asc-launch-workflow

ASC Launch Workflow

Use this skill as the coordinator for a full ASC launch/update pass:

1. localized metadata copy;
2. localized keyword fields;
3. localized ASC screenshot assets;
4. ASC API task JSON and, after explicit confirmation, optional ASC execution.

Each phase runs in a dedicated subagent to keep the coordinator's context lean. The coordinator builds the shared product brief, dispatches subagents with precise inputs, collects their file outputs and summaries, assembles the review bundle, and handles the ASC mutation confirmation directly with the user. The coordinator never performs the work of a child skill inline.

If any child skill is missing or cannot be loaded, stop before the affected phase and report the missing dependency.

Security And Live-Update Rules

Never ask the user to paste App Store Connect secrets, .p8 contents, JWTs, Authorization headers, or ~/.asc_secrets.

Never read or print ~/.asc_secrets, .p8 private keys, tokens, Authorization headers, or full request headers.

Treat App Store Connect writes as a two-step process:

1. Generate final ASC task JSON from reviewed artifacts.
2. Stop and ask the user to confirm the exact JSON before any POST, PATCH, DELETE, upload, submit, or state change.

Read-only ASC discovery may run when the task is clear and local access is available. Mutating ASC operations must never run in the same step that first creates the final JSON unless the user already explicitly approved that exact JSON.

Start Gate

Before starting any project inspection, subagent dispatch, hard dependency check, file generation, screenshot capture, App Store Connect discovery, or API task drafting, verify that the required inputs below are present.

Everything not listed here is either inferred from the project, defaulted automatically, or discovered at runtime. Do not ask the user for anything beyond this list.

Required inputs — the only things to block on:

Input	Required when	Why it cannot be inferred
Phases to run	Always	User decision; cannot be guessed
Target locales	Always	User decision; cannot be guessed
Pixelmator Pro PXD template path	screenshots phase	Stored in user's personal directory; not discoverable
Intended ASC operation (draft-only / generate-task-json / execute-after-confirmation)	asc-json or asc-execute phase	Determines whether live App Store mutations happen; must be explicit

Everything else is automatic — never ask for these upfront:

• Project root → use current working directory
• App name, Bundle ID, Platform, Version → read from Xcode project files
• App category and core user job → infer from README, product docs, existing metadata, feature screens
• Writing style → default to 平静内敛; note it in the run sheet; user can override any time
• Output root → auto-generate as <AppName> ASC Launch Assets YYYY-MM-DD/ in the project root
• Screenshot output folder, device model, UI test target/scheme → discover from the project
• Keyword seed words → infer from app name, feature names, existing metadata; generate a starter set if insufficient
• ASC app identity and resource IDs → discover via read-only ASC API using bundle ID found in project
• Resource scope and fields to update → derive from the requested phases

If a required input is missing, return only a short checklist and wait:

Missing required inputs:
- [field]: why it is required

Please provide these values to start.

Do not begin any workflow work — no project inspection, subagent dispatch, dependency checks, file generation, or ASC queries — until all required inputs for the requested phases are present.

Subagent Dispatch Model

Each child skill runs as a subagent. Use whatever subagent mechanism the current environment provides: the Agent tool in Claude Code, named agents in Codex, or any equivalent. The coordinator never performs child-skill work inline.

Input contract — every subagent receives:

• the absolute path to asc-product-brief.md (written in Step 2);
• the skill name and the specific inputs required by that phase (listed in each step below);
• the output root path so all artifacts land in the same folder tree.

Output contract — every subagent must return to the coordinator:

• a one-paragraph completion summary (phase, locales covered, outcome);
• the absolute paths of all files it wrote;
• any blockers, assumptions, or manual follow-up items.

The coordinator records these summaries and paths. It does not re-read the full content of subagent output files unless assembling the review bundle or constructing the next subagent's prompt.

Error handling — if a subagent reports a hard blocker, the coordinator stops that phase, reports the blocker to the user, and asks whether to continue with the remaining phases.

Orchestration Workflow

1. Normalize Scope

After the start gate passes, parse the user's packet into:

• requested phases;
• target locales;
• project root;
• output root;
• App Store Connect intent;
• hard blockers.

Create a short run sheet before executing:

Run sheet:
- Project:
- Locales:
- Phases:
- Output root:
- ASC mode:
- Stop points:
- Missing inputs:

2. Build One Shared Product Brief

This step runs in the main conversation, not a subagent. The brief is the shared foundation that all subagents read from file — building it in the main conversation ensures the coordinator fully understands the product before dispatching any work.

Inspect the local project read-only. Prefer README, product docs, existing App Store metadata, localization files, app entry points, feature screens, and screenshot folders.

Produce the brief covering:

• app name, platform, category, target audience;
• core user job;
• 3–5 implemented features with file/evidence hints;
• existing title/subtitle/metadata words;
• brand voice and forbidden claims;
• locale-specific caveats;
• assumptions.

Write the brief to:

<output-root>/asc-product-brief.md

All subagents read this file. Do not pass the full brief text in each subagent prompt; pass the file path instead.

3. Draft Metadata

Spawn a subagent using $asc-metadata.

Pass to the subagent:

• path to asc-product-brief.md;
• target locales;
• writing style;
• existing title/subtitle/metadata;
• field scope: Subtitle, Promotional Text, App Description;
• output path: <output-root>/asc-metadata.md.

Expect back:

• completion summary;
• path to asc-metadata.md.

Do not generate keywords in this phase.

4. Draft Keywords

Spawn a subagent using $asc-keywords.

Pass to the subagent:

• path to asc-product-brief.md;
• path to asc-metadata.md (so the subagent can avoid repeating title/subtitle words);
• target locales;
• app category and core user job;
• seed function words per locale;
• competitor/autocomplete evidence when available;
• output path: <output-root>/asc-keywords.md.

Expect back:

• completion summary;
• path to asc-keywords.md.

5. Produce Screenshots

Run this phase only when requested.

Spawn a subagent using $asc-screenshots.

Pass to the subagent:

• path to asc-product-brief.md;
• target locales and folder naming style;
• PXD template path;
• output folder: <output-root>/screenshots/;
• screenshot UI test details (target, scheme, test plan if available).

Note: $asc-screenshots internally dispatches its own subagents for promo point discovery and headline writing. The coordinator does not manage those inner subagents; it only waits for the final output.

Expect back:

• completion summary;
• paths to localized screenshot folders;
• paths to generated .pxd and .png files;
• dependency check result, screenshot acceptance result, export QA result.

Waiting for the screenshots subagent: The screenshot pipeline (UI test run + Pixelmator Pro PXD export) takes significantly longer than the metadata and keywords phases. If the subagent has not returned after an extended period, send it one status check before drawing any conclusions. Only after the second non-response or an explicit failure report should the coordinator treat the phase as blocked.

After receiving the subagent's response, the coordinator must verify that .pxd file paths are present before proceeding. If the response contains no .pxd paths, do not advance to Step 6. Instead, reject the result and ask the subagent to either complete the PXD processing step or report the exact reason it could not be done. Only accept a response with no .pxd files if the subagent explicitly reports a hard blocker that prevented Pixelmator Pro from running — in that case, surface the blocker to the user and ask whether to continue without the PXD step.

Do not let metadata copy automatically become screenshot headlines. Promotional headlines are owned by $asc-screenshots and must be grounded in actual screenshots.

6. Review Bundle

This step runs in the main conversation. The coordinator reads the output files from Steps 3–5 and assembles a human review bundle. Do not re-execute any child skill here; only read the files already produced.

# ASC Launch Review

## Locales

## Metadata

## Keywords

## Screenshots

## Assumptions And Risks

## Manual Checks

Check:

• metadata fields fit App Store limits;
• keyword fields are <= 100 characters and avoid title/subtitle duplication;
• screenshot assets exist and pass QA when requested;
• claims are grounded in project evidence;
• locale names are consistent across metadata, keywords, screenshots, and ASC targets;
• no live ASC mutation has happened yet.

Write the completed review bundle to <output-root>/asc-launch-review.md.

Ask the user to review if Stop after each phase is enabled or if the bundle contains assumptions that affect live App Store content.

7. Generate ASC Task JSON

Run this phase only when requested.

Spawn a subagent using $asc-api for task JSON generation only (read-only discovery + JSON drafting, no execution).

Pass to the subagent:

• path to asc-metadata.md;
• path to asc-keywords.md;
• ASC app identity (bundle ID, app name, or app ID);
• target locales and resource scope;
• target version state if relevant;
• output path: <output-root>/asc-tasks.json.

Typical field mappings for the subagent to verify against OpenAPI:

• Subtitle → subtitle
• Promotional Text → promotionalText
• App Description → description
• Keywords → keywords

Expect back:

• completion summary;
• the full contents of asc-tasks.json (the coordinator must show this to the user);
• the matched ASC target summary (App, Bundle ID, Platform, Version, Locale, Resource, ID, Evidence).

After receiving the task JSON, the coordinator shows it to the user in the main conversation and waits for explicit confirmation. Do not proceed to Step 8 until the user confirms the exact JSON shown.

8. Execute ASC Tasks

Run this phase only if the user explicitly confirms the exact task JSON from Step 7.

Spawn a subagent using $asc-api for execution only.

Pass to the subagent:

• the confirmed asc-tasks.json path;
• explicit instruction that this JSON has been user-confirmed and execution is authorized.

Expect back:

• per-task execution summary: method, endpoint path, status code, resource IDs, changed fields;
• any redacted error bodies for failures.

Never print secrets, tokens, Authorization headers, or request headers.

After execution, the coordinator must run a read-only verification pass. Query ASC to confirm that the live values match what was submitted:

• For metadata tasks: re-fetch the relevant appInfoLocalizations and appStoreVersionLocalizations and compare each submitted field against the live value.
• For screenshot tasks: re-fetch the appScreenshotSets and confirm each uploaded screenshot has COMPLETE state, not FAILED or AWAITING_UPLOAD.

If any field mismatches or screenshot failures are found, report them explicitly. Do not declare the workflow complete until the verification pass confirms all submitted values are live and correct.

Output Layout

Prefer a single top-level folder named by the app and date when the user did not provide one:

ASC Launch Assets YYYY-MM-DD/
├── asc-product-brief.md
├── asc-launch-review.md
├── asc-metadata.md
├── asc-keywords.md
├── asc-tasks.json
└── screenshots/
    ├── en-US/
    └── zh-Hans/

Final Response Contract

At the end of a full run, report:

• phases completed;
• output folder path;
• generated metadata and keyword files;
• screenshot asset folders and generated .pxd/.png files, if any;
• ASC task JSON path or confirmation status;
• hard blockers, assumptions, skipped phases, and manual follow-up;
• whether any live App Store Connect mutation occurred.

Keep the final answer short. The detailed artifact belongs in the generated Markdown report, not the chat.