Operator Surface Capability Matrix

This document is the backend-first decision aid for operator surface scope.

It exists to separate three things that currently get mixed together:

• backend implementation truth
• runtime validation / evidence truth
• frontend scope policy (V1, admin-only, observe-only, internal-only)

It is a current-runtime exposure matrix, not the sole authority for agent promotion maturity. For any task-promotion or “is this agent ready to own this lane?” decision, read this together with docs/architecture/AGENT_CAPABILITY_MODEL.md and docs/architecture/AGENT_CAPABILITY_IMPLEMENTATION_MATRIX.md.

Core Interpretation

V1 in the current repo mostly means:

• the curated maintained specialist scope for operator-s-console
• not the full limit of what orchestrator can route or expose
• not a plan to keep growing a competing generic shell

The orchestrator backend is broader than the current curated frontend scope.

That means:

• a route or task can already exist in backend
• the route or task can even have real success evidence
• but still be intentionally hidden from a normal operator frontend until the operator-facing classification is explicit and stable

That still does not mean every implemented task should be promoted immediately. The intended promotion order is:

1. agent capability reaches the required maturity target
2. runtime evidence supports that claim
3. operator-facing task promotion follows

This document must stay synchronized with the capability docs. It describes current exposure/runtime truth, not the finish line for agent maturity.

Current local-runtime note:

• The separate public-proof/openclawdbot surface is retired in the active local runtime.
• Public proof is now served from orchestrator-owned public routes.
• /operator and /operator/* now serve the built operator-s-console bundle from orchestrator itself.
• operator-s-console/ remains the only supported /operator delivery path, but OpenClaw Control UI/chat is now the preferred general front door for daily use.
• /operator should therefore be treated as a specialist sidecar surface, not as a competing generic operator shell roadmap.

Documentation Maintenance

Whenever operator exposure changes, or when runtime truth materially changes what an operator can credibly rely on, update this file together with:

• AGENT_CAPABILITY_MODEL.md
• AGENT_CAPABILITY_IMPLEMENTATION_MATRIX.md
• ../reference/task-types.md
• ../reference/api.md

If the backend becomes broader than the current frontend on purpose, say that explicitly here rather than leaving the difference implicit.

Future portfolio-growth note:

• when the public repo grows beyond the current catalog, new agents should come in one by one from deliberate external-catalog selection, not by bulk-import
• no candidate should appear in operator surfaces until it owns a bounded lane and has operator-visible runtime evidence that justifies honest exposure

Decision Labels

• Expose now Safe and credible for normal operator use today.
• Admin-only Backend exists, but the runtime path is still low-confidence, side-effectful, or easy to misuse. Keep behind explicit admin intent.
• Observe-only Surface status, telemetry, readiness, and run history. Do not present as a normal user-triggerable workflow yet.
• Internal-only Never expose as a normal operator action. Runtime-only plumbing.

Source of Truth

Primary sources for this matrix:

• workspace/orchestrator/src/index.ts
• workspace/orchestrator/src/middleware/auth.ts
• workspace/docs/reference/task-types.md
• workspace/docs/reference/api.md
• workspace/docs/architecture/AGENT_CAPABILITY_MODEL.md
• workspace/OPENCLAW_CONTEXT_ANCHOR.md

1. Task Capability Matrix

The backend allowlist is broader than the current curated UI task surfaces. Normal operator UIs should follow the matrix below rather than the raw allowlist.

Task Type	Backend Present	Runtime Truth	Safe for /operator	Safe for operator-s-console	Decision	Notes
heartbeat	Yes	internal runtime maintenance path	No	No	Internal-only	Scheduled control-plane upkeep. Keep it out of normal operator task launch surfaces and inspect it only through diagnostics.
build-refactor	Yes	approval-gated; explicit bounded patch mode now executes real edits + optional verification	Yes	Yes	Expose now	Sensitive and real. Keep approval language prominent and require explicit scoped payloads for code surgery.
market-research	Yes	confirmed-working fetch lane with explicit URLs or derived allowlisted source plans; network posture still matters	Yes	Yes	Expose now	Surface source-plan guidance and network caveats, but treat this as a real research lane rather than a placeholder query capture.
doc-sync	Yes	confirmed control-plane path; most useful when pending changes exist	Yes	Yes	Expose now	Safe low-side-effect queue action.
business-value-cycle	Yes	confirmed governed planner path using schema-v2 registry, pipeline, project, risk, and coverage-gap evidence	Yes	Yes	Expose now	Bounded trigger and scheduled lane. It may create internal planning/verification work, but it does not authorize outreach, publishing, deployment, secret/private-source access, or commercial commitments.
nightly-batch	Yes	historical success exists; scheduler and downstream feeds still matter	Yes	Yes	Expose now	Keep schedule/duplication caveat visible in both UIs.
drift-repair	Yes	local smoke proved with real knowledge-pack and repair evidence	Yes	Yes	Expose now	Operator-facing repair lane is live; keep doc-health caveats visible rather than hiding the repair posture.
control-plane-brief	Yes	focused contract proof for bounded control-plane synthesis	Yes	Yes	Expose now	Companion-facing synthesis lane. Keep it bounded and machine-readable rather than turning it into a free-form dashboard scrape.
incident-triage	Yes	focused contract proof for ranked incident queue synthesis	Yes	Yes	Expose now	Expose now as the bounded incident-priority lane rather than forcing operators to cluster incident pressure by hand.
release-readiness	Yes	focused contract proof for bounded release posture synthesis	Yes	Yes	Expose now	Expose now as a release-governance lane. Keep the summary explicit about go, hold, or block and do not overstate it as deploy authority.
deployment-ops	Yes	focused contract proof for bounded deployment posture synthesis	Yes	Yes	Expose now	Expose now as a deployment posture lane. Keep it read-only and explicit about rollback readiness, drift, and pipeline blockers rather than implying deploy authority.
code-index	Yes	live canary proof for bounded code-index posture synthesis	Yes	Yes	Expose now	Expose now as a read-only indexing and retrieval-readiness lane. Keep it explicit about bounded local coverage, freshness, and linkage limits rather than implying unrestricted Codex-style repo authority.
test-intelligence	Yes	live canary proof for bounded multi-suite test posture synthesis	Yes	Yes	Expose now	Expose now as a read-only test-intelligence lane. Keep it explicit about bounded local suite coverage, failure and retry evidence, and release-facing risk limits rather than implying CI execution authority.
compliance-review	Yes	live canary proof for bounded compliance posture synthesis	Yes	Yes	Expose now	Expose now as a read-only compliance posture lane. Keep it explicit about policy/doc coverage, dependency evidence, and release-governance limits rather than implying audit or enforcement authority.
reddit-response	Yes	confirmed-working drafting lane with deterministic local fallback and optional provider polish	Yes	Yes	Expose now	Allow manual queue payloads, and keep the optional provider-polish caveat explicit without understating the core drafting lane.
send-digest	Yes	partial; externally dependent	Yes	Yes	Expose now	Operator-facing now, but outbound channel dependency remains real.
rss-sweep	Yes	externally dependent	Yes	Yes	Expose now	Exposed with config/network caveats, not as a guaranteed-success lane.
agent-deploy	Yes	approval-gated; confirmed local deployment copy path	Yes	Yes	Expose now	Sensitive lane now has a real operator profile; keep approval language prominent because it writes deployment surfaces.
security-audit	Yes	confirmed working in live sweep	Yes	Yes	Expose now	Worker performs real repository/runtime inspection locally rather than a fake success wrapper.
system-monitor	Yes	confirmed working in live sweep	Yes	Yes	Expose now	Targeted monitor passes are now user-runnable and grounded in real runtime/service-state analysis.
summarize-content	Yes	confirmed working in live sweep	Yes	Yes	Expose now	Inline-content path is now a first-class operator task.
content-generate	Yes	confirmed working in live sweep with source-driven generation	Yes	Yes	Expose now	Safe enough as a grounded content lane when source fields are explicit.
integration-workflow	Yes	confirmed working in live sweep with deterministic reroute/replay orchestration	Yes	Yes	Expose now	Exposed with bounded default-plan plus shorthand-step guidance rather than implying a magical empty workflow.
normalize-data	Yes	confirmed working in live sweep	Yes	Yes	Expose now	Exposed with structured input/schema expectations.
data-extraction	Yes	inline lane proven; parser-backed file lanes implemented	Yes	Yes	Expose now	Expose now with parser/format caveats rather than implying only inline extraction is real.
qa-verification	Yes	confirmed working in live smoke	Yes	Yes	Expose now	Exposed with dry-run/live-check distinctions preserved.
skill-audit	Yes	confirmed working in live smoke	Yes	Yes	Expose now	Exposed as a bounded governance task profile.
startup	Yes	internal runtime task	No	No	Internal-only	Boot path only. Must not become a user action.
doc-change	Yes	internal runtime task	No	No	Internal-only	Watcher/buffer plumbing only. Must not become a user action.

Task Policy Notes

• public-triggerable means the trigger route accepts the task. It does not prove the downstream worker/service/dependency path is healthy.
• All non-internal public-triggerable tasks now have explicit operator-facing profiles in OPERATOR_TASK_PROFILES.
• Promotion does not erase runtime caveats. The operator-facing surfaces should expose those caveats rather than flattening every task into a “ready” button.
• The next clean backend move is deeper runtime proof, not another round of catalog hiding.
• Full-capability work should surface governed access gaps honestly. If an agent still lacks a required skill or tool for a task lane, show that as a readiness/policy problem rather than solving it by flattening permissions across the whole fleet.

2. Route Capability Matrix

This matrix is about backend route exposure, not current frontend implementation.

Route / Family	Role	Safe for /operator	Safe for operator-s-console	Decision	Notes
GET /operator	private operator console	Yes	N/A	Expose now	Canonical private operator-console entrypoint served from the built operator-s-console bundle by orchestrator.
GET /operator/*	private operator console route family	Yes	N/A	Expose now	SPA route family for the canonical private operator-s-console bundle, served by orchestrator.
GET /health	public	Yes	Yes	Expose now	Public shallow liveness only. Do not treat returned localhost helper URLs as browser targets.
GET /api/persistence/health	public	Yes	Yes	Expose now	Public dependency truth only, not full operator state.
GET /api/knowledge/summary	public	Yes	Yes	Expose now	Safe public summary; protected query is a separate surface.
GET /api/command-center/overview	public proof	Yes	Yes	Expose now	Public orchestrator-owned proof overview contract for the show-off surface.
GET /api/command-center/control	public proof	Yes	Yes	Expose now	Public orchestrator-owned control-lane summary for the proof page.
GET /api/command-center/demand	public proof	Yes	Yes	Expose now	Public orchestrator-owned demand summary contract.
GET /api/command-center/demand-live	public proof	Yes	Yes	Expose now	Public orchestrator-owned live demand snapshot contract.
GET /api/milestones/latest	public proof	Yes	Yes	Expose now	Public orchestrator-owned latest milestone feed.
GET /api/milestones/dead-letter	public proof	Yes	Yes	Expose now	Public orchestrator-owned proof-risk feed for blocked or at-risk items.
GET /api/openapi.json	public	Optional	Optional	Observe-only	Useful for engineers and diagnostics, not a primary operator panel.
GET /api/auth/me	viewer	Yes	Yes	Expose now	Auth identity and role context.
GET /api/dashboard/overview	viewer	Yes	Yes	Expose now	Aggregate operator summary. Useful, but not stronger than its source routes.
GET /api/companion/overview	viewer	Yes	Yes	Expose now	Canonical read-first companion status surface for plugins and channel clients.
GET /api/companion/catalog	viewer	Yes	Yes	Expose now	Canonical read-first companion task catalog surface.
GET /api/companion/incidents	viewer	Yes	Yes	Expose now	Canonical read-first companion incident summary surface.
GET /api/companion/runs	viewer	Yes	Yes	Expose now	Canonical read-first companion recent-run brief surface.
GET /api/companion/approvals	operator	Yes	Yes	Expose now	Canonical read-first companion approval summary surface. Keep operator role requirement explicit.
GET /api/health/extended	viewer	Yes	Yes	Expose now	Authoritative protected operator-health surface.
GET /api/tasks/catalog	viewer	Yes	Yes	Expose now	Canonical operator capability endpoint for task surfacing.
POST /api/tasks/trigger	operator	Yes	Yes	Expose now	Curated tasks only. Frontends should not assume the full allowlist is user-facing.
GET /api/tasks/runs	viewer	Yes	Yes	Expose now	First-class run visibility surface.
GET /api/tasks/runs/:runId	viewer	Yes	Yes	Expose now	Full run detail / workflow graph / repair linkage surface.
GET /api/approvals/pending	operator	Yes	Yes	Expose now	Approval inbox for sensitive or review-gated flows.
POST /api/approvals/:id/decision	operator	Yes	Yes	Expose now	Action route; keep operator-only.
GET /api/incidents	viewer	Yes	Yes	Expose now	Core incident queue / ledger surface.
GET /api/incidents/:id	viewer	Yes	Yes	Expose now	Core incident detail surface.
GET /api/incidents/:id/history	viewer	Yes	Yes	Expose now	Important lifecycle stream; not admin-only.
POST /api/incidents/:id/acknowledge	operator	Yes	Yes	Expose now	Legitimate operator action, not just diagnostics.
POST /api/incidents/:id/owner	operator	Yes	Yes	Expose now	Legitimate operator action, not just diagnostics.
POST /api/incidents/:id/remediate	operator	Yes	Yes	Expose now	Legitimate operator action; better launched from incident context than a generic task list.
GET /api/agents/overview	viewer	Yes	Yes	Expose now	Canonical agent truth surface: declaration, worker path, explicit serviceExpected/lifecycleMode/host service status, readiness, topology.
GET /api/memory/recall	viewer	Yes	Yes	Expose now	Safe with default redaction. Good for activity/knowledge/operator context.
POST /api/knowledge/query	operator	Optional	Yes	Expose now	Safe operator workflow, but more naturally a deeper console page than the default overview/task rails.
GET /api/persistence/summary	viewer	Yes	Yes	Expose now	Strong operator-facing persistence summary.
GET /api/skills/policy	viewer	Yes	Yes	Expose now	Good for both shells. Summary-level governance posture.
GET /api/skills/registry	viewer	Optional	Yes	Expose now	Better fit for deeper governance pages than the default overview/task rails.
GET /api/skills/telemetry	viewer	Optional	Yes	Expose now	Better fit for deeper governance and diagnostics pages than the default overview/task rails.
GET /api/skills/audit	viewer	Optional	Yes	Expose now	Useful now, but more diagnostic/governance detail than the default overview/task rails.
GET /api/persistence/historical	viewer	No by default	No by default	Observe-only	Real route, but not essential to either current operator surface until a historical ops page exists.
GET /api/knowledge/export	admin	No	No	Admin-only	Export-heavy privileged route. Keep out of normal frontends.
GET /api/persistence/export	admin	No	No	Admin-only	Export-heavy privileged route. Keep out of normal frontends.
POST /webhook/alerts	signed ingest	No	No	Internal-only	Ingest boundary only. Never a UI route.

Route Policy Notes

• operator-s-console should not be forced to mirror every valid backend route.
• GET /api/persistence/historical is valid backend truth, but it is still reasonable to keep it out of both current consoles until a historical ops use case is designed.
• Export routes are real and supported, but they are privileged ops/admin surfaces, not day-to-day operator UI material.

3. Agent Capability Matrix

Agents should be judged on two separate axes:

• can the backend route work to them?
• has runtime evidence matured enough to trust their role-specific claims?

Do not confuse declared capability targets with current operator-safe maturity.

All declared agents should eventually become strong enough that this section is explaining trustworthy operator exposure for a fully capable portfolio, not a long-term excuse for permanent partial maturity.

Agent	Task Lane(s)	Backend Status	Current Maturity Signal	Safe for /operator	Safe for operator-s-console	Decision	Notes
doc-specialist	drift-repair, doc-sync aligned work	Present	Wave 1 gate closed for the current runtime slice; richer truth-spine uplift remains roadmap	Summary only	Full readiness / evidence	Expose now	High-value to surface. Keep action entrypoints contextual through tasks/incidents, not by “run agent” UI.
system-monitor-agent	system-monitor, incident-triage	Present	Wave 1 gate closed for the current runtime slice; deeper operator-action fusion remains roadmap	Summary only	Full readiness / evidence	Expose now	Strong candidate for deeper operator visibility before broader task exposure.
security-agent	security-audit	Present	Wave 1 gate closed for the current runtime slice; broader remediation-closure uplift remains roadmap	Summary only	Full readiness / evidence	Expose now	Surface readiness and evidence, but do not oversell as a finished end-state trust auditor.
qa-verification-agent	qa-verification	Present	Wave 1 gate closed for the current runtime slice; broader acceptance coverage remains roadmap	Summary only	Full readiness / evidence	Expose now	Best surfaced through verification-focused task flows and run detail, not as a generic direct-agent runner.
integration-agent	integration-workflow	Present	Wave 1 gate closed for the current runtime slice; broader workflow productization remains roadmap	Summary only	Full readiness / evidence	Expose now	Strong observability target; task exposure should follow clearer workflow productization.
build-refactor-agent	build-refactor	Present	Wave 3 gate closed for the current runtime slice; deeper applied-edit proof remains roadmap	Summary only	Full readiness / evidence	Expose now	Backing lane for one of the main operator tasks; current readiness now carries code-governance signals rather than a flat confirmed-worker label only.
market-research-agent	market-research	Present	confirmed worker path, external dependency caveats, and readiness now carries delta-capture evidence	Summary only	Full readiness / evidence	Expose now	Backing lane for one of the main operator tasks.
operations-analyst-agent	control-plane-brief	Present	focused bounded synthesis lane with companion-facing contract proof	Summary only	Full readiness / evidence	Expose now	Best surfaced as the reusable control-plane brief source for operator, bridge, and channel consumers.
reddit-helper	reddit-response	Present	service/helper exists, community path is still degraded/external, and readiness now carries provider-posture evidence	Summary only	Full readiness / evidence	Expose now	Keep queue/provider caveats explicit and treat reply drafting as a dependency-sensitive workflow, not a flat always-green lane.
content-agent	content-generate	Present	backend lane proven with bounded operator profile, and readiness now carries publication-policy evidence	Summary only	Full readiness / evidence	Expose now	Expose the bounded generation lane now, while keeping publishing and evidence caveats explicit.
data-extraction-agent	data-extraction	Present	partial by lane; inline source proven; readiness now carries artifact-coverage evidence	Summary only	Full readiness / evidence	Expose now	Good visibility target; do not imply all artifact lanes are equally proven.
normalization-agent	normalize-data	Present	backend lane proven, and readiness now carries comparison-readiness evidence	Summary only	Full readiness / evidence	Expose now	Surface readiness first.
release-manager-agent	release-readiness	Present	focused bounded synthesis lane with release-governance contract proof	Summary only	Full readiness / evidence	Expose now	Best surfaced as a release posture lane, not as a generic deploy button or unbounded release oracle.
deployment-ops-agent	deployment-ops	Present	focused bounded deployment-posture lane with live contract proof	Summary only	Full readiness / evidence	Expose now	Best surfaced as a deployment posture lane that stays read-only and approval-respecting rather than a deploy executor.
code-index-agent	code-index	Present	focused bounded code-index lane with live canary proof and promoted runtime evidence	Summary only	Full readiness / evidence	Expose now	Best surfaced as a local repo/index truth lane, not as an unrestricted generic repo operator.
test-intelligence-agent	test-intelligence	Present	focused bounded test-intelligence lane with live canary proof and promoted runtime evidence	Summary only	Full readiness / evidence	Expose now	Best surfaced as a bounded test-evidence posture lane, not as an executor for tests, CI, or shell workflows.
compliance-agent	compliance-review	Present	focused bounded compliance lane with live canary proof and promoted runtime evidence	Summary only	Full readiness / evidence	Expose now	Best surfaced as a policy/dependency posture lane that stays read-only and does not claim audit enforcement authority.
summarization-agent	summarize-content	Present	backend lane proven, and readiness now carries operational-compression evidence	Summary only	Full readiness / evidence	Expose now	Surface readiness first.
skill-audit-agent	skill-audit	Present	Wave 3 gate closed for the current runtime slice; broader operator adoption remains roadmap	Summary only	Full readiness / evidence	Expose now	Best surfaced with governance/tool audit views.

Agent Policy Notes

• All declared agents are valid to show in operator surfaces as long as the UI labels them honestly.
• The risky mistake is not surfacing them. The risky mistake is surfacing them as equally mature.
• /operator should stay at summary-level agent truth.
• operator-s-console is the better place for full readiness, gap, topology, and relationship history views.
• No current UI should imply “run this agent directly” unless the action is actually modeled as a safe task workflow.
• Full capability should appear here as stronger readiness, richer allowedSkills[], and better evidence-backed task ownership, not as silent universal tool access for every agent.

4. Immediate Surfacing Policy

Use this policy until a stricter product split is chosen.

Private operator console (/operator and /operator/*)

Keep as the canonical private operator console:

• auth identity
• curated task trigger
• approvals
• quick health and incident action
• summary agent truth
• summary governance / skill policy
• run detail only where it directly supports an operator action
• deeper task run history and workflow replay
• richer incident lifecycle and remediation detail
• agent capability readiness and relationship windows
• richer governance and skill surfaces
• richer knowledge atlas and diagnostics
• public proof page as a separate route family inside the same bundle

Current route contract:

• /operator and /operator/* are served from the built operator-s-console bundle by orchestrator
• protected routes remain orchestrator-backed under that console
• route depth within the same console is the exposure distinction now, not a separate “minimal shell versus richer console” product split

Public proof boundary

Keep public proof separate from the private operator control plane:

• /operator/public-proof is the public page path in the same operator-s-console bundle
• it reads only orchestrator-owned public proof routes (/api/command-center/* and /api/milestones/*)
• it must not collapse protected operator state into public proof payloads
• it replaces the retired openclawdbot transport path without reviving a separate proof product boundary

Not for either normal operator surface right now

• raw export routes
• webhook ingest routes
• internal runtime tasks
• unprofiled task lanes promoted as ordinary buttons without operator-facing classification

5. Practical Next Moves

If we want to expose more backend capability without creating chaos, the correct order is:

1. Do not add new generic shell or dashboard scope just because backend capability exists.
2. Prefer OpenClaw Control UI/chat plus the bridge for shared operator needs.
3. Only widen /operator when the added surface exposes unique orchestrator-specific value that OpenClaw does not already cover.
4. Keep everything else visible through runs, incidents, agents, governance, and bridgeable specialist workflows until the operator value is explicit.

That keeps the backend broad, the frontend honest, and the operator experience trustworthy.