Handbook
Product Intersection Map
Agents and workcells: workcells hub — terminology, runners, catalog, and governed workflows with conditional rules (approval gates, DecisionPack, campaign phases).
Updated
Visual flows: Platform reference architecture (per-element flowcharts) and Ecosystem reference (repo topology).
Platform-critical intersections
| Intersection | Purpose | Primary contract |
|---|---|---|
| Lenses <-> Blueprints | make repo policy visible and navigable | BlueprintContextPack |
| Lenses <-> LCDL | make governed reasoning visible | LCDLTraceSummary |
| Lenses <-> Fleet | make controlled execution visible | FleetJobSummary |
| Blueprints/Versonas <-> LCDL | turn discipline guidance into governed reasoning contracts | VersonaSessionReport |
| All products <-> Lenses | unify work through run/evidence state | ForgeRun, EvidencePacket |
| Local runner (alpha) <-> Lenses | approve-and-run bounded repo checks without Fleet | LocalRunnerAction, LocalRunnerResult, .forge/runs/ |
| Campaign orchestrator <-> Fleet | submit workspace, poll job, collect worker_result | FleetJobSummary, workspace upload meta |
| Campaign orchestrator <-> Lenses | ForgeRun + manifest + integrate/PR status | ForgeRun, EvidencePacket, AgentRun |
| Fleet UX worker <-> Fleet | in-container steps via workspace_worker bridge | worker_progress, worker_result |
| KS UX harness <-> forge-workcells | micro-agent run on fixture (--llm) |
WorkcellRequest, WorkcellResult, AgentRun |
| forge-workcells <-> LCDL | local LLM inference for micro-agent | LLM_* profile; LCDLTraceSummary when traced |
| Cockpit <-> forge-cdp-manager | parallel Edge tab leases, fleet, channel cancel | surface lease HTTP + flock; GET /v1/fleet |
| Knowledge Assistant <-> forge-cdp-manager | SharePoint lease on :9223, session metadata | SURFACE_SHAREPOINT; flock + HTTP /v1/leases |
| forge-cdp-manager <-> LCDL | tab match via lcdl_picker; managed lcdl/run gateway |
session_ref, run_task atoms |
| forge-cdp-manager <-> LCDL (AutomationFlow) | declarative probe flows without LCDL in consumer repo | POST /v1/flows/run; forge_lcdl.flow.FlowExecutor |
Operator-tool intersections
| Intersection | Purpose | Primary contract |
|---|---|---|
| Certificators <-> Fleet | bank-health pipeline jobs (tokenize, ingest, workers) | Fleet POST /v1/jobs, bearer FORGE_FLEET_* |
| Knowledge Assistant <-> LCDL | governed SharePoint sync atoms, copilot RAG recipes | run_task, LCDLTraceSummary when traced |
| Knowledge Assistant <-> ICS | Granite comprehension HTTP for case-study copilot | POST /v1/comprehend (ADR-0011) |
| STI <-> LCDL | governed test artifact generation and verification | lmeta flows, LCDLTraceSummary |
| STI <-> Lenses | test campaigns visible on run spine (future) | ForgeRun, EvidencePacket |
| forge-composer <-> consumer repos | deploy reviewed docs/manifests into siblings (e.g. Situ8) | forge_composer deploy --manifest |
| Cockpit memory <-> Fleet | pulse feed ingest via Fleet API | Fleet client + bearer token |
Quality-lane intersections
| Intersection | Purpose | Primary contract |
|---|---|---|
| KS UX harness <-> forge-a11y-checker | deterministic + axe website audits | KS analyze-website-ux.mjs, website-a11y-auditor |
| A11y Studio <-> KS | CDP + axe sealed remediation; UI shell patterns | KS tools/ui-app-audit/ |
fleet_ux_worker <-> KS |
in-container UX audit/remediation under ForgeRun | WorkcellRequest, worker_result |
| forge-llm <-> LCDL | optional Ollama-compatible gateway for LLM_* profiles |
HTTP chat/completions; gateway probes in LCDL docs |
| forge-llm <-> Lenses | LLM settings panel may target local gateway | Ollama-compatible endpoint env |
Agent-value intersections
| Intersection | Purpose | Primary contract |
|---|---|---|
| Hermes <-> Lenses | give Hermes perception and accountability | AgentRun, EvidencePacket |
| Hermes <-> Blueprints/Versonas | ground adaptive operator behavior in canonical policy | BlueprintContextPack, VersonaSessionReport |
| Hermes <-> LCDL | prevent important reasoning from staying unstructured | LCDLTraceSummary |
| Hermes <-> Fleet | give Hermes safe execution requests | FleetTemplate, FleetJobSummary |
Workcell expansion intersections
| Intersection | Purpose | Primary contract |
|---|---|---|
| Factory <-> Fleet | run implementation/review work in a controlled lane | WorkcellRequest, WorkcellResult |
| Factory <-> Blueprints/AGENTS.md | give Factory repo-local operating context | BlueprintContextPack |
| Factory <-> LCDL | verify implementation output independently | LCDLTraceSummary |
| OpenClaw <-> Lenses | convert chat/mobile ingress into governed run state | ChatIntent, ForgeRunStatusSummary |
Anti-patterns
- Hermes owns run state.
- Fleet stores process policy.
- Lenses runs LLM reasoning as its own hidden behavior.
- Blueprints become live runtime state.
- Factory approves its own output.
- OpenClaw treats chat text as sufficient approval for dangerous actions.
- Fleet opens pull requests or owns campaign manifest without Lenses ForgeRun.
- Confusing marketing “campaign” with Forge Campaign in operator docs.
- Product repos submodule forge-platform (use handbook URLs and sibling clones; only forge-platform-website submodules platform for build).