mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-17 23:09:35 +00:00
e11253ab12
Track 1 — Truth Lock: docs/current-vs-target-register.md: Full subsystem maturity register (73 Production, 27 Partial, 2 Pilot, 32 Target, 6 Watch = 52.1% maturity) Track 2 — Document Consistency: docs/governance/document-consistency-audit.md: All 6 checks PASS (no dangling refs, no overclaim, all paths root-safe, naming consistent) Track 3 — Decision Plane: backend/app/schemas/structured_outputs.py: 17 Pydantic schemas with Provenance (LeadScoreCard, QualificationMemo, ProposalPack, PricingDecisionRecord, PartnerDossier, EconomicsModel, ApprovalPacket, TargetProfile, DDPlan, ValuationMemo, SynergyModel, ICMemo, BoardPackDraft, ExpansionPlan, StopLossPolicy, PMIProgramPlan, ExecWeeklyPack) Track 4 — Execution Plane: docs/governance/workflow-inventory.md: 8 short + 8 medium + 6 long-lived workflows classified. 3 Temporal candidates with compensation logic. Track 5 — Trust Fabric: docs/governance/trust-closure-plan.md: 5 live components + Watch adoption criteria for OPA/OpenFGA/Vault/Keycloak Track 6 — Data & Connectors: docs/governance/connector-standard.md: Connector facade contract, semantic metrics dictionary, radar additions (Airbyte, Unstructured, Great Expectations) Track 7 — Operating Plane: docs/governance/operating-plane-checklist.md: GitHub governance, CI/CD enhancements, CODEOWNERS template, OIDC/attestation roadmap Track 8 — Saudi/GCC: docs/governance/saudi-enterprise-readiness.md: PDPL processing register, data classification, NCA ECC readiness, OWASP LLM Top 10, NIST AI RMF Track 9 — Executive Surfaces: docs/governance/executive-surface-closure.md: Wiring plan with real data queries for Executive Room, Approval Center, Compliance Dashboard Track 10 — Market Dominance: docs/governance/market-dominance-plan.md: 3-tier packaging (Core/Strategic/ Sovereign), ROI narrative, competitive wedge, capability moat map, executive sales stories (CEO/CTO/CFO/CISO) Master Checklist: docs/tier1-master-closure-checklist.md 50 items total — 25 Done (documentation), 25 Target (runtime/integration) https://claude.ai/code/session_01W1rJthWDkasijTdXCfxVHs
129 lines
4.3 KiB
Markdown
129 lines
4.3 KiB
Markdown
# Trust Fabric Closure Plan — Track 5
|
|
|
|
> **Parent**: [`trust-fabric.md`](trust-fabric.md)
|
|
> **Plane**: Trust | **Version**: 1.0
|
|
|
|
---
|
|
|
|
## Objective
|
|
|
|
Transform Trust Plane from "policy engine + audit logs" to "no sensitive action without approval + verification + evidence + correlation."
|
|
|
|
---
|
|
|
|
## Live Trust Components Required
|
|
|
|
### 1. Approval Packet Flow (Priority 1)
|
|
**Goal**: At least one path where Class B action goes through structured ApprovalPacket → review → approve/reject → execute → evidence.
|
|
|
|
**Target Path**: WhatsApp outreach to new lead
|
|
|
|
```
|
|
Agent proposes send_whatsapp
|
|
→ ApprovalPacket schema generated (structured_outputs.py)
|
|
→ Policy gate classifies as B
|
|
→ ApprovalRequest created with SLA deadline
|
|
→ Reviewer gets notification
|
|
→ Approve → approval_token issued
|
|
→ OpenClaw gateway executes with token
|
|
→ Tool receipt generated
|
|
→ Evidence logged to ai_conversations + audit_log
|
|
```
|
|
|
|
**Required Wiring**:
|
|
- `ApprovalPacket` schema → `approval_bridge.py` integration
|
|
- SLA deadline field on `ApprovalRequest` model
|
|
- Notification to reviewer (email/WhatsApp)
|
|
- Evidence: approval_token + tool_receipt + audit_log linked by `trace_id`
|
|
|
|
### 2. Tool Verification Receipt Flow (Priority 1)
|
|
**Goal**: At least one tool call produces a verifiable receipt.
|
|
|
|
**Implementation**:
|
|
- `tool_verification.py` already exists
|
|
- `tool_receipts.py` already exists
|
|
- Need: receipts written for WhatsApp plugin calls
|
|
- Need: receipt includes `trace_id`, `tenant_id`, `action`, `result_hash`, `timestamp`
|
|
|
|
### 3. Contradiction Detection (Priority 2)
|
|
**Goal**: Real contradictions detected and flagged.
|
|
|
|
**Implementation Plan**:
|
|
- Wire `contradiction_engine.py` to CI pipeline
|
|
- On governance doc change: run LLM scan against other governance docs
|
|
- Store detected contradictions in `contradictions` table
|
|
- Show in Policy Violations Board frontend
|
|
|
|
### 4. Evidence Pack Viewer (Priority 2)
|
|
**Goal**: Unified evidence pack that links decision → tool → approval → output.
|
|
|
|
**Implementation**:
|
|
- `evidence_pack_service.py` exists
|
|
- Need: `assemble_deal_pack` that queries real data:
|
|
- Deal from `deals` table
|
|
- Lead from `leads` table
|
|
- Activities from `activities` table
|
|
- Messages from `messages` table
|
|
- Approvals from `approval_requests` table
|
|
- AI conversations from `ai_conversations` table
|
|
- Consent from `consents` table
|
|
|
|
### 5. Trace Correlation (Priority 1)
|
|
**Goal**: `trace_id` / `correlation_id` links all related records.
|
|
|
|
**Implementation**:
|
|
- Add `correlation_id` to `DomainEvent` (already exists as field)
|
|
- Pass `correlation_id` through OpenClaw gateway → task router → agent → handler
|
|
- Store in `ai_conversations.correlation_id`, `audit_log.correlation_id`
|
|
- Query by `correlation_id` in evidence pack assembly
|
|
|
|
---
|
|
|
|
## Watch Technologies — Adoption Criteria
|
|
|
|
### OPA (Open Policy Agent)
|
|
**Adopt when**:
|
|
- Policy rules exceed 50 AND are complex (nested conditions, temporal logic)
|
|
- Current `policy.py` becomes maintenance burden
|
|
- ADR demonstrates value with prototype
|
|
|
|
**Spike criteria**:
|
|
- [ ] Prototype: 5 existing policy rules expressed in Rego
|
|
- [ ] Benchmark: latency comparison vs current Python implementation
|
|
- [ ] Integration: OPA sidecar evaluated for performance
|
|
|
|
### OpenFGA
|
|
**Adopt when**:
|
|
- Authorization logic exceeds role-based (needs relationship-based)
|
|
- Multi-tenant permission inheritance becomes complex
|
|
- ADR demonstrates value with prototype
|
|
|
|
**Spike criteria**:
|
|
- [ ] Prototype: tenant → user → resource permission graph
|
|
- [ ] Benchmark: query latency for "can user X do action Y on resource Z"
|
|
- [ ] Integration: OpenFGA as authorization service evaluated
|
|
|
|
### Vault
|
|
**Adopt when**:
|
|
- Secret rotation is needed for compliance
|
|
- 10+ distinct secret types managed
|
|
- Environment variables become unwieldy
|
|
|
|
### Keycloak
|
|
**Adopt when**:
|
|
- SSO requirement from enterprise customer
|
|
- Multi-IdP federation needed
|
|
- Current JWT auth insufficient
|
|
|
|
---
|
|
|
|
## Gate: Trust Closure
|
|
|
|
- [ ] One approval flow live end-to-end with SLA
|
|
- [ ] One tool verification receipt generated and stored
|
|
- [ ] One contradiction detected in real scan
|
|
- [ ] One evidence pack assembled from real deal data
|
|
- [ ] `trace_id` links decision → approval → execution → evidence
|
|
- [ ] Contradiction dashboard shows real data
|
|
- [ ] Approval SLA measured for at least one path
|