# Start with the ASO keyword workbench

The hosted workbench is the main route:

1. describe the app's real features, exclusions, market and language;
2. create pasteable seeds for Google Keyword Planner;
3. upload the CSV, TSV or XLSX export;
4. run the fixed intent classification;
5. calculate word weights and phrase coverage locally from the sheet;
6. optionally check shortlisted terms against Apple's independent popularity signal;
7. review the proposed app name, subtitle and 100-byte keyword field.

The hosted classifier sends the app description and raw keyword text to its configured model provider. Sheet metrics are parsed and calculated in the browser. An optional Apple server query sends shortlisted candidate terms, not the raw sheet. The deployed host must configure and verify both services; this package does not claim that live credentials or entitlement are active.

The portable skill is optional. Use it for local/private analysis, repeatable team work or an assistant that supports skills. Its folder is `aso-keyword-workflow/`; keep the internal structure intact. The released skill ZIP should contain that folder as its single top-level directory.

## Claude

Claude supports personal skill ZIP uploads when code execution and file creation are enabled.

1. Open **Customize → Skills**.
2. Click **+ → Create skill → Upload a skill**.
3. Upload the final `aso-keyword-workflow.zip`.
4. Enable the skill and ask Claude to use the ASO keyword workflow on your files.

Anthropic's current [skills instructions](https://support.claude.com/en/articles/12512180-use-skills-in-claude?utm_source=aso_planner&utm_medium=guide&utm_campaign=aso_workflow&utm_content=claude_skill_install) confirm the ZIP upload route.

## Claude Code

For a personal installation, copy the folder to:

`~/.claude/skills/aso-keyword-workflow/`

For one repository, copy it to:

`.claude/skills/aso-keyword-workflow/`

Start a new session or check `/skills`. Anthropic documents both locations in [Extend Claude with skills](https://code.claude.com/docs/en/skills?utm_source=aso_planner&utm_medium=guide&utm_campaign=aso_workflow&utm_content=claude_code_skill_locations).

## Codex

For a project installation, copy the folder to:

`.agents/skills/aso-keyword-workflow/`

Start a new Codex task in that project so the skill is rediscovered. OpenAI's official migration reference maps reusable Claude skills to `.agents/skills/*/SKILL.md`; see [Migrate from the Claude Agent SDK to the OpenAI Agents SDK](https://developers.openai.com/cookbook/examples/agents_sdk/migrate-from-claude-agent-sdk/readme?utm_source=aso_planner&utm_medium=guide&utm_campaign=aso_workflow&utm_content=codex_project_skill_path). If your Codex client exposes a supported skill installer, you can use that installer with this folder instead of copying it manually.

For an API sandbox, register the skill's parent directory as a capability directory or attach the skill using the runtime's supported mechanism. OpenAI documents current skill structure and API loading in [Skills](https://developers.openai.com/api/docs/guides/tools-skills?utm_source=aso_planner&utm_medium=guide&utm_campaign=aso_workflow&utm_content=openai_skills_guide).

## Other assistants

Attach or paste:

1. `assistant-instructions.md`;
2. `aso-keyword-workflow/references/prompt-v4.md`;
3. the raw keyword export;
4. the literal app description and source/market settings.

The plain instructions work without a skill installer. Ask the assistant to preserve the exact eight-column classifier contract and show the scoring contributions. If it cannot run the bundled scripts, use `references/no-code-fallback.md`.

## Optional local calculation

The bundled CLI requires Node.js 18 or newer:

```sh
node aso-keyword-workflow/scripts/rank.mjs \
  --classified classified.csv \
  --demand keyword-planner.csv \
  --metric google_volume \
  --weighting raw \
  --out-dir aso-output
```

The CLI validates the exact classifier columns before scoring and reads UTF-8 plus BOM-marked UTF-16LE/UTF-16BE exports. Convert any other text encoding to UTF-8 first.

The skill can still classify and calculate manually without Node. The optional Apple adapter is included both beside this guide and in the skill's `scripts/` folder. It requires Python 3 plus already authorized Apple Ads API access. Verify the local parser first with:

```sh
python3 aso-keyword-workflow/scripts/apple_ads_popularity.py --self-test
```

Before using any output in live metadata, review product truth, source/market consistency, terms of two characters or fewer, trademarks and current platform rules. The workflow prepares research candidates; it does not submit metadata or change an account.