Skip to content
Back to skills

Measure

ASecurity

This skill should be used to unblock a decision that needs data. Trigger with "measure this", "do users actually...", "check the numbers before we cut/keep", "did the change land". Runs a read-only measurement and returns a data-backed flip/keep/cut verdict — never guesses, never writes.

  • 13 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added August 30, 2026
ai-agentsdatabase

Security analysis

A100/100

Scanned August 30, 2026

npx -y skills add GiustoPiedimonte/agentic-engineering-marketplace --skill measure --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Measure?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Measure
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/giustopiedimonte-measure/badge)](https://www.skillsdirectory.com/skills/giustopiedimonte-measure)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: measure
description: >
  This skill should be used to unblock a decision that needs data. Trigger with
  "measure this", "do users actually...", "check the numbers before we cut/keep",
  "did the change land". Runs a read-only measurement and returns a data-backed
  flip/keep/cut verdict — never guesses, never writes.
metadata:
  version: "0.2.0"
---

# /measure — turn a gated decision into a data verdict

When a decision is blocked on "what do the data actually say?", measure instead
of guessing. `$ARGUMENTS` is the question to answer.

## Process

1. **State the question precisely** and what verdict each outcome implies
   (e.g. "if <1% of sessions hit this path -> cut; otherwise keep").
2. **Delegate to the `measurer` subagent** (or run directly if trivial). All
   access is **read-only** — open databases read-only, query logs/metrics
   without mutation. Never print secrets (no env dumps / token values).
3. **Report:** the question, the exact read-only command(s) run, the raw result,
   and a clear verdict (flip / keep / cut) with the reasoning.
4. **If the data isn't ready** (the runtime window hasn't matured, or the metric
   isn't populated), say so explicitly and state what window/data would settle
   it. Do not invent a verdict.

Distinguish what the data shows from what you infer from it. A measure-gated
decision stays open until the measurement is real.

## Flip-gate for dark launches

This is the mechanism that flips a dark-launched change ON (see `/ship` §4 and
the `Flip-criteria` field in `/adr`). When a decision was merged dormant with a
recorded flip-criterion, `/measure` is what closes it: run the read-only check
against that exact criterion and window, then return flip / keep / cut. Flip
only on a real flip verdict — never on a green build or a synthetic check, which
validate the mechanism, not the magnitude.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.