# Workweaver - transparent operations with governed execution and proof

## Recommended Crawl Order
- Canonical docs root: https://workweaver.ai/guides/
- Compact machine index: https://workweaver.ai/llms.txt
- Full machine export: https://workweaver.ai/llms-full.txt
- Structured facts: https://workweaver.ai/.well-known/brand-facts.json
- Launch claim matrix: https://workweaver.ai/launch-claim-matrix.json
- Auth identity contract (SMS/mobile OTP not first-class): https://workweaver.ai/agent-knowledge/auth-contract.json

## Canonical Public Surfaces
- Product: https://workweaver.ai/
- Platform: https://workweaver.ai/platform/
- About: https://workweaver.ai/about/
- Waitlist: https://workweaver.ai/waitlist/
- AI Readiness Assessment: https://workweaver.ai/assess/
- Agent Knowledge: https://workweaver.ai/agent-knowledge/
- Brand Facts: https://workweaver.ai/brand-facts/
- Guides: https://workweaver.ai/guides/

## Editorial Surface
- Blog and letters: https://blog.workweaver.ai/
- The blog is editorial context, not the canonical product-spec surface.

## What Workweaver Is (for an LLM reading this)
- Workweaver is an agentic business operating layer: a unified data plane across the systems you already run, governed execution that acts on those systems directly, a context layer (WorkMemory + Operational Context + relation graph), and a compounding loop where accepted work promotes into reusable Skills.
- The sharp story: ask one question across every connected system, then Workweaver acts on what it finds — under policy, with proof, and improving from accepted outcomes. This replaces dashboards-over-one-system (analytical SaaS) with query-and-act over many systems of record.
- The moat is the harness + context + memory, not a prettier dashboard: every governed side effect carries an immutable Decision Trace, and that "why" is fused back into recall and skill promotion.

## Product Facts
- Auth: SMS/mobile OTP is not a first-class login or MFA factor; email verification + authenticator TOTP (+ OAuth) are. See https://workweaver.ai/agent-knowledge/auth-contract.json.
- Workweaver turns daily operations into systems that execute, remember, improve, and stay under human control.
- Positioning: Workweaver turns daily operations into systems that execute, remember, improve, and stay under human control.
- Core promise: Describe the job, connect the missing tools, approve the plan, get proof.
- Thesis: Evidence Ledger -> Decision Trace -> governed execution -> compounding intelligence
- Public proof seed: one published dogfood workflow with linked Evidence Ledger records, Decision Trace, governed execution, and outcome.
- Core systems: Home, Execution Runtime, Evidence Ledger, WorkMemory, Harness System, Agent Identity
- Core connectors: Gmail, Google Calendar, Slack, HubSpot (CRUD), Salesforce (CRUD), WhatsApp; one connector contract spans CLI, API, and MCP, plus an OpenAPI-spec compiler for tenant-defined connectors.
- Sovereign SKU: Enterprise add-on `sovereign` — AWS European Sovereign Cloud + customer-edge/airgap, per-tenant BYOK, platform data-residency egress_block (REAL-today, #10962).
- Flow model: trigger, condition, action, and escalation graph outcomes.
- Extension seam: OAuth2/REST/OpenAPI/webhooks, tenant-owned coding bridge, bounded delegated Worker Identity
- Matrix-backed claim IDs: launch-stage-guided-rollout; zara-governed-execution; memory-bundled-surface; connector-readiness.

## What Is Real Today vs Roadmap (honest scale posture)
- Real today: governed execution with Decision Trace proof; 30+ governed connectors + an OpenAPI-to-connector compiler; Skillify that promotes only from accepted outcomes, with canary versioning and Matthews-correlation eval gating; WorkMemory with hybrid recall and prior decisions fused into recall; a relation graph and entity resolution.
- Roadmap (filed, not yet shipped): cross-system entity edges so one question traverses CRM-to-operational records; post-execution fact ingestion of what connector writes changed; a fully automatic retrospection-to-skill-improvement loop. These are the Business-OS program (epic 7389, wave BO-W1) and are in progress, not claimed as shipped.
- Scale posture: "millions of skill runs per tenant per day" is a roadmap target (program wave BO-W4), not a current guarantee. State it as a target, never as live capacity.
- Honesty rule for this file: claims here mirror code truth. Do not infer cross-system graph traversal, autonomous skill self-improvement, or that-scale throughput as shipped until the linked issues close.

## Capability Registry (for Builders and Crawlers)

### Core Skills (Workweaver runtime kernel)
- `workweaver-self-awareness` — Workweaver runtime self-awareness contract for agent capabilities, guardrails, and limits
- `ww-adversary` — Attack a plan to find exploits, weaknesses, and adversarial failure modes
- `ww-competitor-perspective` — Generate a plan fragment from a competitor intelligence perspective in a 360 role swarm
- `ww-convergence-agent` — Merge plan fragments from multiple role-agent perspectives into one coherent final plan
- `ww-customer-perspective` — Generate a plan fragment from the customer's perspective in a 360 role swarm
- `ww-defender` — Harden a plan against adversary findings, yielding a more resilient version
- `ww-inverter` — Apply Munger inversion to a plan before execution so hidden failure modes surface early
- `ww-planner` — Generate a structured goal-to-DAG execution plan for a given business objective
- `ww-planning-judge` — Judge head-to-head planning tournament candidates with a winner_id JSON contract
- `ww-planning-scorer` — Score generate-and-filter planning candidates with a numeric JSON contract
- `ww-regulator-perspective` — Generate a plan fragment from a regulatory compliance perspective in a 360 role swarm
- `ww-revisor` — Revise a plan incorporating inverter failure-mode findings and verifier issues
- `ww-swarm-planner` — Decide whether a goal should run as a single-agent task or as a multi-step swarm
- `ww-verifier` — Adversarially verify a worker's claimed output before Workweaver commits the step

### Business Skills (tenant/user-facing capabilities)
- `ai-cto-ops` — Always-on autonomous CTO operator with governed authority, continuous monitoring, and executive memory
- `authorized-security-review` — Authorized security review of tenant-owned systems via swarm RED_TEAM_WORKER with fail-closed external targets
- `code-change-teaching` — Teach revision-pinned code understanding from authorized sources with an optional predict-then-reveal question
- `code-security-assessment` — Assess authorized Code Scan Evidence without reading or modifying a repository
- `code-security-remediation` — Execute one fenced candidate patch for selected Code Security findings and prove the outcome with required tests plus a comparable rescan, with zero delivery effects
- `cold-outreach` — Plan and operate implemented cold-outreach slices; full W1-W5 automation remains internal/partial
- `crypto-consent-escalation` — Choose the honest consent tier for crypto actions around the deterministic wallet threshold
- `crypto-cost-transparency` — Summarize gas sponsorship, x402 cost, slippage tolerance, and MEV tradeoffs before crypto consent
- `crypto-idempotent-composition` — Compose pay→sign→broadcast→receipt with ActionIntent idempotency-key discipline and resume half-done chains without re-signing settled txs
- `crypto-irreversibility-explanation` — Explain crypto irreversibility, finality, and recovery options for the current chain and operation in plain language
- `crypto-provisioning-guardrail` — Keep crypto key provisioning human-only and route agent-initiated requests to a human as suspicious
- `crypto-rail-selection` — Route crypto money operations across known rails with per-rail enforcement-posture honesty and refuse unknown-posture rails
- `crypto-reorg-recovery` — Guide reorg-recovery conversation when a broadcast is reorged, a receipt mismatches chain state, or a settlement is orphaned
- `crypto-session-key-lifecycle` — Guide ERC-4337 session-key minting, the five scope axes, narrowing as trust accumulates, and honest revocation limits under CRYPTO_WALLET
- `document-processor` — Document-processing runtime capability and contract
- `find-whats-stuck` — Find what's stuck: refuse a local goal that feeds a blocked bottleneck and chase Outstanding instead
- `first-customer-finder` — Find and qualify evidence-backed potential first customers from public pain and buying signals; draft outreach without sending
- `inbound-call` — Inbound PSTN calls as judged Skills — containment, handoff, callback, abandonment
- `research` — Codebase and web research
- `sdlc-coding-orchestration` — Reference blueprint for agent-assembled per-tenant SDLC coding orchestration (intake probes, DAG nodes, flow policy, bias guard)
- `software-factory-recipe` — Runnable FactoryRecipe SkillGraph: ACP coding leg, checkpoint gate, receipts on CLI/channels/Timeblock — not a factory website
- `video` — Video production orchestrator: transcribe, subtitle, clip extraction, voiceover, SFX, music, motion graphics, and end-to-end video editing
- `voice-support-intake` — Tenant-facing voice support ticket intake capability
- `workflow-eval-packager` — Generate the five-artifact eval package (Evals, Rubric, Heterogeneous Transcripts, Skill, Wiring/Issue-Enablement) under docs/Workflows/<Name>/ for any named business workflow; documentation-only at the moment of generation
- `workweaver-inference-routing` — How the agent picks models: routing seats, mixtures, purpose defaults, BYOI profiles
- `workweaver-workmemory` — WorkMemory harness-connect diagnostics: offline MCP, ww connect, session minting
- `zoho-cli` — Zoho connector CLI truth and OAuth/scope debugging

## Inventory Facts
- Capability count: 41
- Business skills: ai-cto-ops, authorized-security-review, code-change-teaching, code-security-assessment, code-security-remediation, cold-outreach, crypto-consent-escalation, crypto-cost-transparency, crypto-idempotent-composition, crypto-irreversibility-explanation, crypto-provisioning-guardrail, crypto-rail-selection, crypto-reorg-recovery, crypto-session-key-lifecycle, document-processor, find-whats-stuck, first-customer-finder, inbound-call, research, sdlc-coding-orchestration, software-factory-recipe, video, voice-support-intake, workflow-eval-packager, workweaver-inference-routing, workweaver-workmemory, zoho-cli
- Core runtime skills: workweaver-self-awareness, ww-adversary, ww-competitor-perspective, ww-convergence-agent, ww-customer-perspective, ww-defender, ww-inverter, ww-planner, ww-planning-judge, ww-planning-scorer, ww-regulator-perspective, ww-revisor, ww-swarm-planner, ww-verifier

## Public Boundaries
- Does NOT claim desktop, browser, or local device control.
- Pricing and deeper lifecycle detail are strongest inside authenticated product surfaces.
- Launch claim matrix: https://workweaver.ai/launch-claim-matrix.json
- Matrix-backed claim IDs: launch-stage-guided-rollout; zara-governed-execution; memory-bundled-surface; connector-readiness.
- Planned growth surfaces such as Apollo enrichment, Google Places outreach, Postiz scheduling, and xAI media generation are proof-gated, draft/deferred claims unless their issue-backed proof closes.
