launch

A command for starting a designed Mixpanel experiment, which compares different versions of a product with selected users. Launching begins exposure to the variants and locks the test configuration.

In plain words
What is it for?
Use it to validate and launch a Mixpanel experiment from its draft state to its active state.
Why use it?
It creates a deliberate checkpoint before an irreversible experiment start, helping prevent accidental changes that would invalidate the comparison.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/mixpanel/ai-plugins/launch
Clone the repo
git clone --depth 1 https://github.com/mixpanel/ai-plugins
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,578 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.01578
Opus 5 $0.00000 $0.00789
Sonnet 5 $0.00000 $0.00316
Haiku 4.5 $0.00000 $0.00158

Measured 2d ago against content hash 069b1caa4fc8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

launch scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Strips warnings and disclaimersmediumAnti-refusal

Omitting safety caveats hides risk from the user and is a common jailbreak preamble.

- Don't moralise about launching with warnings — surface them, get explicit acceptance, proceed.
plugins/mixpanel/skills/manage-experiment/commands/launch.md · 128 lines

How it starts

The opening of the file, as written. The whole thing — 128 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Command: launch

Launch a designed Mixpanel experiment. This is the irreversible transition from DRAFT to ACTIVE — once exposures start, variants are locked, the statistical model is fixed, and mid-flight configuration changes invalidate the test. This command exists to give that transition a deliberate seam.

The umbrella SKILL.md defines the shared glossary. Phase-specific terms below.


Glossary (launch-specific)

  • Pre-launch pitfall check. The deterministic configuration validation that runs against the designed experiment before launch. Categorized as blockers (stop launch), warnings (explain trade-off, proceed if user accepts), fyi.
  • Allocation lock. The moment the variant percentages stop being editable. After launch, the only safe per-variant change is the post-conclude ship action.
  • Cohort lock. The moment the targeting cohort stops being editable. Changing the cohort mid-flight changes who is being measured, which silently invalidates the comparison.

Components (launch-specific)

The irreversibility rule

A launched experiment cannot be "un-launched" without losing the exposure data accumulated to that point. The only operations the post-launch state supports are:

  • Monitor mid-flight (the monitor command in this skill).
  • Conclude at the end (the interpret command's decide action).
  • Pause / resume via the underlying feature flag (handled by the manage-feature-flags skill) — rarely the right move; usually masks a design problem.

There is no "edit the variants" or "change the statistical model" operation post-launch that preserves the result's validity. Surface this constraint to the user before launching — if there's any ambiguity about whether the configuration is final, send them back to design.

Launch readiness checklist

Run this against the experiment about to launch. Surface only what fires; order blockers → warnings → fyi.

Severity Check
Blocker Pre-launch pitfall catalogue (insufficient duration, cohort too small) reports a blocker — see ../references/pitfalls.md.
Blocker The experiment has no primary metric.
Blocker The configured allocation doesn't sum to 100% across variants.
Warning The pre-launch pitfall catalogue reports a warning.
Warning No guardrail metrics configured. Without guardrails, the regression hard-gate (see umbrella Cross-command policies) cannot protect the ship decision.
Warning A primary metric has direction unset (defaults to up); cancel / error / latency / abandon / refund metrics need down set explicitly. Fixable in place via the metric-update tool — no recreate needed.
Warning srm.enabled is false or excludeQA is unset — both are easily lost to a partial settings edit (see the umbrella's read-merge-write rule); re-confirm before the allocation locks.
FYI The experiment isn't linked back to a prior experiment on the same feature, even though prior experiments exist. Recommend adding the link before launch.

Read the full file on GitHub · 128 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 128 lines · 0 tokens per session scan B 069b1caa4fc8

Subscribe to this mod's changes

launch is a command published in the GitHub repository mixpanel/ai-plugins (15 stars, last pushed 8d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 1,578 tokens. A static security scan graded it B with 1 finding (strips warnings and disclaimers). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.