mancitrus/system-prompts-and-models-of-ai-tools
1
0
Fork 0
mirror of https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git synced 2026-06-18 15:29:36 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
0f8dd825c3
system-prompts-and-models-o.../docs/adr/0001-tier1-execution-policy-spikes.md
Sami Assiri b4531f0a4c feat(tier1): docs-governance CI, evidence gate, closure artifacts, trust/execution docs 
- Replace repo-preflight with docs-governance workflow and check_docs_links.py
- Class B bundle: require correlation_id for external_*; AuditMetadata trace fields
- Root-safe TIER1 §2; optional .githooks pre-push for main
- Add RELEASE_READINESS_MATRIX_AR, SOURCE_OF_TRUTH_INDEX, operational severity, external index
- ExecWeeklyGovernanceContract; expand trust-fabric, execution-fabric, ADR-0001, ws5, Saudi overlays
- Wire MASTER TOC, enterprise-readiness, completion-program, architecture_brief paths

Made-with: Cursor
2026-04-16 16:46:36 +03:00

3.6 KiB
Raw Blame History

ADR 0001 — Tier-1 spikes: Temporal, OPA, OpenFGA (gated)

  • Status: Proposed
  • Date: 2026-04-16
  • Context: Dealix Tier-1 roadmap targets Temporal for crash-proof long workflows and OPA / OpenFGA (or Cedar) for separable policy and fine-grained authorization. The repo today uses FastAPI, Celery, LangGraph, and in-app policy patterns.

Decision (interim)

Do not merge production dependencies on Temporal, OPA, or OpenFGA until all spike exit criteria below are met for each technology independently.

Until then, document them only as targets in:

Spike — Temporal

Goal: One non-customer-critical workflow (or shadow mode) proving resume after worker kill and worker versioning story.

Exit criteria:

  1. Workflow code and tests in CI (docker-compose or Temporal test server).
  2. Idempotency keys on external side effects; compensation path documented.
  3. Security review for secrets and network boundaries.
  4. Runbook: deploy, rollback, and “what happens on double start.”
  5. Product owner sign-off to move additional flows.

Spike — OPA (or equivalent PDP)

Goal: One policy (e.g. environment promotion or “may send message”) evaluated from structured input (JSON) with decision logged.

Exit criteria:

  1. Policy bundle versioned in repo; CI runs opa test (or equivalent) on policies.
  2. No duplication of the same rule in prompts and OPA without explicit ownership.
  3. Latency and failure mode documented (fail closed vs open).

Spike — OpenFGA (or Cedar)

Goal: One authorization model (e.g. “user U may read DD room R for tenant T”) with tuple writes on lifecycle events.

Exit criteria:

  1. Model file + migration plan for existing RBAC.
  2. Performance test for hot paths.
  3. Audit: who changed tuples; backup/restore considered.

Evidence pack (per spike)

Each spike produces: design note, threat assumptions, test commands and results, rollback plan, and explicit Approved / Rejected / Extend pilot outcome recorded in this ADR or a superseding ADR.

Consequences

  • Engineering avoids split brain between agent narration and system-of-record workflows.
  • Compliance and security can review one policy and auth surface per decision type.

Related

Annex — Temporal operational criteria (high-risk paths)

When Temporal moves from Planned to Pilot for high-risk workflow types (money movement, regulated messaging, cross-tenant admin):

  1. Pinned workflow types — register each workflow type + task queue in the service catalog; no anonymous “catch-all” workers for governed domains.
  2. Rollback rehearsal — run a simulated rollback (cancel + compensation or documented manual playbook) in staging before every production worker upgrade that changes workflow code.
  3. Worker versioning — adopt Temporal’s worker versioning policy before expanding beyond the pilot; track preview/beta notes in ../references/tier1-external-index.md (Temporal changelog link).

These criteria are release gate rows in ../RELEASE_READINESS_MATRIX_AR.md when Temporal is claimed for a candidate build.