Pantheon Hub — Product Requirements Document

Version: 3.1 (Wave-2 sequenced) · Date: 2026-05-05 · Author: Athena

This is the single canonical PRD for Pantheon Hub. It supersedes and absorbs: - pantheon/docs/pantheon/PRD.md v1.0 (2026-03-02 — pre-constellation, predates fleet rebuild) - pantheon/docs/pantheon/DASHBOARD-BIBLE.md (2026-03-26 — vision + architecture decisions) - constellation-athena/docs/specs/pantheon-hub-navigation-restructure.md (2026-04-17 — page architecture) - constellation-athena/docs/specs/pantheon-hub-v2-session-plan.md (2026-04-22 — implementation plan) - constellation-athena/docs/specs/pantheon-hub-prd-2026-04-28.md (2026-04-29 — PM MVP scope cut) - constellation-athena/docs/specs/planning-view-reference.md (2026-04-22 — planning surface details) - constellation-athena/docs/specs/athena-subpage-architecture.md (2026-04-08 — Athena page detail) - constellation-athena/docs/audits/pantheon-hub-project-mapping.md (2026-04-08 — comprehensive map)

Source documents move to pantheon/docs/pantheon/archive/superseded-2026-04-30/ on consolidation.

1. Problem

Multi-agent work across multiple machines and surfaces creates a visibility problem. Without a central coordination layer:

• A coding task discussed in a war room generates work that is never tracked.
• Project context lives in scattered repos, devlogs, and chronicles — agents and operator both pay re-orientation cost every session.
• The operator cannot answer "what is happening across the fleet right now?" without grep-walking the filesystem.
• Strategic decisions get made and lost — there is no single place where Athena's directives, project priorities, and assessment scores live.
• Three separate task systems coexist (file backlog, Dropbox dual-write, table mirrors) without a unified read surface.

The Hub exists to solve this. Every agent reports to the Hub. Every project lives there. Every decision lands there. The operator opens one URL and sees the fleet.

2. Vision

The Pantheon Hub is a strategic portfolio command surface — not an ops console, not a project management tool. It exists to answer three questions:

1. Where are my resources going? (current state — agents, projects, costs, dispatches)
2. Where should they be going? (strategic — Athena directives, priorities, milestones)
3. What is producing value and what is consuming it? (financial — Plutus oversight)

The primary user is Boas / Kairos (human operator) working with Athena (strategic AI counsel). Agents (the constellation) consume the Hub for context and write to it as they work. The Hub is the gravitational center — every surface else (Discord, Claude Code sessions, Arsenal, war rooms, n8n) reports here.

Hierarchy

Portfolio (all projects at a glance)
  └── Project (strategic container — long-lived, multi-repo, multi-agent)
        ├── Repos (1..N per project, e.g. hub+spoke patterns)
        ├── Clients (downstream consumers / customers)
        ├── Constellations (agent groups working on the project)
        ├── Tasks (live fleet_tasks scoped to project)
        ├── Milestones (planned + in-progress + complete, descending)
        ├── Decisions (war_room_decisions FK'd to project)
        ├── Artifacts (briefs, mockups, fleet_shares deep-links)
        ├── Activity (event timeline across all repos)
        └── Strategy Docs (athena/metis-authored prep, athena-012)

3. Users + Stakeholders

The Hub is human-in-the-loop by design. Agents execute, research, monitor; Kairos approves, steers, decides. There is no path to autonomous merge. Even the Night Shift protocol operates against a queue Kairos pre-approved.

4. Architectural Principles

These are non-negotiable design commitments.

1. Live data only. No mocks in the dev runtime. No MSW handlers active. If a section can't show real data yet, it says "Not connected" — never displays template content. Origin: 2026-04-28 fleet policy after Hub v3 0 ACTIVE on real backend because MSW shapes diverged.

2. Append-only everywhere. Events, war room outputs, decisions, chronicle entries, audit trails. Nothing is mutated. Reproducibility, time-travel debugging, trustworthy audit by construction.

3. Projects are the gravitational center. Everything is a projection within project context. The flat 39-item repo dump becomes ~25 meaningful projects with repos nested inside.

4. Goal-backward planning. Set an epic / end-goal with Athena. Build roadmap backwards. Progress reflects distance to goal, not phase-count.

5. Descending roadmap. Future goals at top, completed work sinks. The destination leads.

6. Protocol-agnostic planning layer. GSD, Superpowers, or whatever comes next wraps underneath. The Hub does not assume a specific execution protocol.

7. Depth through navigation, not density. Top-level shows health + activity signals. Details live one click deeper. No page requires scrolling 30 items to find one.

8. Every agent page is an operations console. Not a profile card. Open Hermes → see active monitors, pending research, recent findings — not a static description of what Hermes is.

9. Real data or honest "Not connected". Mock data in production creates false confidence.

10. Pantheon brand SSOT. JetBrains Mono primary, dark-only, phosphor accent #7CE4C9. Defined in constellation-apollo/projects/pantheon/pantheon.brand.json. Never cross-wire with Galdr (EB Garamond / Iron Manuscript) or boas-dev (JetBrains Mono dual-mode).

11. Single PR per task. No bundling. Scope-hygiene rule from 2026-04-25.

12. 3-surface deploy smoke after every restart. Backend curl + frontend HTML render + bundle hash check. Backend-only smoke false-positives on stale SPA.

13. Offline-capable, Supabase-optional. Hub must work from a laptop without internet (SQLite local). Supabase is the productization path (offering Pantheon to clients), not a hard dependency.

14. Wire existing gods, do not add new ones. The roster is sufficient. Priority is connecting them — provisioning runtimes, closing dispatch loops — not expanding the cast.

15. Zod at all boundaries. Every external input validated before processing. Hard-fail on invalid data; never clean up afterward.

5. Page Architecture

The Hub has three navigation sections: COMMAND (the daily heartbeat), CONSTELLATION (per-agent ops consoles), SYSTEM (infrastructure visibility).

Navigation

COMMAND
  ◈  Overview        /                Fleet pulse + activity + alerts
  ◇  Portfolio       /portfolio       Projects, clients, engagements
  ⚔  War Room        /war-room        Active + historical sessions

CONSTELLATION
  ✦  Athena          /athena          Strategic governance, assessments
  ⬢  Atlas           /atlas           Fleet ops, skills, Argus, enforcement
  ◇  Metis           /metis           Strategic research, proposals
  ⚡  Hermes          /hermes          Intelligence, monitoring, feeds
  ☀  Apollo          /apollo          Creative output, deliverables
  $  Plutus          /plutus          Costs, budgets, financial oversight
  ☢  Hades           /hades           Security, credentials, guard rails
  ⚙  Proteus         /proteus         Infrastructure, pipelines, compliance
  ◑  Mnemosyne       /mnemosyne       Knowledge graph, memory, Aion

SYSTEM
  ♡  Pulse           /pulse           Health rings, probe history
  ⊕  Fleet           /fleet           Machines, agents, connectivity
  ◉  SSOT            /ssot            Fleet table browser
  ▶  Dispatch        /dispatch        Pipeline, work items, ask_human

5.1 COMMAND Section

Overview (/)

The page Boas opens first. Answers: "Is everything OK? What happened? What needs attention?"

Layout: - Fleet status bar (top): online agent count, machine count, Pulse status, active war rooms, critical Argus findings - Activity feed (left, 60% width): cross-agent event stream, time-sorted, max 20 items - Attention panel (right, 40% width): Argus findings, stale agents, ask_human pending, pipeline blocked - Active work (right, below Attention): live war rooms + active fleet_tasks - Constellation status (bottom, full-width): 9 agent dots, color = last_seen recency, click → constellation page

Data sources (all real): /api/agents, /api/pulse, /api/argus-status, /api/war-room/sessions?phase=active, /api/cross-agent-feed, /api/pipeline?status=waiting_approval, /api/dispatch-responses?status=pending, /api/active-tasks.

Portfolio (/portfolio)

Strategic project + client view. Three collapsible sections, each showing project cards:

• Agency — type: client-engagement (revenue work). Visually prominent, shown first.
• Products — type: product (Pantheon Hub, Arsenal, Nexus, Trading War Room). Distinct accent.
• Constellation — type: agent-constellation (compact card grid of 9-10). Links to constellation page, not project detail.

Filters: lifecycle (active/maintenance/archived), priority (focused/critical/planning/paused), type, sort (activity/progress/name).

Each card surfaces: project name, workspace badge, lifecycle chip, priority chip, owner agent, current milestone, progress bar (milestones complete / total), last-activity timestamp (max of: commits, war_room_sessions, fleet_tasks updates), repo count, brief excerpt, deep-link to detail page + latest share.boas.dev artifact + latest war room session.

Project Detail (/project/:slug)

Sticky header with progress ring (Athena assessment score if available, milestone-derived otherwise), project name, priority + lifecycle badges, subtitle stats. Below: Overview Strip (epic goal display + edit, Athena assessment row conditional, milestone progress bar, focus derivation).

Tab structure (PR-v3-03 8-tab refactor, merged 2026-04-25 PR #21): flat URL /project/:slug/:tab. 8 canonical tabs, classified by purpose (group classification is visual badge only, not URL prefix):

Generic EntriesTabList consumes canonical /api/v3/projects/:slug/entries?type= per PR #19 contract. Fallback ?view=raw-docs renders Wave-2 filesystem-scanner view per spec §13 v3-06 escape hatch.

Supersession sidebar widget (Q5) renders on every project page — sidebar layout only (NOT breadcrumb, NOT modal, NOT inline chain walker). Surfaces the supersession chain for the project's canonical research/context entries.

Strategy Docs tab (athena-012, post-Hub-functional, blocked on PM MVP closure + Hub auth) — would slot as a 9th tab between artifacts and code, scoped private + agent-accessible. NOT in current 8-tab spec.

Empty Tasks tab shows clean empty-state, NOT <NotConnected> (project IS connected, just no tasks). <NotConnected> reserved for tabs whose backing data path is genuinely absent (Issues today).

War Room (/war-room)

Session history + governance decisions. Active sessions pinned to top with live indicator. Otherwise unchanged from current implementation — well-structured against real war_room_* tables.

5.2 CONSTELLATION Section

Every constellation page follows the same shell:

• Agent header — name, role, last_seen, machine, status (online/stale/offline)
• Domain stats strip — 3-5 key metrics, agent-specific
• Tab bar — Activity (universal) + 2-4 domain-specific tabs

The Activity tab is universal — events filtered by agent_id. Recent commits, war room participation, task completions.

Per-agent domain tabs:

Honest "Not connected" placeholders where real data path absent today: Metis (Arsenal-isolated, no Hub registration), Hermes Intelligence (Arsenal queue not yet wired), Apollo Gallery (Dropbox mockups path not yet wired), Apollo Brand (table needed), Plutus Budget (config system needed), Plutus Optimization (logic needed), Hades Findings (security_findings table needed), Hades Guard Rails (state surface needed), Mnemosyne Knowledge (Aion CLI integration).

Athena chat shell stays as a placeholder. Don't make it prominent — it doesn't work yet (Mastra agent backend not provisioned).

5.3 SYSTEM Section

Pulse (/pulse)

Health monitoring. Replaces "System" naming — Pulse is the actual health system with ring probes. 6 health rings with status (healthy/degraded/critical), findings list with severity, 30-day trend sparklines per ring, "Check Now" button. All real from pulse_reports table.

Fleet (/fleet)

Machine + agent inventory. Machines section: hostname, type, location, last_seen, CPU/RAM/disk sparklines (24h). Agents section: name, role, machine, model, last_seen, current priority. Each agent row links to its constellation page. All real.

SSOT (/ssot)

Fleet table browser. Lookup tool — fleet_agents, fleet_services, fleet_repositories, fleet_credentials, fleet_machines, fleet_naming_conventions. Works as-is.

Dispatch (/dispatch)

Pipeline + human-in-the-loop. ask_human queue at top (most urgent), active work items by stage, dispatch log, dispatch sessions. All real from pipeline + dispatch tables.

6. Data Model

Core project model — fleet_projects_registry

-- Naming: fleet_projects_registry (not fleet_projects, not fleet_portfolio_projects)
-- Renamed 2026-04-25 to avoid collision with cornerstone fleet_projects (Supabase project registry).

CREATE TABLE fleet_projects_registry (
  id              uuid PRIMARY KEY DEFAULT gen_random_uuid(),
  slug            text NOT NULL UNIQUE,
  name            text NOT NULL,
  project_type    text NOT NULL CHECK (project_type IN
                    ('client-engagement','agent-constellation','internal-initiative','subsystem')),
  workspace       text NOT NULL CHECK (workspace IN
                    ('Agency','Products','Constellation','Personal')),
  status          text NOT NULL CHECK (status IN
                    ('proposal','active','maintenance','shipped','shelved','archived')),
  priority        text CHECK (priority IN ('focused','critical','planning','paused')),
  owner_agent     text,
  brief_markdown  text,
  created_at      timestamptz NOT NULL DEFAULT now(),
  updated_at      timestamptz NOT NULL DEFAULT now()
);

Renamed columns from earlier draft: lifecycle → status, type → project_type. 4-value project_type enum: client-engagement (3 rows), agent-constellation (10 rows), internal-initiative (renamed from earlier product + roadmap, 9 rows), subsystem (4 rows). Coverage today: 26 rows after Athena PRD-Ath2 (2026-04-29, registry expansion shipped) + fleet-backup-redesign (2026-04-30, v31.0 ratification PR #33). Workspace distribution: Products 13, Constellation 10, Agency 3 (Personal unused so far).

Related tables

-- Repo membership (M:N)
CREATE TABLE fleet_project_repos (
  project_id  uuid NOT NULL REFERENCES fleet_projects_registry(id) ON DELETE CASCADE,
  repo_name   text NOT NULL,
  role        text CHECK (role IN ('primary','secondary','subsystem','archive')),
  PRIMARY KEY (project_id, repo_name)
);

-- Milestones (descending roadmap, goal-backward)
CREATE TABLE fleet_portfolio_milestones (
  id           uuid PRIMARY KEY DEFAULT gen_random_uuid(),
  project_id   uuid NOT NULL REFERENCES fleet_projects_registry(id) ON DELETE CASCADE,
  title        text NOT NULL,
  target_date  date,
  status       text NOT NULL DEFAULT 'planned' CHECK (status IN
                 ('planned','in_progress','complete','missed','cancelled')),
  ordering     int NOT NULL DEFAULT 0
);

-- Project dependencies
CREATE TABLE fleet_portfolio_dependencies (
  project_id              uuid NOT NULL REFERENCES fleet_projects_registry(id) ON DELETE CASCADE,
  depends_on_project_id   uuid NOT NULL REFERENCES fleet_projects_registry(id) ON DELETE CASCADE,
  kind                    text CHECK (kind IN ('blocks','informs','shares_repo')),
  PRIMARY KEY (project_id, depends_on_project_id)
);

Foreign-key bridges (existing tables, optional FKs)

• fleet_tasks.project_id uuid NULL REFERENCES fleet_projects_registry(id) — which project does this task serve. Coverage 94.9% as of 2026-04-30 (35 NULL of 683 rows). Decreased from 95.5% snapshot because new tasks accreted faster than re-scan picked them up. PRD-M1 resolver + Ath2 registry expansion handle the backfill mechanic. NOT NULL constraint deferred until backfill held permanent.
• war_room_sessions.project_id uuid NULL REFERENCES fleet_projects_registry(id) — session focused on a project.
• fleet_shares.project_id uuid NULL REFERENCES fleet_projects_registry(id) — published brief tied to project.
• crm_deals.project_slug text NULL — TEXT reference (cross-database). Validation app-level. Deferred until Hub coupling needed.
• research_reports.project_slug text NULL — same pattern. Deferred.

Strategy docs (post-Hub-functional)

Per backlog/pantheon-hub--strategy-docs-surface.md (athena-012, blocked on PM MVP closure + Hub auth):

CREATE TABLE fleet_documents (
  id              uuid PRIMARY KEY DEFAULT gen_random_uuid(),
  project_id      uuid REFERENCES fleet_projects_registry(id),
  document_type   text CHECK (document_type IN
                    ('meeting-prep','negotiation-prep','client-profile','sector-delta',
                     'decision-record','strategy-narrative','post-meeting-debrief')),
  title           text NOT NULL,
  body_markdown   text NOT NULL,
  author_agent    text NOT NULL,
  meeting_id      uuid REFERENCES meetings(id),
  access_class    text CHECK (access_class IN ('private','internal','shared')) DEFAULT 'private',
  tags            text[] DEFAULT '{}',
  created_at      timestamptz DEFAULT now(),
  updated_at      timestamptz DEFAULT now()
);

RLS: anon SELECT only when access_class='shared'. Service-role full access. Single-user (Boas) authenticated access via Supabase Auth.

Themis-PM bridge

File backlog ({repo}/backlog/{area}--{slug}.md) is the authorship surface. The fleet_tasks table (Pantheon Supabase) is the dashboard / query surface. Themis (Metis-owned subsystem) mirrors files → table on schedule. One-way bridge — Themis reads files, writes table, never writes back.

Agents author via files. Hub reads via table. Cross-repo task views query the table. Marshal dispatch reads files (authority) but cross-checks table for recency.

7. Surfaces / Integration Points

Dashboard

React 19 + Vite SPA, served by Hub Express server. shadcn/ui + Tailwind CSS. TanStack Query + query key factory. 6 routes (Portfolio, Athena, Fleet, Dispatch, System/Pulse, Project Detail). Pantheon brand applied via pantheon.brand.json token bridge.

REST API

~70 endpoints on http://localhost:3847/api/. Bearer token auth for protected endpoints; localhost bypass for dev. v3 API parallel mount at /api/v3/* during PR-v3 cutover (2026-04-25):

• GET /api/v3/projects — list, registry-backed
• GET /api/v3/projects/:slug — detail with milestone + tab counts
• GET /api/v3/projects/:slug/entries — knowledge entries from harvester (when live)
• GET /api/v3/review — sensitivity + curation queues

MCP

Stdio transport. Context management (hub-set-context), telemetry (hub-task-start, hub-task-complete, hub-ping), repository management, planning sessions, pipeline operations, system diagnostics. Stateless per-request — no session state persisted.

This is how agents authenticate as themselves at session start. Boas's CLAUDE.md per-repo files reference hub-set-context + hub-ping + hub-task-start as session-start conventions.

Discord (OpenClaw)

Kronos and Hermes connected via OpenClaw sandboxes for Discord-side operations. Connection from Docker sandbox to Hub via host.docker.internal:3847.

Pipeline (Arsenal Dispatch)

Work items enter via REST or MCP, flow through stages with typed outputs + transition audit, dispatch to Arsenal's /execute endpoint when reaching execution stage. Arsenal HTTP server is Phase 65 pending — Pantheon's side of the loop is closed; Arsenal cannot yet receive dispatched work items.

8. Sequencing

Current wave: PM MVP (in flight, 2026-04-28 → )

Goal: project/task logging surface up and running with REAL data, fastest path. Operator mandate: "smallest viable, ship fastest."

Ground truth (verified 2026-04-30T13:00Z against remote master + Pantheon Supabase): - fleet_projects_registry exists, 26 rows (25 post-Ath2 + fleet-backup-redesign 2026-04-30) - fleet_tasks.project_id FK exists, 94.9% coverage (35 NULL of 683 rows). Decay from 95.5% is task accretion outpacing re-scan, not regression. - All 26 brief_markdown ≥ 300 chars (verified — distribution 301–961 chars, mean ~365) - Themis-PM file→table mirror works (pnpm scan in metis-themis/) - Hub frontend Portfolio + Project Detail render real backend (PR #19 + #23 merged 2026-04-25; PRD-A3 portfolio render fix shipped PR #30 2026-04-29) - Hub backend /api/fleet-tasks?project_id={UUID} LIVE (PR #29 merged 2026-04-29) - Project list/detail surface id: string UUID per card (PR #32 merged 2026-04-29) — required for Tasks tab to call useFleetTasks({project_id}) - MSW stripped from dev runtime (PR #28 merged 2026-04-29) - PR-v3-03 8-tab L3 + /review operator dashboard SHIPPED (PR #21 merged 2026-04-25) — 3 review queues live (sensitivity, curation, taxonomy_unknowns), 27 seeded curation rows - PR-v3-02a 3 LOW nits cleanup SHIPPED (PR #24, pantheon-007 resolved)

MVP task status (as of 2026-04-30T13:00Z):

Acceptance for PM MVP closure: - [x] /portfolio renders 26 real project cards (no MSW fallback) — PRD-A3 fix landed - [x] Click any card → /project/:slug Tasks tab populated with real fleet_tasks rows — PRD-A1 merged - [x] Each project's Tasks tab row count matches SELECT count(*) FROM fleet_tasks WHERE project_id = (SELECT id FROM fleet_projects_registry WHERE slug=$1) — verified post-merge against pantheon-hub (200-row cap, sort transitions match spec) - [x] All 26 brief_markdown ≥300 chars - [x] Zero MSW handlers active in dev runtime — PRD-A2 stripped - [ ] 3-surface deploy smoke green on hub launchd restart — verify on next restart (Wave-2) - [ ] fleet_tasks.project_id NULL count ≤ 12 — currently 35; rerun pnpm scan post-A1 (Wave-2)

PM MVP wave-1 CLOSED. All UI deliverables shipped; remaining acceptance items (deploy-smoke automation, NULL count drop) are Wave-2 ops.

Deferred (filed to backlog, NOT gating Wave-1)

• athena-013 — Hub backend stubs for /api/v3/projects/:slug/strategic-counts + /supersession-chain (4 console 404s on project detail page; pre-existing master defect surfaced during A1 smoke).
• themis enum migration — fleet_tasks.status + priority CHECK constraints to enforce single vocabulary at DB layer (long-term fix; client-side sort normalization is interim band-aid).
• A1 worktree cleanup — git worktree remove ~/Dev/_K4120S/pantheon-prd-a1 (branch deleted on remote).

Wave-2: Hub Functional (in flight, 2026-05-05 → )

Goal: Boas opens Hub and uses it end-to-end without crashes, dead-renders, or undisclosed-mock surprises. Stable URL + auth come last in the wave; structural correctness first.

Sequencing principle: structural truth before cosmetics. Each task either closes a regression, removes a silent-degrade, or wires a real-data surface. Order = dependency-respecting, not priority-ranked — every item is "required to call Wave-2 done."

Wave-2 ship list

Wave-2 acceptance

• [ ] PR #36 merged + deployed (W2-01)
• [ ] All 5 missing fields populated on /api/v3/projects/:slug (W2-02) — verify via curl + FE renders without nulls
• [ ] athena-013 endpoints either ship 200 or removed from FE call sites (W2-03) — console clean of repeat 404s on project detail
• [ ] fleet_tasks.project_id NULL count ≤ 12 (W2-04) — verify via Pantheon Supabase SELECT count(*) FROM fleet_tasks WHERE project_id IS NULL
• [ ] launchd-driven smoke green on hub restart (W2-05) — exit 0, three checks log success
• [ ] Zero "lorem-ipsum / mock-data" surfaces on Hub (W2-06) — every component either renders real data or "Not connected" sentinel
• [ ] At least 3 projects render branded /project/:slug (W2-07)
• [ ] Single-user login flow works end-to-end (W2-08)
• [ ] Hub reachable at stable hostname from Boas's machines (W2-09)

Out of scope for Wave-2 (deferred to Wave-3+)

• Multi-user auth, RLS-per-user, shared Hub
• Per-agent ops console layout (Wave-3)
• Athena Priorities tab inline edit (Wave-3 — depends on agent_priorities table decision pantheon-005)
• Strategy Docs tab on project detail (Wave-3 — athena-012 dependency)
• System-section restructure (Wave-4)
• Motif primitives + design library (Wave-5)

Wave 3: Constellation Ops Consoles

After PM MVP + Hub Functional. Per nav-restructure spec §CONSTELLATION:

Wave 4: System Section

Wave 5: Cosmetics Resume

After structural truth lands. Only then:

• Motif primitives (.stripe, .status-dot, .coordinate-gutter, .telemetry-strip, .sec-expand) — apply per Hub Design Bible
• Per-project theme injection on /project/:slug reading {slug}.brand.json
• /apollo/gallery design library
• Narrative stream + mockup gallery

Long-horizon (post-MVP, separate roadmaps)

• Athena chat backend — Mastra agent, conversation persistence, Mnemosyne integration. Separate spec.
• Sprint model — multi-repo coordinated work UI on Portfolio (conditional section, only when sprint active). Schema not yet built.
• SSE + real-time updates — event bus, smart polling, live updates on Overview / Portfolio / Fleet / Dispatch.
• Arsenal HTTP server (Phase 65) — Arsenal accepts EXEC_REQ. Closes the dispatch loop.
• Topology graph (React Flow) — visualization of WORLD.md topology. Researched, not specced.
• Pantheon Harvester v1 — separate roadmap (docs/specs/pantheon-harvester-v1-spec.md). Knowledge ingestion via n8n + classifier + adapters. Descoped from PM MVP per PRD v2.3 §6.
• Athena sub-pages (/athena/assessments, /athena/directives, /athena/priorities) — governance layer above MVP. Specced in athena-subpage-architecture.md.
• Constellation action surface — Plutus financial reports, Metis strategy reports, Hermes research dispatches accessible from per-project pages.
• agent_priorities table — currently NULL athena_read on cards; consumer treats null as "no priority." Three options under decision (pantheon-005): dedicated table, harvester subtype, or defer indefinitely.
• taxonomy_unknowns_queue — currently derived from curation_queue.queue_reason='type_ambiguous'. Promote to dedicated table only if classifier audit needs split semantics (pantheon-006).
• Pipeline closure end-to-end — Pantheon dispatches but cannot deliver results back to originating surface (e.g., Discord webhook to notify requester). Deferred.
• Plutus runtime — cost endpoints exist, no execution surface provisioned.
• Hades runtime visibility — security findings surface needed.
• Mnemosyne Hub registration — runs as Ollama subsystem on K20N0S (192.168.2.195), no fallback.

9. Acceptance — Long-Term

Pantheon Hub is "complete" when:

• Every surface reports to the Hub (Discord, Claude Desktop, Arsenal, CLI, Dashboard, n8n, war rooms). No blind spots.
• Every agent has an operational page populated with real data — no mocks, no <NotConnected> for primary tabs.
• Boas opens the Hub and answers "what is the fleet doing right now?" in under 30 seconds.
• Athena can write priorities to any agent via the dashboard and the agent sees them at next session.
• Strategic documents (meeting prep, negotiation prep, sector deltas) live in Hub and are reachable per-project under auth.
• Pipeline is closed end-to-end: work item enters → flows through stages → dispatches to Arsenal → returns results → notifies originating surface.
• Audit trail is unbroken: from any line of code, walk backward to the human decision that caused it.
• Hub runs offline on a laptop. Online sync to Supabase optional.

10. Out of Scope (current wave)

11. Constraints (non-negotiable)

1. Live data only. Memory feedback_no-mockups-live-only (2026-04-28). PRD-A2 strips MSW from dev runtime.
2. Pantheon brand. JetBrains Mono + dark + #7CE4C9. Memory feedback_brand-ssot-cross-wire-3way. SSOT lives in constellation-apollo/projects/pantheon/, NOT pantheon/docs/.
3. 3-surface deploy smoke. Backend curl + frontend HTML route + bundle hash. Memory feedback_deploy-smoke-must-include-frontend.
4. PM discipline. Each PRD-NN task = one fleet_tasks row tied to pantheon-hub project_id. Memory feedback_pm-system-discipline.
5. Single PR per task. No bundling. Memory feedback_clean-branch-scope-applies-during-cadence.
6. No --no-verify. No new dependencies for MVP scope.
7. Append-only data. No mutation of events, decisions, audit trails, chronicle entries.
8. Cross-room dispatch coordinator-only. Memory feedback_cross-room-coordinator-only. Per-room workers stay in their rooms.
9. Live data verification before claims. Memory feedback_athena-quality-bar-self-qa (2026-04-30) — Athena MUST inspect rendered artifacts before surfacing links.

12. Roles

13. Open Decisions

These are tracked as athena/decisions backlog items, blocked on data signal or operator input:

• agent_priorities table fate (pantheon-005) — three options: dedicated table, harvester subtype, defer indefinitely. Real path returns athena_read: null until resolved.
• taxonomy_unknowns_queue (pantheon-006) — derive from curation_queue.queue_reason='type_ambiguous' (current, working — /review queue uses this). Promote to dedicated table only if classifier audit needs split semantics.
• ~~PR-v3-02a 3 LOW nits cleanup (pantheon-007)~~ — RESOLVED by PR #24 merged 2026-04-25.
• Mimir-cluster project mapping — 14 fleet_tasks rows (mimir-agent, mimir, client-VISKA-mimir-agent) pending taxonomy session. Three options: collapse all to one Viska tile, split agent code vs n8n spoke, defer until Mimir repurposed.
• n8n credential gap closure long-term (metis-027) — Hades deposits 3 credentials into nono metis profile, Atlas owns all n8n VPS pushes, or keep simulator pattern indefinitely.
• Hub URL + auth model — single-user (Boas) + agents-only at v1; multi-user later.

14. References

Authoritative source documents (current PRD inherits from these)

• pantheon/docs/pantheon/DASHBOARD-BIBLE.md — vision, AD-01..AD-11, verbatim user directions, document registry, session log
• constellation-athena/docs/specs/pantheon-hub-navigation-restructure.md — nav structure, per-page specs, repo consolidation plan
• constellation-athena/docs/specs/pantheon-hub-v2-session-plan.md — implementation waves, cross-room coordination
• constellation-athena/docs/specs/pantheon-hub-prd-2026-04-28.md — PM MVP scope cut, task list

Adjacent / supporting

• constellation-athena/docs/specs/athena-subpage-architecture.md — Athena page deep-dive (post-MVP)
• constellation-athena/docs/specs/planning-view-reference.md — planning view detail (post-MVP)
• constellation-athena/docs/audits/pantheon-hub-project-mapping.md — comprehensive Hub mapping, Apr-08 baseline
• constellation-athena/backlog/pantheon-hub--strategy-docs-surface.md (athena-012) — strategy docs surface task

Out of scope

• constellation-athena/docs/specs/pantheon-harvester-v1-spec.md — separate roadmap
• constellation-athena/docs/specs/governance-intelligence-model.md — governance domain, not Hub PRD
• constellation-athena/docs/specs/supabase-governance-schema.md — governance schema, not Hub PRD
• constellation-athena/docs/specs/dependency-intelligence-system.md — Proteus's domain, separate spec
• constellation-athena/docs/specs/cross-repo-front-end-operations.md — separate scope

Operational SSOTs (non-PRD, but consulted for accuracy)

• ~/Dev/_K4120S/constellation-atlas/WORLD.md — fleet topology
• ~/Dev/_K4120S/constellation-atlas/OPS.md — fleet operational rules
• fleet_* tables on Pantheon Supabase — runtime SSOT for everything in §6
• ~/.claude/rules/*.md — global rules surfaced into agent context

15. Change Log

End of PRD. This document is the single source of truth for Pantheon Hub product requirements. Implementation work references this document for scope, sequencing, and acceptance.