, all 12 in-tree packages migrated
v0.6.13	Responsive & Spacing	Spacing token scale (4px grid), tablet breakpoint
v0.6.14	Visual Polish	Stale fallback colors purged, fonts self-hosted, radius tokens
v0.6.15	User Display Audit	Batch user resolve API, sw.users SDK module
v0.6.16	Usability Survey Gate	Four audit scripts, contrast/touch-target fixes
v0.6.17	Bug Fixes & Welcome	Notes folder fix, team member add fix, welcome auto-disable, zero default bundle
v0.6.18	CI Bundle Wiring	BUNDLED_PACKAGES env var wired into Gitea CI pipeline

---

v0.7.x — Test Infrastructure + Quality Gate

The v0.6.x series built the kernel. v0.7.x makes it provably correct.

A full surface audit (docs/AUDIT-surfaces.md) found 7 cross-surface issues and 18 surface-specific issues across Settings, Admin, Team Admin, and Docs. Only Docs is properly built. The fix is three phases: (1) establish a shell contract and bring all four primary surfaces to parity, (2) build a runner framework for end-to-end browser tests, (3) automate those runners headlessly in CI.

Design docs:

• docs/DESIGN-shell-contract.md — two-slot topbar, three navigation patterns, surface migrations
• docs/DESIGN-surface-runners.md — runner framework, requires declarations, headless E2E
• docs/AUDIT-surfaces.md — full audit findings

v0.7.0 — Shell Contract + Surface Audit + Rebrand Cleanup

Design doc: docs/DESIGN-shell-contract.md

Shell Infrastructure

Step	Status	Description
Shell topbar (two-slot model)	done	Kernel injects topbar into surface-extension template. Two named slots: left (defaults to manifest title) and center (flex: 1, for tabs/pickers/search). Home link, notification bell, and user menu always present.
Topbar customization API	done	sw.shell.topbar.setLeft(vnode) overrides left slot. sw.shell.topbar.setSlot(vnode) sets center slot. sw.shell.topbar.setTitle(str) shorthand for text-only left. sw.shell.topbar.hide() / .show() for full-bleed surfaces.
Kernel tab CSS	done	.sw-topbar__tabs and .sw-topbar__tab classes for consistent tab styling in the center slot. Surfaces use these for free or style their own slot content.
Notification read broadcast	done	Backend emits notification.read and notification.all_read WS events. Bell listens for .created, .read, .all_read. Cross-surface sync without refetch.
User menu reactivity	done	Emit package.changed (install/uninstall/enable/disable) and auth.changed (role/membership) WS events. UserMenu listens and re-fetches surface list. Most impactful single fix.
Shell announcement global dismiss	done	Dismissed state persisted to localStorage keyed by content hash. Dismiss once, dismissed everywhere.

Surface Migrations

Step	Status	Description
Settings → Pattern B (flat tabs)	done	Delete custom topbar + sidebar nav. 6 sections become flat tabs in topbar center slot. Content full-width. Fix Teams section: add team admin link, role display, leave action. Remove sessionStorage return URL logic.
Admin → Pattern C (category tabs + sidebar)	done	Delete custom admin-topbar. setLeft() for favicon + "Administration". setSlot() for category tabs (People / Workflows / System / Monitoring). Admin sidebar (sub-navigation) unchanged — surface-owned, below the topbar. Bell + user menu come free from shell. Delete bespoke CatIcon renderer if using standard SVGs.
Team Admin → Pattern B (flat tabs)	done	Delete custom topbar + sidebar nav. 5 sections (Members / Connections / Workflows / Settings / Activity — Groups removed) become flat tabs. Content full-width. setTitle() for team-specific name. Remove sessionStorage return URL. Fix signoff user display (user_id → sw.users.displayName()).
Team Admin: remove Groups tab	done	37-line dead-end. Read-only "No groups" with no create/docs/link. Remove until team-scoped group management is properly designed.
Docs → Pattern A (default)	done	Delete explicit Topbar import. Shell topbar auto-renders with manifest title. Docs sidebar (document list) is in content area, unaffected.

Error Handling + UX Pass

Step	Status	Description
Inline error states	done	Replace catch { toast } with inline error + retry on all list endpoints. New .sw-inline-error CSS primitive. Systematic pass across Settings, Admin, Team Admin.
Empty state guidance	done	Every "No X" message gets one-line explanation + primary action (create button or doc link). Admin Groups, Workflows, Teams; Team Admin Workflows; Settings Notifications.

Bug Fixes

Step	Status	Description
evil-chat cleanup	done	ICD security tier: finally cleanup block + tighten 409 assertion.
Workflow demo error surfacing	done	Replace silent catch with inline error + retry.
Hello dashboard removal	done	Delete packages/hello-dashboard/.

Rebrand

Step	Status	Description
Light-mode icon SVG	done	New favicon-light.svg — square icon, transparent bg, dark node fills. Rename current favicon-light.svg (wordmark) to wordmark.svg.
Dark-mode wordmark SVG	done	New wordmark-dark.svg — light text for dark backgrounds.
Light-mode raster assets	done	favicon-light-32.png, favicon-light-256.png.
PWA manifest description	done	"Self-hosted extension platform — build, compose, and run extensions."
REBRAND-SPEC.md		Land into docs/. Find/replace patterns, validation checklist, asset inventory.
base.html favicon swap	done	Verify theme swap works with new square light icon.

Tests

Step	Status	Description
Shell topbar renders for extensions	done	Home, left slot, center slot, bell, user menu present.
Topbar API (setLeft, setSlot, hide)	done	Custom content renders. Hide removes topbar.
All 4 surfaces use shell topbar	done	No double topbars. Each pattern (A/B/C) renders correctly.
User menu reactive to package install		Install → menu updates without reload.
Notification bell cross-surface sync		Dismiss on Notes → clears on Chat.
Inline error on API failure		Error + retry shown, not empty list.

v0.7.1 — Surface Runner Framework

Design doc: docs/DESIGN-surface-runners.md

Step	Status	Description
Runner framework (sw.testing)	done	SDK module: suite/test/assert, lifecycle hooks, structured JSON results.
requires declarations	done	Runner manifests declare dependencies. Missing packages → clean skip.
Cleanup enforcement	done	s.track(type, id) auto-deletes in afterAll. No leaked state.
Warning tier	done	pass / fail / warning. No silent catch swallowing.
ICD runner migration	done	Refactor to sw.testing. Test logic preserved.
SDK runner migration	done	Refactor to sw.testing. Domain suites preserved.
Runner registry surface	done	/s/test-runners — list runners, run-all, results dashboard.

v0.7.2 — Package Runners + CI Gate

Step	Status	Description
Notes runner	done	requires: ["notes"]. 3 suites (crud, folders, tags-search), 12 tests.
Chat runner	done	requires: ["chat", "chat-core"]. 2 suites (conversations, messaging), 9 tests.
Schedules runner	done	requires: ["schedules"]. 1 suite (crud), 5 tests.
Workflow runner	done	requires: ["content-approval"]. 1 suite (lifecycle), 5 tests.
Renderer runner	done	requires: ["mermaid-renderer"]. 1 suite (contract), 4 tests.
Runner result API	done	POST/GET /api/v1/admin/test-runners/results. In-memory store, 4 Go tests.
CI integration	done	test-runners stage in Gitea CI. Playwright driver, wait-for-healthy.sh.
CI DinD networking fix	done	Resolve container IP via docker inspect — DinD port mapping doesn't expose to runner localhost.

v0.7.3 — Extension Shell Migration

Chat, Notes, and Schedules still use the old sw.shell.Topbar component, producing a double topbar (shell-injected + surface-owned). Migrate all three to the v0.7.0 shell contract (sw.shell.topbar.setTitle/setSlot), same pattern as the kernel surface migrations.

Step	Status	Description
Chat → shell topbar	done	Delete <Topbar>, use setTitle('Chat') + setSlot() for thread title/people button.
Notes → shell topbar	done	Delete <Topbar>, use setTitle('Notes') + setSlot() for action buttons.
Schedules → shell topbar	done	Delete <Topbar>, use setTitle('Schedules') + setSlot() for count/new button.
Update package runner tests	done	Shell-topbar test suite in each runner — assert no legacy Topbar, uses shell API.

v0.7.4 — Documentation + Deferred Surface Work

Step	Status	Description
Docs category grouping	done	Backend Category field on doc entries. Frontend groups sidebar by category with headings. Four categories: Getting Started, Platform, Extension Development, Operations.
Permissions & Groups guide	done	RBAC model, 7 permission slugs, system/custom groups, settings cascade, extension permissions.
Workflows user guide	done	Entry modes, stages, team roles, signoff gates, SLA, public forms, branch rules, Starlark hooks.
Starlark Reference	done	Sandbox constraints, 10 modules with function signatures, permission gates, example hook script.
Frontend JS Guide	done	Preact+htm runtime, 16 SDK modules with API reference, shell topbar patterns, CSS contract.
Extension config_section docs	done	Manifest schema, backend discovery, frontend contract, example component. Added to Extension Guide.
Docs content refresh	done	All 10 user-facing docs reviewed for v0.7.x accuracy: rebrand volume names, stale CSS vars, TLS_MODE env, self-hosted font notes, Architecture frontend section.
Team Admin Workflows split	done	722-line workflows.js split into 3 modules: workflows.js (router+CRUD), workflow-editor.js (editor+stages), workflow-monitor.js (assignments+monitor+signoff).
Stale CSS variable fix	done	--bg-2 references in sw-shell.css and sw-primitives.css replaced with --bg-secondary.

v0.7.5 — Headless E2E + CI Gate

Step	Status	Description
CI gating redesign	done	Fix path rules: VERSION/scripts/* deploy-only, docs/* no longer skips deploy, test-runners trigger on any code change.
E2E smoke test harness	done	ci/e2e-smoke-test.sh + ci/e2e-smoke-driver.js — Playwright visits every surface, asserts topbar, no JS errors, home link.
Screenshot-on-failure	done	Full-page screenshot + console log saved as CI artifacts.
CI pipeline integration	done	Re-enable test-runners stage, add e2e-smoke stage. Failure blocks merge.
Design docs landed	done	docs/DESIGN-storage-primitives.md, docs/DESIGN-extension-composability.md. Roadmap expanded through v1.0.

v0.7.6 — Code Hygiene + Test Coverage

Codebase analysis (April 2026) found 9 code smells from the chat gut and a test coverage gap in the store layer. Fix before v0.8.x kernel expansion.

Critical Fixes

Step	Status	Description
Remove channels from allowedViews	done	db_module.go — ext_view_channels does not exist. db.view("channels") crashes.
Fix workflow.html dead routes	done	Template calls /api/v1/channels/{id}/workflow/* — routes moved to /api/v1/public/workflows/. Update template to match. Also fixed RenderWorkflow handler to use route param as entry token.

Dead Code Removal

Step	Status	Description
Delete SeedTestChannel()	done	database/testhelper.go — inserts into nonexistent channels table.
Remove RunContext.ChannelID	done	sandbox/runner.go — vestigial field, unused by any module.
Remove webhook.ChannelID	done	webhook/webhook.go — vestigial struct field from chat era.
Fix stale comments	done	storage.go key format, prometheus.go route pattern, workflow_module.go param name — reference dead /channels paths.

Migration Hygiene

Step	Status	Description
Align SQLite migration numbering	done	SQLite 013 placeholder + renumber test_runner_type to 014. Compat rename in migrate.go.
Document missing 008	done	Comment in both 009 files explaining the gap (merged into 007).

Test Gap Closure

Step	Status	Description
store/ unit tests		Deferred to v0.7.8. Direct tests for PG + SQLite store implementations. Requires live DB. Target: ≥30% SLOC ratio.
workflow/ routing unit tests	done	82 tests: ResolveNextStage, ResolveStageByName, ParseStageConfig, evaluateCondition (all operators).
middleware/ unit tests	done	10 tests: RequirePermission, RequireAdmin, permission caching, RateLimiter (allow/deny/fail-open).
InstallPackage decomposition		Deferred to v0.7.8. 224 cognitive complexity. Split into: validate → create → DDL → permissions → triggers.

Bug Fixes

Step	Status	Description
Remove Admin Storage tab	done	Dead weight under System — backend endpoint preserved for future Monitoring use.
Fix backup download	done	sw.auth.token() → sw.auth._getToken(). Both "Download Backup" and server backup download buttons now work.

v0.7.7 — API Tokens + Extension Permissions

Personal access tokens (PATs) for programmatic API access, plus extension-declared user permissions for backend RBAC enforcement. Unblocks: E2E test auth, CI automation, extension permission gating, sidecar auth (v0.10.x).

API Tokens (PATs)

Step	Status	Description
api_tokens table + migration	done	id, user_id, name, token_hash, permissions (JSON array), expires_at, last_used_at, created_at. PG + SQLite.
Token store interface	done	Create, GetByHash, ListForUser, Revoke, CleanExpired, UpdateLastUsed.
Token generation endpoint	done	POST /api/v1/auth/tokens — name, permissions (subset of user's), expires_at. Returns token once (plain text), stored as SHA-256 hash.
Token management endpoints	done	GET /api/v1/auth/tokens (list mine), DELETE /api/v1/auth/tokens/:id (revoke).
Admin token endpoint	done	POST /api/v1/admin/tokens — creates token for any user. Explicit admin action, audit logged.
Auth middleware update	done	Accept Bearer arm_pat_... tokens alongside JWTs. Resolve user + scoped permissions from token. PAT requests skip refresh token flow.
Permission scoping	done	Token permissions ⊆ creating user's resolved permissions at creation time. Validated on create. If user loses a permission later, token retains it until revoked (git model).
Settings UI	done	Settings → API Tokens tab. Create, list, revoke. Show permissions, expiry, last used. Token value shown once on creation.
Admin UI	done	Admin → People → user detail → "Create API Token" action.
E2E test migration	done	Replace ci/e2e-smoke-test.sh login flow with PAT-based auth. Seed token via ARMATURE_BOOTSTRAP_PAT=true.

Extension-Declared User Permissions

Step	Status	Description
user_permissions manifest field	done	Extensions declare permissions users need: "user_permissions": ["image-gen.use", "image-gen.admin"].
Dynamic permission registry	done	On install: merge into valid permission set. On uninstall: remove. AllPermissions becomes KernelPermissions + ExtensionPermissions.
Group UI update	done	Admin → Groups → permission checkboxes include extension-registered permissions, grouped by source package.
permissions Starlark module	done	permissions.check(user_id, "image-gen.use") → resolves user's groups, returns bool. New permission: none required (read-only check against kernel data).
gate_permission manifest field	done	Optional. If set, ext_api.go checks this permission before calling on_request. Extension doesn't execute for unauthorized users.
req["permissions"] in request dict	done	ext_api.go resolves user permissions and includes them in the request dict. Extensions can check inline without the module.

v0.7.8 — Deferred Test Coverage

Remaining test gap items deferred from v0.7.6.

Step	Status	Description
store/ unit tests		Direct tests for PG + SQLite store implementations. Priority: packages, ext_data, workflows. Target: ≥30% SLOC ratio.
InstallPackage decomposition		224 cognitive complexity. Split into: validate → create → DDL → permissions → triggers. Each independently testable.
workflow/ engine integration tests		Engine lifecycle tests with mock stores (Start, Advance, Cancel, signoff gate).

---

Planned

v0.8.x — Storage Primitives

Design doc: docs/DESIGN-storage-primitives.md

The last major kernel expansion. Completes the primitive set that the entire extension ecosystem builds on. After v0.8.x, the kernel is feature-complete for 1.0 — all new capabilities are extensions.

v0.8.0 — files Module

Bridge ObjectStore (PVC/S3) into the Starlark sandbox. Extension-scoped key namespacing (ext/{packageID}/). Permissions: files.read, files.write. Metadata companions stored as sibling objects.

New: sandbox/files_module.go. Modified: runner wiring, permission constants.

v0.8.1 — workspace Module

Managed disk directories for tools that need a real filesystem (git, compilers, media tools). workspace.create(name) → scoped path. Permission: workspace.manage. Quota enforcement via WORKSPACE_QUOTA_MB.

New: sandbox/workspace_module.go. Modified: runner wiring, permission constants.

v0.8.2 — Capability Negotiation

Extensions declare environment requirements in manifest (capabilities.required, capabilities.optional). Kernel validates at install time. Runtime query via settings.has_capability(). Admin endpoint: GET /admin/capabilities.

Detected capabilities: pgvector, workspace, object_storage, s3, postgres.

New: handlers/capabilities.go. Modified: install handler, settings module.

v0.8.3 — Vector Column Type

"vector(N)" in db_tables manifest. Maps to native vector(N) on PG+pgvector, JSONB on PG without, TEXT on SQLite. New db.query_similar() with dual-path dispatch (native operator vs Go-side brute force).

Modified: ext_db_schema.go, db_module.go.

v0.8.4 — Extension Composability

Design doc: docs/DESIGN-extension-composability.md

Manifest declarations for slots (host surfaces declare injection points) and contributes (extensions declare UI contributions to other packages' slots). lib.require() relaxed from library-only to any package with exports — enables full packages (with surfaces, settings, UI) to also expose callable backend functions. sw.slots.renderAll() SDK helper. Admin slot aggregation endpoint.

The composability primitive that enables: LLM tool registration, UI contribution (toolbar buttons, image actions), and cross-extension function calls. No new tables, migrations, or permissions.

Modified: sandbox/lib_module.go (~1 line), handlers/extensions.go, src/js/sw/sdk/slots.js. New: handlers/admin_slots.go.

---

v0.9.x — Reference Extensions

The kernel is complete. This series proves the platform by shipping first-party extensions that exercise every primitive. These are extensions, not kernel code — they ship as .pkg files and can be uninstalled.

v0.9.0 — vector-store Library

Library extension: document ingestion, chunk storage, embedding persistence, semantic similarity search. Three-tier vector strategy (pgvector → JSONB fallback → TEXT fallback). Consumes: db.write, files.read.

v0.9.1 — llm-bridge Library

Library extension: model abstraction, tool-use routing, provider BYOK via connections. Exposes complete(), embed(), classify(), register_tool() to consumer extensions. Tool extensions (image-gen, code-exec, web-search) register at load time — LLM discovers available tools automatically. Consumes: connections.read, api.http, db.write.

v0.9.2 — file-share Extension

File upload via api_routes, storage via files module, download links, optional team/group ACLs via resource grants. First extension to exercise the full files primitive end-to-end.

v0.9.3 — code-workspace Extension

Managed code repositories via workspace module. Git clone/pull via api.http to Gitea/GitHub API. File browser surface. Structural indexing (AST-aware) for code navigation. Consumes: workspace.manage, files.write, api.http.

v0.9.4 — image-gen + image-edit Extensions

Image generation via external APIs (DALL-E, Stable Diffusion). Registers as LLM tool with llm-bridge. Contributes regen/edit buttons to chat:image-actions slot. Image-edit extends with inpaint, outpaint, upscale, and style transfer — each contributing action buttons to the same slot independently. First full demonstration of the composability pattern: three extensions contributing UI to a fourth's surface.

v0.9.5 — Chat System

Generic 1-to-N human messaging. chat-core library (conversation/message CRUD) + chat surface. Declares chat:message-actions, chat:image-actions, chat:composer-tools slots for extension contribution. LLM participation as a follow-on consuming llm-bridge. Consumes: db.write, realtime.publish, files.write (attachments).

---

v0.10.x — Sidecar Tier + Polish

v0.10.0 — Sidecar Tier

Autonomous out-of-process extensions for workloads that can't run in Starlark: ML inference, media transcoding, language servers, local git operations. The kernel does NOT manage container lifecycle — sidecars are independent processes (Docker, systemd, bare binary) that connect inward to the kernel, following the cluster registry pattern:

• Sidecar boots with cluster config (instance URLs, credentials).
• Connects and self-registers, declaring capabilities ("sklearn", "gpu", "ffmpeg", "sentence-transformers").
• Goes inert until an admin authorizes (same flow as extension permissions).
• Kernel dispatches work; sidecar returns results.
• Heartbeat keeps registration alive; missed heartbeat = eviction.

Communication channel TBD (WebSocket over existing event bus, gRPC, or HTTP callback model — design doc required). Authentication via mTLS certs or registration tokens.

This model eliminates k8s RBAC dependency for local deployments. The kernel stays thin — it accepts registrations and dispatches work. The sidecar brings its own compute and manages its own lifecycle.

v0.10.1 — Native Dialog Audit

Replace prompt()/confirm()/alert() with sw.dialog SDK primitives. Accessible, themed, promise-based. Small kernel SDK change that improves every surface.

v0.10.2 — Stability + Migration Tooling

Proper versioned migrations (post-MVP discipline). armature migrate CLI command. Backup/restore validation against reference dataset. Pre-1.0 schema freeze.

---

v1.0.0 — Stable Release

The kernel API surface is frozen. Extensions are the product.

Gate criteria:

• All kernel Starlark modules documented with examples
• All api_routes covered by OpenAPI spec
• Upgrade path tested from v0.8.0 → v1.0.0
• At least 3 reference extensions shipped and tested (vector-store, llm-bridge, chat)
• At least 1 cross-extension composability demo (image-gen → chat:image-actions)
• Headless E2E green on PG + SQLite
• armature-ca.sh + mTLS deployment guide
• Single-binary + Docker + K8s deployment paths documented

---

Post-1.0 Horizon

These are candidates, not commitments. Each requires a design doc.

• Federation — cross-instance package sharing, identity federation
• Package marketplace — signing, review, discovery registry
• Desktop app — Tauri wrapper for local-first deployment
• Offline/sync — SQLite-first with PG sync for field deployments
• Multi-tenant SaaS mode — tenant isolation at the team boundary

---

Design Principles

Principle	Implication
Extensions are the product	Chat, tasks, LLM, vector search, file sharing — all extensions. Zero kernel awareness of domain logic.
Kernel stays thin	New kernel primitives require justification. If it can be an extension, it must be.
Progressive enhancement	Every feature works on SQLite. PG adds performance. pgvector adds native vectors. S3 adds scalable storage.
KISS-first	No unnecessary dependencies. Preact+htm (3KB), single binary, dual-DB from one codebase.
Changeset discipline	Each CS independently CI-green. Design docs as implementation contracts.
Pre-1.0 migration freedom	Schema changes fold into existing migrations. Post-1.0: proper versioned migrations.

---

Design Decisions Log

Decision	Rationale
Tasks → extension	Three trigger primitives replace the monolithic scheduler.
Sessions removed	Workflow instances with dedicated storage replace kernel sessions.
custom stage mode	Delegates to a surface package, proving extension composability.
Providers removed from kernel	Connections + Starlark provider.complete as the interface.
Kernel permissions simplified	6 platform permissions. Extensions define their own.
Preact+htm retained	3KB runtime, no build step, KISS.
Single Docker image	Go binary + assets + migrations.
Admin → RBAC group	Grant check replaces role check.
Settings cascade	Scope auth + user_overridable. Two orthogonal axes.
No new migrations pre-MVP	Proper versioned migrations post-MVP.
Chat as extension, not kernel	Zero kernel awareness. Proves extensibility thesis.
PG as consensus layer	UNLOGGED node_registry + LISTEN/NOTIFY. No etcd/Consul/Redis.
Two trigger tiers	Extension-declared (full sandbox) vs user ad-hoc (restricted).
Builtin package rationale	Must enhance kernel surfaces or demonstrate platform capabilities.
Two-slot topbar model	Left slot (title/branding) + center slot (flex: 1, tabs/pickers). Two named slots cover every navigation pattern: simple title (Pattern A), flat tabs full-width (Pattern B), category tabs + surface-owned sidebar (Pattern C). Surfaces without sub-items get full content width; surfaces with hierarchical navigation add their own sidebar below the topbar. Shell provides the stage; surface decides the theater.
All four primary surfaces migrate to shell topbar	Admin was "keep custom topbar" initially. The two-slot model makes it unnecessary — category tabs fit in the center slot, sidebar is surface-owned below. One topbar implementation replaces four. Bell + user menu + reactivity come free on every surface.
Settings / Team Admin → flat tabs (Pattern B)	Both had thin sidebars (~140px) that consumed width without justification. 5–6 sections fit cleanly in topbar tabs. Full-width content is a better use of space for these surfaces.
Team Admin Groups removed	37-line read-only dead-end. Admin Groups has full CRUD. Restore when properly designed.
Docs is the reference surface	Only surface with shell Topbar, bell, user menu, inline errors. Others converge.
Surface runners over expanding ICD	Different test tier, different failure class.
Headless E2E via Playwright	Runners produce structured JSON; Playwright navigates and reads output.
db module is the structured store	No separate KV primitive. Extensions declare tables — infrastructure exists.
files module rides ObjectStore	No new kernel tables. Metadata as companion objects. Works on PVC and S3 identically.
workspace for real filesystem	Flat blob store can't serve git/compilers/ffmpeg. Managed disk paths are the explicit escape hatch.
Capability negotiation at install	Fail loud with actionable message, not silently at runtime.
Vector column with three-tier fallback	Works everywhere, works fast with pgvector. Documentation > hard gates.
Sidecar deferred to v0.10.x	HTTP module covers external APIs. Local compute is a v0.10.x concern. Sidecars connect inward (like cluster nodes), not spawned outward (no k8s RBAC needed).
No second scripting runtime	Starlark for glue, HTTP for external compute, sidecars for local compute. Three tiers with clear boundaries. Lua/WASM middle tier rejected — Turing-complete enough to be dangerous, not isolated enough to be trustworthy.
Reference extensions prove the platform	First-party .pkg files, not kernel code. Uninstallable. Dogfooding.
Composability via slots + exports, not kernel	sw.slots + sw.actions + lib.require() are the composition primitives. Manifest declarations add admin visibility. No inter-extension message bus — too complex, too fragile.
Gut cleanup before kernel expansion	Codebase analysis found 9 smells from the chat gut. Fix before adding new primitives — dead views and broken templates compound when new code builds on them.
PATs over OAuth client credentials	PATs are simpler (one table, one middleware check), cover all use cases (E2E tests, CI, sidecars, user automation), and follow the git hosting convention users already understand. OAuth adds client registration, token exchange, scopes — complexity without benefit for a self-hosted platform.
Extension-declared user permissions	Extensions register custom permissions in the kernel's group system. Admin assigns them to groups. Backend enforces via permissions.check() or gate_permission. Frontend hides via sw.can(). The kernel owns the permission registry; extensions contribute to it.
Admin token as explicit action	Following Venice.ai pattern — admin token creation is a separate endpoint, audit logged, requires surface.admin.access. No ambiguity about the privilege level.