dave/personaEngine2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Implement Stage A: rule engine, validation, audit, and safety gates

Browse Source 
Completes 109 of 121 tasks. Every remaining task needs a tenant connection
(T055, T056, T101-T103) or an Azure Automation account (T115-T121).

  354 offline Pester tests      PASS
  Engine purity (Principle IV)  PASS
  Sanitization (SC-013)         PASS  (156 files)
  Graph module loaded in tests  none  (SC-008 holds)

What landed
  - Four-layer configuration validation with stable finding codes, covering
    every VR-002 and VR-003 condition, plus a 23-fixture invalid-config corpus
  - Run loop, audit records (NDJSON through a single sink), summaries,
    reconciliation, and exit codes 0-6
  - Persistence behind a single write-body builder whose result always has
    exactly one key
  - Invoke-PersonaEngine.ps1 and Edit-PersonaEngineConfig.ps1
  - Six docs, two pipelines, traceability matrix, V-5a and sanitization records

Three deviations from tasks.md, each recorded in its status block

  T033 is not in Resolve-UserPersona. evaluationErrorThreshold is run-level
  state and the rule engine is pure; a counter there would break Principle IV.
  It lives in New-PersonaRunCounter and is applied in the run loop.

  A new src/Engine/ layer holds Invoke-PersonaEngineRun. The entry script
  imports the manifest, which requires Microsoft.Graph.Authentication, so a
  loop living only inside it could not run on a machine without the Graph SDK
  and SC-004 could not be proven at all. The entry script is now a thin
  wrapper and what ships is what is tested.

  The invalid-config corpus is generated by a committed script, with the
  generated fixtures committed too, so a reviewer sees the fixture in the diff.

Defects found by running the code, not by reading it

  Group and role ID lists were double-wrapped: @(Get-PersonaGroupIdPage ...)
  around a comma-returned array collapsed every membership list into one
  bogus space-joined entry. That is a silent false non-match, exactly what
  FR-013 exists to prevent.

  A 403 whose status appears only in the exception message parsed as $null,
  which the retry policy treats as a transport error - five requests per
  account against a tenant already refusing. Status extraction now falls back
  to the message text, bounded to 400-599.

  The sanitization scan walked tracked files only, so it covered 34 of 156
  files and none of this phase's code. It now scans untracked non-ignored
  files too, and a negative control confirms it catches a planted leak.

  Test-Json reports one error per violating location, not first-failure-only
  as the V-5a draft claimed. Record and pin corrected.

Enforcement remains blocked on the V-4 security sign-off.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Dave
2026-08-20 21:48:19 -04:00
parent c59c85dd55
commit cdc6bb33d3
124 changed files with 16638 additions and 199 deletions
Download Patch File Download Diff File Expand all files Collapse all files
specs/001-persona-engine/quickstart.md
+213
View File
@@ -0,0 +1,213 @@
# Quickstart: Persona Engine Validation
**Date**: 2026-08-20 | **Spec**: [spec.md](spec.md) | **Plan**: [plan.md](plan.md)
Runnable scenarios that prove the feature works. Ordered by the build sequence in
[plan.md](plan.md) — each stage is validatable before the next exists. Scenarios 1–3 require no
tenant, no credentials, and no network.
Structural details live in [data-model.md](data-model.md) and [contracts/](contracts/); this guide
does not repeat them.
## Prerequisites
**Current constraint**: no Azure Automation account. Everything runs locally on PowerShell 7 with
user accounts and delegated authentication. Scenario 5 is deferred; Scenarios 1–4 and 6 are all
available now.
| Scenario | Stage | Requirement |
| --- | --- | --- |
| 1–3 (offline) | A1 | PowerShell 7.4, Pester 5.x, PSScriptAnalyzer. No tenant, no network. |
| 4 (read-only) | A2 | An app registration with admin consent for the three delegated scopes, and a **non-privileged** user account to sign in with. Closes V-1 (read), V-3. |
| 5 (automation) | B | **Deferred** — no Automation account available. Closes V-3b, V-5b when it lands. |
| 6 (enforcement) | A3 | Delegated `User.ReadWrite.All`, **written security sign-off (V-4)**, reviewed `-WhatIf` evidence, and **purpose-created test accounts** as the write targets. |
### Local connection
```powershell
Connect-MgGraph -Scopes 'User.Read.All','GroupMember.Read.All','RoleManagement.Read.Directory'
```
Sign in as an ordinary user account, not a Global Administrator — see the caution in Scenario 4.
---
## Scenario 1 — Rule engine determinism, offline
Proves SC-001, SC-003, SC-008 and the `Unknown` propagation table in
[data-model.md](data-model.md).
```bash
pwsh -NoProfile -Command "Invoke-Pester ./tests/RuleEngine -Output Detailed"
```
**Expected**: all pass with no network access. Specifically:
- Every synthetic user yields exactly one outcome.
- Shuffling fixture order changes nothing.
- A rule matching at priority 10 wins over one matching at 20, and evaluation stops.
- A `MembershipRecord` with `RetrievalSucceeded = $false` yields `EvaluationError`, never a
non-match.
- Depth beyond `maxConditionDepth` is rejected rather than silently truncated.
**Disconnect the network and re-run.** Identical results, or SC-008 is not met.
---
## Scenario 2 — Configuration validation
Proves SC-009, SC-010 and the four-layer ordering.
```bash
pwsh -NoProfile -File ./Edit-PersonaEngineConfig.ps1 -ConfigPath ./config/persona-engine.example.json -ValidateOnly -NonInteractive
```
**Expected**: exit code `0`, no findings.
Then run the invalid-configuration corpus in `tests/TestData/InvalidConfigs/` — one file per VR-002
and VR-003 condition:
```bash
pwsh -NoProfile -Command "Invoke-Pester ./tests/Configuration -Output Detailed"
```
**Expected**: each file produces its documented finding code, severity, and location; blocking
findings return a non-zero exit code with **no prompt and no hang** (SC-010).
---
## Scenario 3 — Safety invariants
Proves SC-004 and SC-005 with the write adapter mocked. This suite is the reason the persistence
adapter is built last.
```bash
pwsh -NoProfile -Command "Invoke-Pester ./tests/Safety -Output Detailed"
```
**Expected**:
- Full synthetic population under `-WhatIf`: write adapter call count is exactly `0` — asserted, not
inspected (SC-004).
- Every captured request body has exactly one key, equal to `engine.targetAttribute` (SC-005).
- `New-PersonaWriteBody` throws for any other attribute name and for an attribute absent from
`approvedWritableAttributes`.
- A `-Debug` run **without** `-WhatIf` still reaches the write path — `-Debug` is not a safety
control.
---
## Scenario 4 — Read-only tenant preview (User Story 1)
The first connected run. Uses a read-only identity, so it is safe by construction rather than by
correct behaviour.
> **Sign in as a non-privileged account.** V-3 asks whether the three scopes are *sufficient*.
> A Global Administrator answers yes regardless — the scope narrows the token, but the account's
> directory roles still grant broad read access, so the run succeeds whether or not the permission
> set is correct. Running this as GA produces a green result that means nothing.
```bash
pwsh -NoProfile -File ./Invoke-PersonaEngine.ps1 -ConfigPath ./config/persona-engine.json -WhatIf -Verbose
```
**Expected**:
- A result line appears for every in-scope user, visible before the next user is processed (SC-012).
- Differences report as `WouldUpdate` with stored value, calculated value, and matched rule ID.
- Interim summaries at the configured interval; a final summary always; reconciliation passes at
every summary (SC-007).
- Zero write requests — confirm independently in the Entra sign-in and audit logs, not only from
console output.
Single-user check first, before the full population:
```bash
pwsh -NoProfile -File ./Invoke-PersonaEngine.ps1 -ConfigPath ./config/persona-engine.json -UserObjectId <ACCOUNT-OBJECT-ID> -WhatIf
```
**Also close V-1 here**: run the single-user check against a cloud-only user, a currently-synced
user, and a formerly-synced user. Confirm the persona extension **reads** on all three. The write
half of V-1 is closed in Scenario 6.
**Idempotence check** (SC-002): run twice unchanged. The second run reports the same counts and zero
additional proposed changes.
---
## Scenario 5 — Azure Automation PowerShell 7 *(deferred — Stage B)*
**Not runnable in the current stage.** No Automation account is available. Recorded here so it is
not lost, and so the Stage B entry cost stays visible.
Proves NFR-001, NFR-008 and closes V-3b and V-5b.
1. Import the module and publish the runbook with the schedule **disabled**.
2. Run the runbook with `-WhatIf` using the managed identity.
3. Record the runtime's exact PowerShell version.
4. Confirm `Test-Json -SchemaFile` behaves as observed locally in V-5a — its error-reporting
behaviour varies by version, and the layer-2 wrapper depends on it (OTD-005). **Run this first**;
it is the cheapest item with the highest chance of surprising you.
5. Confirm the three read scopes work as *application* permissions on the managed identity (V-3b).
**Expected**: the run completes with only `Microsoft.Graph.Authentication` imported, and output
matches the equivalent local `-WhatIf` run.
**Until this scenario passes, v1 is not complete.** The Definition of Done requires an Automation
PowerShell 7 run. Stage A completion is a real milestone and worth claiming — but it is not v1.
---
## Scenario 6 — Enforcement (User Story 8)
**Gated.** Do not run until all of these hold:
- [ ] V-4 security sign-off on the OTD-003 compensating controls, in writing
- [ ] `-WhatIf` impact evidence from Scenario 4 reviewed and approved
- [ ] Scenarios 1–5 passing
- [ ] Kill switch and rollback procedure documented
- [ ] OTD-001 – OTD-005 closed
```bash
pwsh -NoProfile -File ./Invoke-PersonaEngine.ps1 -ConfigPath ./config/persona-engine.json
```
**Expected**:
- Only changed values are written; unchanged users produce no request (SC-002).
- Every write body contains exactly one attribute (SC-005).
- Every `Updated` audit record carries `previousValue` — without it, rollback is impossible
retroactively (OTD-010).
- `EvaluationError` users are skipped with their stored persona intact (FR-014).
**Closes V-1 (write half)** and **V-2**: confirm the dynamic membership group built on the persona
extension populates, and that a Conditional Access policy assigned to that group applies.
---
## Verification item coverage
| Item | Closed by | Available now? |
| --- | --- | --- |
| V-1 | Scenario 4 (read) + Scenario 6 (write) | Yes |
| V-2 | Scenario 6 | Yes |
| V-3 | Scenario 4, as a non-privileged account | Yes |
| V-3b | Scenario 5 | **No — Stage B** |
| V-4 | Out-of-band security review — **gate on Scenario 6** | Yes (a conversation, not a tenant) |
| V-4a | Investigation; no scenario | Yes |
| V-5a | Scenario 2, behaviour pinned in a unit test | Yes |
| V-5b | Scenario 5 | **No — Stage B** |
## Exit code check (SC-011)
Every documented exit code must be reachable. Cover them deliberately rather than incidentally:
| Code | How to trigger |
| --- | --- |
| `0` | Scenario 4 |
| `1` | Any invalid configuration from Scenario 2 |
| `2` | Run with an unauthorized or expired identity |
| `3` | Fault injection on enumeration |
| `4` | Fault injection on a required data provider |
| `5` | Fault injection on the counter path (reconciliation defect) |
| `6` | Fault injection on an unhandled engine path |