Indicative Roadmap

---

M0 — Platform charter and boundaries

Outcome: everyone knows what each product owns.

Deliverables:

• platform charter,
• product boundary matrix,
• intersection map,
• ecosystem reference (ecosystem_reference.v1),
• MVP scope,
• non-goals,
• initial ADRs.

In flight: Local LLM micro-agent pilot — KS UX harness --llm + forge-workcells local_llm_worker (see Forge micro-agent and workcells hub).

M1 — Forge Run Spine

Outcome: one shared run object connects all products.

Deliverables:

• ForgeRun v1,
• EvidencePacket v1,
• ApprovalRequest v1,
• AgentRun v1,
• WorkcellRequest v1,
• WorkcellResult v1,
• static samples.

M2 — Lenses as local control plane

Outcome: humans can see and manage the run spine.

Deliverables:

• ForgeRun list/detail,
• evidence panel,
• approval card,
• agent run panel,
• Fleet/LCDL placeholder panels.

M2-lite — Lenses static/control (Self-Host prerequisite)

Outcome: smallest Lenses slice to open/import a ForgeRun from disk, show approval + evidence + local-runner result + follow-on Sparks, and transition decision state safely (local JSON writes only until policy hardens).

Deliverables:

• Studio route backed by GET /api/project/<slug>/forge-runs
• Read-oriented panels for EvidencePacket + ApprovalRequest
• Optional POST decision patch (loopback-gated consistent with other Lenses writes)

SELFHOST-ALPHA — Local platform improvement loop

Outcome: approval-gated allowlisted local runner + importer + .forge/runs/ store documented.

Deliverables:

• LocalRunnerAction / LocalRunnerResult v1
• scripts/selfhost_runner.py, scripts/selfhost_import_evidence.py
• docs/selfhost-run-store.md
• Validated static chain + demo fixture under sprints/selfhost-alpha/

M3 — Blueprints + Versonas + LCDL reasoning

Outcome: raw intent becomes a governed, verified plan.

Deliverables:

• BlueprintContextPack,
• VersonaSessionReport,
• LCDLTraceSummary,
• decompose/DecisionPack/contradiction scan flows.

M4 — Fleet controlled execution

Outcome: approved jobs run through Fleet and return evidence.

Deliverables:

• FleetTemplate v1,
• docs-health template,
• FleetJobSummary v1,
• approval-gated submit,
• Lenses job card,
• LCDL output classification.

M5 — Agent workcells

Outcome: Hermes and Factory can operate inside the Forge envelope.

Deliverables:

• Hermes skill pack,
• Hermes MCP configs,
• Factory WorkcellRequest/Result lane,
• Factory Fleet templates,
• Lenses agent output panel.

M6 — Channel ingress

Outcome: OpenClaw can start and monitor runs without becoming the authority.

Deliverables:

• ChatIntent v1,
• OpenClaw-to-Lenses bridge,
• status relay,
• evidence summary relay.

M7 — Dogfood and hardening

Outcome: the platform is used to improve the platform.

Deliverables:

• cross-product smoke test,
• docs-health evidence packets,
• release readiness check,
• public MVP narrative,
• known limitations.