Ivx Aso Brief

ASO ideation + screenshot brief

Wraps content-factory/pipelines/aso/ + pipelines/intel/ideate.py into an actionable workflow. The brief becomes the input to the ivx-content-factory-pipeline skill (specifically trigger_screenshot_localizer or trigger_app_store_deployer).

When to use this skill

• User mentions an app by name + Apple/Google listing concerns.
• "What keywords should we target for X?"
• "Compare us against competitors Y and Z."
• "Generate App Store screenshots for the new release."
• "Audit our listing for app <bundle-id>."

Prerequisites

• content-factory MCP server reachable.
• firecrawl MCP server configured (this skill uses Firecrawl as the live ASO signal source — see pipelines/aso/signals_provider.py LIVE_PROVIDER slot, which is wired to Firecrawl after PR for Gap C lands).
• For council voting: OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY (any one is enough — missing keys fall back to a deterministic local heuristic, flagged in the council_log).

The workflow

1. Establish the app. Ask for App Store URL or Play Store URL. Derive bundle ID and locale.

2. Pull live signals via Firecrawl (already wired into signals_provider.py::LIVE_PROVIDER):

   • SERP snapshot for the tracked keywords
   • Top 10 competitor listings (title, subtitle, screenshots, reviews snippet)
   • Listing audit (missing localizations, short subtitle, stale build, low review volume)

3. Trigger the ideation pipeline:

   trigger_aso_keywords(
     config_path="configs/pipelines/aso_keywords.yaml",
     app_url="<store URL>",
     locales=["en-US","es-MX","ja"]
   )
   

   Or for the full brief:

   trigger_competitor_analysis(
     config_path="configs/pipelines/competitor_analysis.yaml",
     app_url="<store URL>",
     competitors=["<url1>","<url2>","<url3>"]
   )
   

4. Read the resulting IdeationBrief (returned in the task response under artefacts.ideation_json):

   • feature_inventory — features merged from app metadata + GitHub
     • Figma + signals.
   • hooks, taglines, screen_captions — council-elected.
   • image_prompts — 2 flavours per slot per locale (pastel_design_sheet + ui_close_up_dark) with captions pre-injected.
   • signals — weighted ASO signals (the why behind each ranking).
   • council_log — every voter's score + rationale per candidate.

5. Hand off to the screenshot renderer:

   trigger_screenshot_localizer(
     config_path="configs/pipelines/screenshot_localizer.yaml",
     ideation_brief="<task_id of step 4>",
     locales=["en-US","es-MX","ja"]
   )
   

   Output: screenshots/<locale>/*.png (A/B variants per slot) and storyboard.html (interactive switcher).

Surfacing results

Always show the user:

1. The council_log table — model, vote, score, one-line rationale. This is the audit trail for the creative choice and is the single most useful signal for the user to trust the output.
2. The image_prompts with locale-specific captions filled in, before kicking off rendering — gives the user a chance to course- correct before burning GPU credits.
3. The signals delta vs. the last run — "your quiz keyword rank moved from #14 to #8 last week" is far more actionable than raw numbers.

Quality + safety

• Always disclose when a council vote was a deterministic fallback (missing API key) — flagged in council_log per voter.
• Never trigger app_store_deployer (which pushes a listing change via App Store Connect) without explicit user confirmation showing the diff.
• If Firecrawl returns 0 signals, the brief still builds against the curated baseline — call that out in the response so the user knows the brief is generic, not data-driven.

Common failures