convez.ai
AgenteIntegração & APIsIntermediário
meticalabs

Metica SDK Agents

meticalabs/metica-sdk-agents

Integração automática do MeticaSDK em Unity com monitoramento de ads em runtime

Ver no GitHub
InicialPluginScore · de 0 a 100

Conjunto de agentes Claude Code que integra o MeticaSDK em projetos Unity automaticamente, incluindo suporte a AppLovin MAX, e monitora comportamento de anúncios em runtime através de análise de logs de dispositivo (Android logcat / iOS idevicesyslog). Inclui integrador automático e skill de monitoramento de anúncios com protocolo de comparação A/B.

Instalação

$npx convez-ai meticalabs/metica-sdk-agents

1 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

Estrelas
2
Forks
0
Acompanhando
0
Saúde
37%
Último commit
há 3 dias
Versão
v0.1.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Integrar MeticaSDK em projetos Unity existentes com uma menção
  • Monitorar comportamento de anúncios em dispositivos reais via análise de logs
  • Comparar performance de anúncios entre rotas de controle e teste (holdout vs trial)

Veredito

Ferramenta especializada para o ecossistema MeticaSDK/Unity com automação robusta de integração e QA estruturado de comportamento de anúncios. Reduz friction significativamente se você trabalha nesse stack específico, mas escopo é estreito fora dele.

README

A tradução para o português deste README ainda está na fila.

metica-sdk-agents

Claude Code subagents and a skill that integrate MeticaSDK into Unity projects in one pass — including projects that already use AppLovin MAX — and verify integrations at runtime by analysing device logs (Android logcat / iOS idevicesyslog).

Install

Via Claude Code marketplace (recommended):

/plugin marketplace add meticalabs/metica-sdk-agents
/plugin install metica-sdk-agents@metica-sdk-agents

That's it. Claude Code clones the repo, registers three agents and one skill, and sets $CLAUDE_PLUGIN_ROOT for you.

Verify: launch Claude Code in your project and type /agents — you should see the three Unity agents under the metica-sdk-agents plugin. The ad-log-monitor skill shows up via / autocomplete (/metica-sdk-agents:ad-log-monitor) and via natural-language triggers (see "Use the ad-log-monitor skill" below).

Use the integrator

From your Unity project's root in Claude Code, mention the integrator. The name depends on how you installed:

# Marketplace install (recommended):
@metica-sdk-agents:unity-integrator

That's the whole invocation. The integrator auto-detects the Unity project (walks up from $(pwd) looking for ProjectSettings/) and fills missing API keys with placeholders you swap in later. You'll be shown a plan and asked to approve before any file is written.

Tip: unsure which form your install registered? Type @metica-sdk-agent and let autocomplete show it.

If you're outside the project, or you have several Unity projects in one workspace, pass it explicitly:

@agent-metica-sdk-agents:unity-integrator
PROJECT=/absolute/path/to/your/unity/project

Use the ad-log-monitor skill

Use this when you have a build on a device and want to QA the runtime ad behaviour. It works for any Metica integration — fresh or already shipped — and is built around a two-route protocol (record the holdout user, then the trial user, then compare).

How to trigger it. Any of these in the main conversation:

FormExample
Slash (explicit)/metica-sdk-agents:ad-log-monitor
Natural language (auto-loads the skill from its description)"start ad-log monitoring", "capture logcat for ads", "analyse this logcat at ./trial-user-android-….log", "compare trial vs holdout"

(A skill is not invoked with the @-mention syntax used for agents — use the / slash form or just a trigger phrase.)

The skill will pick the right phase from your prompt: a bare invocation starts a new capture (Phase 1); "done playing" stops and analyses (Phase 2); "compare" produces the trial-vs-holdout verdict (Phase 3).

Two-route protocol. Plan to run it twice on the same device, same build, same network — once for each route — so the comparison is apples-to-apples. Target ~5 interstitials and ~5 rewarded ads per route.

"We'll run two captures so we can compare. First with the holdout user (control), then with the trial user (Metica active). On each run, play until you've seen roughly 5 interstitials and 5 rewarded ads. Don't change device, network, or app version between the two runs."

Optional baseline (recommended). Holdout and trial are dev builds you provide; you can also capture a third baseline route on the live App Store / Play Store build. It's not a third A/B arm — it's a fidelity anchor that confirms your holdout dev-build reproduces production before anyone trusts the trial-vs-holdout numbers. The comparison treats it as context only, never as the pass/fail gate.

Minimal example. In the main conversation, three turns end-to-end:

you  >  /metica-sdk-agents:ad-log-monitor start the holdout capture
claude > [runs log-monitor-start.sh --label=holdout-user, prints session block,
         "now play the game"]
you  >  [plays ~5 interstitials + ~5 rewarded ads, comes back]
       done
claude > [runs log-monitor-stop.sh, reads the log, writes ./holdout-user-analysis.md]
you  >  [swaps to the trial user on the device, repeats: "start the trial capture"
         → play → "done" → analysis lands at ./trial-user-analysis.md]
you  >  compare
claude > [reads both analyses + both raw logs, writes ./compare-trial-vs-holdout.md]

If you already have logs from elsewhere and want to skip the capture:

you  >  /metica-sdk-agents:ad-log-monitor analyse the 2 existing logs at
        ./holdout-user-android-….log and ./trial-user-android-….log

Output artifacts (all written to the current working directory; multiple captures coexist by label and timestamp):

FileSourceWhat it is
./<label>-<platform>-<YYYYMMDDThhmmssZ>.logstart.shRaw capture (Android logcat or iOS syslog)
./<label>.sessionstart.shShell-readable handoff state (label/platform/pid/log path). Removed by stop.sh; never sourced.
./<label>-filtered.txtthe skillGrep-filtered subset, useful for browsing
./<label>-analysis.mdthe skillPer-route analysis (stack inventory, per-format stats, network attribution, revenue, floor handoff, load strategy, errors)
./compare-<trial>-vs-<holdout>.mdthe skillPhase 3 side-by-side verdict with the n=5 caveat

The Phase 1 script clears the device's main logcat buffer for a clean capture — that wipes existing log history for every app on the device, not just the target. The skill warns the user before it runs the script.

The four components

NameTypeRole
@metica-sdk-agents:unity-compat-checkerAgentDetects Unity / Java / MaxSDK / Android API / MeticaSDK install. PASS or BLOCK with a precise remediation hint.
@metica-sdk-agents:unity-integratorAgentOrchestrator. Discovers whether MaxSDK is present, presents a plan, snapshots git, generates code, invokes the validator.
@metica-sdk-agents:unity-validatorAgentIndependent verification of any integration. Reads the project's code and reasons about every rule (structural + behavioral); behavioral verdicts carry line-cited evidence. Plus a Unity batch compile.
/metica-sdk-agents:ad-log-monitorSkillRuntime QA on a connected device. Captures Android logcat / iOS idevicesyslog while QA plays, extracts ad unit IDs / network / revenue / lifecycle / Metica→MAX floor handoff, and compares holdout vs trial.

Most users only ever invoke the integrator. The compat-checker and validator are called by the integrator automatically (and are available standalone if you want to spot-check an existing integration). The ad-log-monitor skill is a separate runtime-QA workflow — invoke it directly when you have a build on a device.

Compatibility matrix

Defined in metica-versions.yaml (single source of truth):

MeticaSDKUnityJavaMAXAndroid API
2.4.0 (latest)≥ 2021.3≥ 11≥ 8.2.0≥ 23
2.2.7≥ 2021.3≥ 11≥ 8.0.0≥ 23

The compat-checker reads this file. To add a new SDK version, add an entry and bump latest:.

Running the tests

cd ~/.metica-sdk-agents   # or wherever you cloned
bash tests/run-all.sh

Test suites cover the four surviving scripts: resolver, download, compile, and log-monitor. The verification logic now lives in agent prose (reviewed by the user at run time), so it is not golden-tested.

The download suite needs a local SDK build at ../Metica SDK builds/MeticaSdk-2.4.0.unitypackage and silently skips when it's absent; the other three run on any clean clone.

Repo layout

metica-sdk-agents/
├── .claude-plugin/
│   ├── marketplace.json               # Claude Code marketplace manifest
│   └── plugin.json                    # plugin manifest
├── install.sh                         # one-line installer
├── metica-versions.yaml               # compat matrix
├── agents/
│   ├── contracts.md                   # JSON schemas for sub-agent outputs
│   ├── unity-compat-checker.md
│   ├── unity-integrator.md
│   └── unity-validator.md
├── skills/
│   └── ad-log-monitor/
│       └── SKILL.md                   # runtime ad-lifecycle verification + trial-vs-holdout comparison
├── scripts/                           # only what an agent can't do in prose
│   ├── resolve-plugin-dir.sh          # auto-detects plugin root for agents + skill
│   ├── compile-check.sh               # batch-mode Unity build behind the validator's compiles_cleanly rule
│   ├── download-metica-sdk.sh         # offered by integrator when compat-check finds MeticaSDK missing
│   ├── log-monitor-start.sh           # ad-log-monitor Phase 1: background capture + health checks
│   ├── log-monitor-stop.sh            # ad-log-monitor Phase 2a: stop capture + summary (analysis is the skill's job, Phase 2b)
│   └── templates/standalone/          # MeticaAdService.cs.tmpl — one MonoBehaviour, per-format @fmt regions
├── references/
│   ├── max-metica-api-map.tsv         # machine-readable MaxSdk → MeticaSdk map; read by both validator + integrator
│   └── max-vs-metica-2.4.0-api.md     # narrative parity doc (keep in sync with the TSV)
└── tests/                             # 4 suite scripts (+ run-all.sh) + log-monitor fixture

License

Apache License 2.0 — see LICENSE.

Support

Issues and feedback: https://github.com/meticalabs/metica-sdk-agents/issues — dev@metica.com.

Ver o original no GitHub

Bom para

Relacionados