Pretty Fly.ai

The agent skills behind the work. Not demos, not templates. The ones we actually run, packaged so they run on your machine.

See the bundles Start free
Four pass both source and portability checks today
Scroll

Bundles

Stop broken work before it ships.

One release system. Four original skills. One price.

Ready now

Ship Stack

AI helped you move faster. Now one small change can break three other things.

$179
ONE-TIME PURCHASE

Set up release checks, map what changed, prove the work, and catch prompt regressions.

  • Release Proof Starter: sets up performance, access, bundle, and dead-code checks.
  • Release Reach Map: finds files, tests, settings, and docs a change may affect.
  • Verify Phase: checks finished work against the plan and its acceptance list.
  • Regression Test: keeps prompt changes from quietly hurting past results.

Includes all four skill folders, one-command install, setup notes, and email help. If it does not install, reply to the receipt. We will fix it or refund you.

4 agent skills$179

Free

Start with the one that remembers what good looked like.

One skill, free, no card. It is original work and it passed the same source check as everything we sell.

Regression Test

You improved the prompt. Something else got worse, and you found out from a user.

Keeps a golden set of the cases that matter, committed next to your code, and shows which ones moved across your last five runs. One score tells you about today. A history tells you what your change did.

Will not write your prompts. Exits clean when the set passes and non-zero when it does not, so a build can fail on a prompt the way it fails on a test.

regression-testFree

What that looks like. Someone said the support bot sounded cold, so two lines went into the prompt. The bot did get warmer.

Rules: +- Be warm and helpful. +- Always offer the customer a next step. - Answer in at most two sentences. - If the answer is not in the policy above, reply exactly: I don't know.
case run 1 run 2 run 3 ------------------------------------------------------------ refund-window PASS PASS PASS two-sentence-limit PASS FAIL PASS unknown-stays-unknown PASS FAIL PASS ------------------------------------------------------------ passing 3/3 1/3 3/3 exit 0 non-zero 0 run 1 -> run 2: 2 regressed (two-sentence-limit, unknown-stays-unknown) run 2 -> run 3: 2 recovered (two-sentence-limit, unknown-stays-unknown)
Nobody touched the two-sentence rule or the "I don't know" rule. The refund answer never moved, which is how you know what did. That table is printed from the recorded runs, not typed. The full walkthrough runs on your machine with no API key: the worked example.

Get the files

Where should it go?

Drop your email and the skill file comes back to you. Sent by a person, usually within a few hours.

No list rental, no drip sequence you cannot leave. One email when new skills ship.

What these are not

Not a framework. Not a subscription. Not a wrapper around an API you already pay for. They are instruction files your agent reads. If that is not worth money to you, take the free ones and go.

They run in Claude Code, Cursor, Codex CLI, Gemini CLI, Copilot, and OpenClaw, because all of them read the same format. Before anything is listed here it passes a portability check: no paths to our machines, no missing files, no credentials.

Your code stays yours. Release Reach Map makes no network calls. Other tools use only the services you choose. Prompt tests may send test cases to the model provider you set.

Your license is simple. One buyer may use the skills on their own work and client work, across their own machines. Do not resell or share the files.

Delivery is manual right now. You pay, and a person emails you the file, usually within a few hours. Reply to the receipt for help. If it will not install, we will fix it or refund you.

Rather talk

Twenty minutes. No pitch.

Book a free first look

You leave with two or three specific observations that are yours to keep, whatever you decide.