Detector de Conflitos do Microsoft Intune
aharwelik/intune-conflict-detector
Detecta conflitos ocultos de configuração no Intune offline
Ver no GitHubSkill para Claude que detecta conflitos de configuração no Microsoft Intune offline, a partir de um export do Microsoft Graph. Normaliza identidades de configuração entre todos os tipos de política, expondo conflitos cross-type e OMA-URI customizados que o painel de administração esconde. Sem dependências externas, usa apenas stdlib Python.
Instalação
npx convez-ai aharwelik/intune-conflict-detector1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Identificar conflitos de política cruzada (Settings Catalog vs Endpoint Security vs Administrative Templates)
- Detectar conflitos envolvendo configurações OMA-URI customizadas invisíveis ao painel nativo
- Validar políticas Intune em CI/CD antes do deployment
- Auditar e eliminar duplicatas de configuração entre múltiplas políticas
Veredito

Ferramenta sólida para administradores Intune detectarem conflitos reais que o console nativo omite. Implementação offline com zero dependências a torna confiável para pipelines de validação e auditoria de políticas de endpoint.
README
A tradução para o português deste README ainda está na fila.
intune-conflict-detector
Find true Microsoft Intune policy conflicts across every policy type -- including the custom OMA-URI and cross-policy-type conflicts the admin center hides -- offline, from a Graph export, as a Claude Code skill.
Keywords: Microsoft Intune, Intune conflict, configuration profile conflict, settings catalog, endpoint security baseline, OMA-URI, MDM, endpoint management, policy as code, Microsoft Graph, Claude Code skill, Intune troubleshooting, device configuration conflict.
The problem
Intune does not arbitrate conflicting settings: if two policies set the same setting to different values on the same device, the result is a conflict the device cannot resolve. The native experience makes these hard to run down:
- the dashboard shows a conflict as a summary banner; you then drill into per-device, per-setting status to find the policies actually involved;
- custom OMA-URI settings do not appear in the conflict report at all, because Intune treats the custom payload only as a delivery mechanism;
- the same setting is named differently across policy types (an Administrative Template name versus its Security Baseline name, for example), so a real cross-type conflict is invisible until the settings are matched on a canonical identifier.
This tool closes those gaps. It works offline on an export of your policies and assignments, normalizes every setting to a canonical identity, intersects the assignments, and reports the true conflicts and duplicates -- including the ones the native report misses.
What it catches
- Cross-policy-type conflicts (Settings Catalog vs Endpoint Security vs Administrative Templates), matched on a canonical setting id even when the display names differ.
- Native-invisible conflicts that involve a custom OMA-URI setting.
- Assignment-aware conflicts: a conflict is only reported when the two policies actually land on overlapping groups or devices.
- Duplicates: the same setting set to the same value in multiple policies (a drift and management-overhead risk), reported at low severity.
Install
No dependencies. Python 3.10+.
git clone https://github.com/aharwelik/intune-conflict-detector
cd intune-conflict-detector
Use it
python -m intuneconflict scan examples/export.sample.json
# add --json for machine-readable findings, --strict to also fail on duplicates
Exit code is 1 when conflicts are found, 0 when clean, so it drops into CI on
a committed export. Sample run on the bundled example:
Intune configuration conflict scan
policies=7 canonical settings=4 skipped (no identity)=0
conflicts=3 duplicates=1 native-invisible conflicts=1
CONFLICTS
[HIGH] conflict device_vendor_msft_policy_config_defender_allowrealtimemonitoring
- Win - Settings Catalog - Defender (settingsCatalog) = '1'
- Finance - Endpoint Security AV (endpointSecurity) = '0'
overlap: shared devices: device:WS-002
note: cross-policy-type conflict
[HIGH] conflict ./Device/Vendor/MSFT/BitLocker/RequireDeviceEncryption
- Custom OMA-URI - BitLocker (customOmaUri) = '1'
- Settings Catalog - BitLocker (settingsCatalog) = '0'
overlap: shared devices: device:WS-002
note: not visible in the native Intune conflict report
(The third conflict, Internet Explorer logon options, is matched across an Administrative Template and a Security Baseline that name the setting differently -- see the full output by running the command.)
Use it as a Claude Code skill
This repo ships a Claude Code skill in skill/SKILL.md.
Install it so Claude can run the analyzer and explain conflicts on request:
mkdir -p ~/.claude/skills/intune-conflict-detector
cp skill/SKILL.md ~/.claude/skills/intune-conflict-detector/SKILL.md
Then, with this repo on your path, ask Claude things like "scan this Intune export for conflicts" and it will run the tool, show the evidence, and propose remediation that configures each setting in exactly one place.
Producing an export
The analyzer is read-only and offline. Producing the export needs Microsoft
Graph with a read scope (DeviceManagementConfiguration.Read.All). The export
schema and a PowerShell starting point are in docs/COLLECTOR.md.
Theory
This tool is an application of the Identity Normalization invariant from proof-carrying-ops: no conflict comparison is sound until the two settings are reduced to the same canonical identity. The canonical identity here is the Settings Catalog id or OMA-URI path; an alias map reconciles the documented "same setting, different name" cases. Once identities are normalized, conflict detection is an exact comparison of values across policies whose assignments overlap.
Honesty boundary
The analyzer compares literal values. Representation differences across policy
types (for example 1 versus Enabled) are reported as conflicts and flagged
for human confirmation, not silently reconciled. The collector snippets in
docs were not run against a live tenant as part of building this tool.
Design
Standard library only, tested on Python 3.10 through 3.13, with a no-emoji guard in CI.
Author
Anthony Harwelik -- aharwelik@gmail.com -- https://github.com/aharwelik
License
MIT. See LICENSE.