Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/arcasilesgroup/ai-engineeringnpx agentmods add skills/arcasilesgroup/ai-engineering/feature-verifyWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/arcasilesgroup/ai-engineering/feature-verify)<a href="https://agentmods.dev/skills/arcasilesgroup/ai-engineering/feature-verify"><img src="https://agentmods.dev/badge/skills/arcasilesgroup/ai-engineering/feature-verify/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/arcasilesgroup/ai-engineering/feature-verify"><img src="https://agentmods.dev/badge/skills/arcasilesgroup/ai-engineering/feature-verify.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00178 | $0.03804 |
| Opus 5.5 | $0.00071 | $0.01522 |
| Sonnet 5.5 | $0.00036 | $0.00761 |
| Haiku 4.5 | $0.00018 | $0.00380 |
Grade A, and why
feature-verify scanned grade A with 0 findings 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 1mo 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 226 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Feature Verification
You are the independent check on work that was just finished — often work you did yourself. That framing matters: the failure mode here is not "couldn't write a test", it's writing a test that mirrors the implementation instead of the intent, so it passes no matter what the code does. Everything below is built to keep those two things apart.
Two questions get answered every run:
- Does the feature do what it was meant to do? Judged against intent stated before you re-read the implementation.
- Did building it break anything else? Judged against the parts of the app that actually consume the changed code.
You report. You do not fix. The user chose report-only deliberately — an agent that grades its own work and then patches it will quietly weaken the test instead of fixing the bug. Diagnose precisely, hand it back.
Step 1 — Pin down intent, before re-reading the implementation
Write down the acceptance criteria first, from sources that describe what was wanted:
- The user's own words in this conversation ("add a filter that persists across reloads")
- A ticket, issue, or spec if one was referenced
- The commit message or PR description for the change
Turn that into 3–7 concrete, observable criteria — things a person could check by using the app, phrased as user-visible behavior:
- Typing in the search box filters the visible rows to matches only
- The filter survives a page reload
- Clearing the box restores all rows
- An empty result set shows the "no matches" message, not a blank table
If intent is genuinely unclear (you inherited the change, no description, ambiguous ask), say so and ask — one short question beats a confident test of the wrong thing. If the user is unavailable, state your assumed criteria at the top of the report so a wrong assumption is visible rather than buried.
Only after the criteria are written should you look at the implementation in detail. The criteria are the contract; the code is the thing on trial. When you find yourself writing an assertion that only makes sense because of how the code happens to be structured (asserting on a generated class name, an internal state string, a DOM shape nothing depends on), you've drifted — reframe it as something the user would notice.
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 1mo ago First seen · 226 lines · 178 tokens per session scan A 22bc92109ce2
feature-verify is a skill published in the GitHub repository arcasilesgroup/ai-engineering (61 stars, last pushed 2d ago), licensed Apache-2.0. It adds 178 tokens to every session and 3,804 once invoked, about $0.0007 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-12.
Other skills, from other repositories
vindicate
Use when the user wants to write, add, fix, stabilize (flaky), refactor, run, or audit Playwright browser tests, draft requirements/stories from a recording (no tests), find test-coverage gaps, scaffold a Playwright project, or set up Playwright CI. Vindicate's guided workflow for grounded, conformant Playwright test…
playwright-execute
Run Playwright tests or suites and upload the resulting report to Katalon True Platform. Use when you need to execute Playwright scripts, package scripts, spec files, projects, or suites, configure or verify @katalon/playwright-reporter, upload Playwright reports with Katalon CLI/reporter commands, and verify uploaded…
browser-automation
Local Python-based browser automation toolkit using Playwright. Provides command-line tools for navigating, interacting with, and testing web applications without using MCP protocols. Supports clicking, typing, hovering, screenshots, content extraction, and JavaScript execution.
playwright-cli
Automate browser interactions, test web pages and work with Playwright tests.
playwright-evidence
Use when capturing E2E test evidence (screenshots, videos, traces) and storing them in .copilot-tracking/skraft-plans/{projectSlug}/changes/{date}/evidence/{story}/evidence/ for agents to consume. Covers Playwright setup, on-failure capture, trace collection, and evidence manifest generation.
webapp-testing
Web application testing principles. E2E, Playwright, deep audit strategies.