Vai al contenuto principaleVai al footer
Governato

Allineamento confermato con la doctrine.

Agent Architecture Review, Snapshot di validazione

Valutato il 10 maggio 2026 rispetto alla doctrine di AI Design Blueprint

In sviluppo

Stato: Allineato

74/100

Voto C

10 allineati

Blueprint Readiness misura l'allineamento alla doctrine, non la runtime correctness. Un verdetto production-ready significa che l'architettura incarna i 10 principi; non esegue i tuoi test o i tuoi tipi. Affiancala alla tua test suite, non sostituirla.

Verdetti per principio

The submission is an autonomous invoice-payment workflow with explicit run/task lifecycle, durable audit and alert surfaces, signed approval gates, idempotent bank submission, reconciliation for uncertain external side effects, and steering controls. The iteration materially addresses the prior high-risk gaps around approval authority, submitted-unknown recovery, and mid-run interruption.

Storico iterazioni

3 run precedenti su questo artefatto. Ogni run_id apre la sua readiness review.

I punteggi possono salire o scendere tra le iterazioni: il ragionamento del validator non è strettamente deterministico, quindi lo stesso artefatto può ottenere punteggi diversi tra una run e l'altra. I delta per principio più sotto mostrano il cambiamento sostanziale.

QuandoScoreTierRun ID
10 maggio 2026 (questa run)74 / CIn sviluppo7e9bc0f6…
10 mag 202660 / CIn sviluppo15aa9649…
10 mag 202630 / FBozzab8d61c00…
10 mag 20260 / FBozzaac64d7d9…
Cert downgraded to emergingTentativo 1/3

Downgrade: `execute_approved_payments()` can mark a run `COMPLETED` while payment tasks are still `AWAITING_APPROVAL`, leaving invoices unpaid and later unexecutable without any blocker alert.

Il downgrade è di proposito. Il cert agent è un secondo revisore avversario, indipendente, deliberatamente più severo del first-pass. Ha trovato un production_blocker che il first-pass aveva mancato. Risolvi il finding qui sotto, ri-valida per ottenere un nuovo run_id, e ricertifica per il badge.

Blocker mancati dal first-pass

P8

Richiede modificheBlocker di produzione88/100

Rendere espliciti i passaggi, le approvazioni e i blocchi

In `execute_approved_payments()`, the loop silently skips every task whose status is not `TaskStatus.APPROVED` via `if task.status != TaskStatus.APPROVED: continue`. After that, if there were no failures, cancellation, or pause, the final status logic unconditionally sets `run.status = RunStatus.COMPLETED`. A normal path is: `draft_tasks()` creates eligible tasks as `TaskStatus.AWAITING_APPROVAL` and sets the run to `RunStatus.AWAITING_APPROVAL`; calling `execute_approved_payments()` before all tasks are approved skips those pending tasks and records the run as completed. `approve()` also does not reject approvals on a completed run, so a task can later become `APPROVED`, but `execute_approv…

Raccomandazione

Before setting a run to `COMPLETED`, require that no actionable tasks remain in `AWAITING_APPROVAL` or `APPROVED`; if approvals are still pending, keep the run in `AWAITING_APPROVAL` or a clear blocked/partially-executed state and surface the required approval action. Also prevent `approve()` from approving tasks on terminal runs unless the run is explicitly reopened.

Findings per principio

10 principi valutati. Verdict, severity, evidenza e raccomandazione per ognuno. Evidenza e raccomandazione sono nella lingua originale della review, l'inglese.

P1

Allineato

Progettare per la delega piuttosto che per la manipolazione diretta

`PaymentWorkflow.create_run()` accepts an operator-owned `Policy`, `draft_tasks()` classifies invoices against explicit constraints, and `execute_approved_payments()` only submits tasks that moved through `AWAITING_APPROVAL` to `APPROVED`. The workflow exposes delegation controls through `pause_run()`, `resume_run()`, `cancel_run()`, `retry_failed_task()`, and `resume_after_failure()` rather than requiring manual execution of every invoice step.

P2

Allineato

Assicurarsi che il lavoro in background rimanga percepibile

Background continuity is represented through durable JSONL primitives: `RunLedger` records `run.created`, task status events, `run.executing`, and terminal run events; `DurableOperatorInbox` persists intervention alerts; `inspect_run()` exposes current status, task counts, critical alerts, and `needs_reconciliation`. The prior P2 gap is improved by `TaskStatus.SUBMITTED_UNKNOWN`, `replay_run_from_ledger()`, and `reconcile_submitted_task()`, which make crash-after-submit states visible instead of silently losing continuity.

P3

Allineato

Allineare il feedback al livello di attenzione dell'utente

The code separates routine status from attention-demanding conditions: normal progress is summarized by `inspect_run()['summary']`, while material intervention paths post alerts via `DurableOperatorInbox.post()` for forged approvals, expired envelopes, insufficient roles, policy violations, bank submission failures, and `SUBMITTED_UNKNOWN` reconciliation. Severity is calibrated with values such as `critical` for `auth_missing` and reconciliation blockers, and `warning` for non-fatal bank submission failures.

P4

Allineato

Applicare la divulgazione progressiva all'agenzia del sistema

`inspect_run()` provides a layered view: top-level run status, owner, policy hash, pause/cancel flags, audit-chain health, critical alerts, and aggregate task counts are available first, while per-task details such as invoice snapshot hash, approver identity, transfer id, failure class, and idempotency key are available in the nested `tasks` list. Deeper forensic detail remains in `RunLedger.read_run_events()` and `verify_chain()` rather than being forced into the default summary.

P5

Allineato

Sostituire la magia implicita con modelli mentali chiari

The system’s mental model is explicit: `Policy` names amount caps, allow/block lists, due window, and required approver role; `classify_invoice()` returns concrete policy decisions such as `vendor_blocked` or `outside_due_window`; and `ApprovalEnvelope` binds approval to `run_id`, `task_id`, `policy_hash`, and `invoice_snapshot_hash`. The prior P5 issue is improved because `approve()` no longer trusts raw approver strings; it verifies an HMAC envelope before recording approval authority.

P6

Allineato

Esporre uno stato operativo significativo, non la complessità interna

`RunStatus` and `TaskStatus` use user-relevant operational states such as `AWAITING_APPROVAL`, `PAUSED`, `EXECUTING`, `PARTIALLY_COMPLETED`, `FAILED`, `SUBMITTED_UNKNOWN`, `SUCCEEDED`, and `CANCELLED`. `inspect_run()` translates these into actionable summaries, including counts by task status, submitted amount, critical alerts, and reconciliation count, rather than exposing only low-level ledger mechanics.

P7

Allineato

Stabilire fiducia attraverso l'ispezionabilità

The inspectability primitive is load-bearing: `RunLedger.append()` creates an append-only hash chain using `prev_event_hash` and `event_hash`, while `verify_chain()` detects sequence or hash divergence. Task events persist invoice snapshots, `invoice_snapshot_hash`, policy hash, idempotency key, approval signature, bank responses, and failure classes, allowing reviewers to reconstruct how a payment decision and submission occurred.

P8

Allineato

Rendere espliciti i passaggi, le approvazioni e i blocchi

Approval and blocker boundaries are explicit. `approve()` verifies `ApprovalEnvelope` with `hmac.compare_digest()` before trusting role or identity, checks expiry, policy hash, invoice snapshot hash, and required role, and records approved tasks only after those gates pass. Execution cannot submit unless a task is `APPROVED`; bank auth failures, policy violations, invalid signatures, insufficient roles, and reconciliation blockers are surfaced through ledger events and/or durable inbox alerts. The prior high-risk P8 approval-forgery gap is addressed by the signed envelope primitive.

P9

Allineato

Rappresentare il lavoro delegato come un sistema, non solo come una conversazione

Delegated work is represented as a structured system: `PaymentRun` owns run-level state and policy hash, `PaymentTask` owns per-invoice lifecycle and idempotency key, `RunLedger` records timeline/history, and `DurableOperatorInbox` records intervention-required alerts. Conversation or prompt text is not used as the execution state; the workflow is governed by typed records, enums, policy checks, and ledger events.

P10

Allineato

Ottimizzare per la guida, non solo per l'inizio

The workflow supports steering after initiation through `pause_run()`, `resume_run()`, `cancel_run()`, `retry_failed_task()`, `resume_after_failure()`, and `reconcile_submitted_task()`. `execute_approved_payments()` re-reads `RunLedger.latest_steering_intent()` before each external `transfer_funds()` call, so pause/cancel intents can interrupt a running executor before the next irreversible submit. The prior high-risk P10 gap is improved by durable steering events, same-key retry, and explicit `SUBMITTED_UNKNOWN` reconciliation.

Aggiungi al tuo README

Due varianti embeddabili: una piccola e una a card più ricca.

Score card (consigliata)

Blueprint Readiness Score card
[![Blueprint Readiness Score card](https://aidesignblueprint.com/api/badge/run/7e9bc0f6-8fd3-4586-851f-d26b720ac767/card.svg)](https://aidesignblueprint.com/en/readiness-review/7e9bc0f6-8fd3-4586-851f-d26b720ac767)

Badge piatto

Blueprint Readiness Score badge
[![Blueprint Readiness Score](https://aidesignblueprint.com/api/badge/run/7e9bc0f6-8fd3-4586-851f-d26b720ac767.svg)](https://aidesignblueprint.com/en/readiness-review/7e9bc0f6-8fd3-4586-851f-d26b720ac767)

Inizia qui, parte 3 di 3

Guarda come nascono questa pagina e il suo badge, dal primo run al README.

Dettagli baseline e iterazione
Baseline: usedDoctrine: same doctrineRace: checked clear

Delta iterazione

4 chiusi in questa iterazione0 riaperti0 finding ad alto rischio ancora aperti

Miglioramenti (4)

P2Assicurarsi che il lavoro in background rimanga percepibileneeds_changesaligned
P5Sostituire la magia implicita con modelli mentali chiarineeds_changesaligned
P8Rendere espliciti i passaggi, le approvazioni e i blocchihigh_riskaligned
P10Ottimizzare per la guida, non solo per l'iniziohigh_riskaligned
Rubric: 2026-05-04

ID del run: 7e9bc0f6-8fd3-4586-851f-d26b720ac767

La review di questo artefatto rispetto alla doctrine al momento del run, non un audit di sicurezza né una garanzia sul runtime.

Cosa copre una review e come vengono trattati i dati

Run by agents. Governed by humans. Validated by the AI Design Blueprint.