A growth team that lives in your coding agent

Skills for Claude Code, Codex, Cursor and the rest, six specialists that read everything, and a weekly review that turns it into one pull request.

Then ask your agent: “Set up Appeeky Spec for this app.”

Skills teach your agent the workflow

The spec is YAML and markdown in your repo and a CLI that answers in JSON. Coding agents are good at exactly that: they work on the listing the way they work on your code.

Install once per repo

A skill is a short guide your agent reads when a task calls for it. Claude Code, Codex, Cursor, Gemini CLI, opencode and the others read the same 17. pnpm dlx, yarn dlx and bunx work too.

Then ask in plain words: “we're launching in Germany next week”, or “why did we drop for white noise?”. The agent picks the skill and runs the same commands you would.

zsh
$ npx skills add appeeky/skills

The same 17 skills, in every agent

  • Claude Code
  • Codex
  • Cursor
  • Grok
  • T3 Code
  • Gemini CLI
  • opencode
  • Zed
  • appeeky-spec

    Set up the spec and doctor; find gaps, edit, check and push listing changes.

  • appeeky-spec-targets

    Personas, keyword research and which searches to target.

  • appeeky-spec-localize

    A new language or country, and spare keyword space in secondary locales.

  • appeeky-spec-review

    Day-7 and day-14 results, keep or revert, the playbook.

  • appeeky-spec-cofounder

    The weekly review in any agent: one listing change, a brief and a pull request.

  • appeeky-spec-reviews

    Review themes with counts, and replies drafted within each store's limit.

  • appeeky-spec-competitors

    Finding competitors, reading their changes, what to do when one moves.

  • appeeky-spec-screenshots

    App Store screenshot sets, their sizes and counts, screenshot tests.

  • appeeky-spec-social

    Posts as files for every platform, the check, comments and replies.

  • appeeky-spec-ugc

    Creator personas, video briefs, the cost first, which hooks worked.

  • appeeky-spec-ads

    Apple Search Ads on your key, Meta and TikTok, paid searches worth targeting.

  • appeeky-spec-attribution

    Where installs and revenue come from: the SDK, smart links, data pull.

  • appeeky-spec-experiments

    Any experiment's hypothesis, metric and result; pricing tests.

  • appeeky-spec-visibility

    Whether AI assistants name the app, and the category chart rank.

  • appeeky-spec-seo

    The website: seo check, llms.txt, Search Console, Google ranks.

  • appeeky-spec-automation

    The GitHub workflows and their secrets, other CI, failed runs.

  • appeeky-spec-server

    Running it on a VM or container host: Docker, the .env, logs, updates.

--json on every command

Every command takes --json and prints the same facts as structured data, so the agent reads results instead of guessing from text. --cwd points it at another folder, and --via picks your own store keys or your Appeeky connection.

The same output works in your own scripts and tools.

a typical agent loop
$ npx @appeeky/spec coverage --json # which targets are missing
$ npx @appeeky/spec check --json # store limits and ASO rules
$ # the agent edits appeeky/store/ and explains the change
$ npx @appeeky/spec diff --json # what would go live
$ # a person pushes, or merges the pull request

Set up by your agent, in one prompt

Install the skills, paste one prompt, and watch the agent set the spec up with its own commands. It tells you which keys are missing; it never handles your key files.

  1. 01

    Teach your agent the workflow

    Claude Code, Codex, Cursor, Gemini CLI and the rest read the same skills.

  2. 02

    Paste one prompt

    In the repo of your app, ask your agent:

  3. 03

    Add the keys, then merge

    The agent never handles your key files. It tells you which ones are missing, you add them, and nothing reaches the store before you merge.

your agent - sleep-sounds

skill appeeky-spec

you

The listing, targets, workflows and specialists in one pull request. Your keys and the merge stay with you.

Six specialists, each with one job

agents install adds six specialist subagents. Each reads its part of the repo and reports back. None of them edits a file, posts or pushes.

  • appeeky-aso

    Reads Coverage, ranks, competitors, AI visibility, the category chart.

    Returns Missing targets, spare keyword space, up to two listing changes with character counts.

  • appeeky-reviews

    Reads Reviews and their summary, competitors' low ratings, revenue data.

    Returns What users ask for, a listing change that answers it, replies to draft.

  • appeeky-ads

    Reads Apple Search Ads, Meta, TikTok and attribution data.

    Returns Paid searches worth targeting, and spend findings for a person. Spec never changes ads.

  • appeeky-releases

    Reads Releases and their day-7 and day-14 checks, loops.md, the app's git history.

    Returns What won and lost, what waits for a new version, What's New text.

  • appeeky-ugc

    Reads UGC experiments, smart link and attribution results, personas, review words.

    Returns Which creators and hooks worked, and the next video as a brief for a persona.

  • appeeky-social

    Reads What shipped and won, What's New, review words, the channels that bring installs, unanswered comments.

    Returns Posts per connected account with the link to use, and comment replies as commands for the owner.

Every answer has the same shape

Each answers in the same short shape: findings with their numbers, up to two candidates, and the data that is missing.

A specialist whose data is missing is skipped, and the brief says which command fills it.

One plain CLI command per shell call, with --json, and files read with the agent's own tools, so a scheduled run can allow exactly those commands.

appeeky-aso returned

Findings
Play US covers 4 of 6 targets
“offline” is nowhere on Play
Candidates · up to two
Play en-US short description leads with offline, 78/80
Data missing
none

The co-founder review

The specialists only read. /appeeky-cofounder is the one that acts: it starts them together, picks the single change with the strongest evidence, makes it, and writes it all down for the pull request.

  1. 01

    Where things stand

    status --json says which stores are free: the others still have a release being measured.

  2. 02

    Ask the specialists

    All six start together. Each reads its part of the repo and returns findings and up to two candidates.

  3. 03

    Decide

    One change with the strongest evidence, for a free store, and never one the playbook says lost.

  4. 04

    Make it

    It edits the listing file and runs check and diff until both are clean.

  5. 05

    Write it down

    The brief, the post drafts and the video brief, and the proposal the pull request is opened from.

claude - sleep-sounds

> /appeeky-cofounder

A brief, every time

Where things stand, what each specialist found with the numbers, the change and the metric that judges it, and the good candidates it did not pick. What is for a person becomes something you approve by merging.

  • appeeky/briefs/<date>.mdthe brief
  • appeeky/social/drafts/posts, as drafts
  • appeeky/ugc/briefs/the next video, as a brief
  • .appeeky/proposal.mdbecomes the pull request

No free store, or no strong candidate: it changes nothing and says why.

appeeky/briefs/2026-11-23.md

## Where we stand

  • App Store: the 2026-11-12 release is checked on day 14, Nov 26.
  • Google Play is free.

## What the specialists found

  • aso: 4 of 6 Play US targets covered; “offline” missing.
  • reviews: “offline” in 11 reviews, 8 at 1–3★.

## The change

  • Play en-US short description leads with offline, 78/80.
  • Metric: Play US conversion, checked on day 7 and day 14.

## For a person

  • 3 comment replies to run. 2 posts and one video brief in this pull request.

## Not now

  • A new Play title: wait until this change is measured.

Installed in your repo, with clear limits

The specialists and the review are plain markdown files in your repo. Commit them, and edit them to fit your app.

Claude Code and opencode

agents install writes the specialists to .claude/agents/ and the review to .claude/commands/appeeky-cofounder.md. --for opencode writes .opencode/agent/ and .opencode/command/; --for all writes both.

Running it again keeps the files you changed, unless you pass --force. In an agent without subagents, the review works through each specialist's checklist itself, one after another.

$ npx @appeeky/spec agents install

  • .claude/

    • agents/

      • appeeky-aso.md
      • appeeky-reviews.md
      • appeeky-ads.md
      • appeeky-releases.md
      • appeeky-ugc.md
      • appeeky-social.md
    • commands/

      appeeky-cofounder.md

What the agent does

  • Read the playbook, the loops, the last brief and every file under appeeky/
  • Run read-only commands with --json: status, coverage, check, diff, log
  • Edit the listing files, and write the brief, drafts, briefs and proposal
  • Run social check, and ugc generate --dry-run to show a cost
  • Open a pull request with one change

What it never runs

push · revert · pull
nothing reaches the stores, and your files are never overwritten
reviews push
replies to reviews are drafts until you send them
social push
posts go out after the merge
social reply · hide · like · delete-comment
comment replies are commands for you to run
ugc generate --yes
clips use credits or your fal key only after the merge
visibility scan --local --yes · seo ranks --yes
nothing is spent on your own keys
  1. The agentopens a pull request
  2. A personreviews it
  3. The mergethe only gate
  4. The pushlisting-push.yml
  5. The stores✓ live

The review's instructions forbid every ✗ command above; the weekly routines also deny the ones that send or spend.

Every week, a pull request

One change per store, about once a week, is as fast as the playbook can learn: a push is judged on day 7 and day 14. So the review runs on a schedule, on your own Claude plan.

  • GitHub Actions

    examples/weekly/appeeky-spec-weekly.yml

    Mondays 07:00 UTC

    Pulls the data, asks status for the free stores, runs the review and opens the pull request. A PROPOSAL_TOKEN lets your pull request check run on it.

  • Cron on any machine

    examples/weekly/weekly.sh

    0 7 * * 1 in your crontab

    The same steps, with git, gh, Node 20 and claude installed. A few minutes a week on a laptop, a server or a small VM.

  • Docker

    WEEKLY_AT=0 7 * * 1

    the latest-agent image

    The container that runs the daily and push routines, with Claude Code and gh added. GITHUB_TOKEN needs pull requests write.

The agent

CLAUDE_CODE_OAUTH_TOKENfrom claude setup-token: your Claude Pro or Max plan

or ANTHROPIC_API_KEYbilled per use

Only for stores that are free

  • app-storemeasured until Nov 26
  • google-playfree · proposes here

status decides. When every store is busy, the week ends with nothing to propose.

Your own version of the review

  1. .github/appeeky-propose.mdread first
  2. .claude/commands/appeeky-cofounder.mdotherwise

Copy the review there to keep your own version; the routines use it when it exists.

The commands

npx skills add appeeky/skills
Teach your coding agent the 17 skills.
agents install
Write the six specialists and /appeeky-cofounder, for Claude Code, opencode or both. Files you changed are kept.
/appeeky-cofounder
The review in Claude Code: specialists, one change, the brief, the pull request.
status
What is being measured, and which stores are free to change.
doctor
What is set up, what is missing, how to fix it.
--json
Machine-readable output on every command.
Every command in the docs

Guides

More in Appeeky Spec