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:
| Form | Example |
|---|
| 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):
