adaria-ai is a single-user, local-first marketing operations agent that runs as a macOS background service. It connects to Slack, orchestrates data collection from 8 marketing sources, runs Claude-powered analysis skills, and delivers weekly growth briefings with approval-gated actions.

Built for the Adaria.ai app portfolio (Fridgify, Arden TTS, Tempy), but configurable for any mobile app team via apps.yaml.

Features

8 Analysis Skills — ASO keyword tracking, review sentiment & reply drafts, onboarding funnel analysis, SEO blog generation, short-form content ideas, SDK request aggregation, content strategy, and social media publishing.

8 Data Collectors — App Store Connect, Google Play Console, ASOMobile, Eodin SDK, Eodin Blog + Search Console + GA4, Fridgify Recipes API, YouTube Data API, Arden TTS metrics.

6 Social Platforms — Generate and publish platform-optimised marketing content to Twitter/X, Facebook, Threads, TikTok, YouTube Community, and LinkedIn. Every post is approval-gated.

Two Interaction Modes

| Mode | Trigger | Example | |------|---------|---------| | Mode A — Skill command | @adaria-ai aso fridgify | Runs full ASO analysis with keyword rankings, competitor diffs, and metadata proposals | | Mode B — Conversational | @adaria-ai how are reviews this week? | Claude answers using 4 read-only MCP tools against your marketing data |

Automated Briefings — Weekly growth reports and daily threshold alerts delivered to Slack on a launchd cron schedule.

Approval-Gated Write Actions — Blog publishing, review replies, metadata changes, and social posts require explicit Slack button approval before execution.

Architecture

                           Slack
                             |
                       +-----+-----+
                       |  daemon   |  (always-on, launchd)
                       +-----+-----+
                             |
                   +---------+---------+
                   |                   |
              Mode A              Mode B
          (skill dispatch)    (Claude + MCP tools)
                   |                   |
            skills/*.ts          tools/*.ts
                   |              (read-only)
          collectors/*.ts             |
                   |              SQLite DB
            safety.ts
        (approval gates)
                   |
         Slack Block Kit
       [Approve] [Reject]

Three separate launchd processes:

| Process | Schedule | Purpose | |---------|----------|---------| | com.adaria-ai.daemon | Always on | Reactive Slack event handler | | com.adaria-ai.weekly | Sun 23:00 UTC | Full weekly analysis + briefing | | com.adaria-ai.monitor | Daily 23:00 UTC | Threshold-based alerts |

Quick Start

# 1. Install
npm install -g @anthropic-ai/claude-code
claude /login
npm install -g adaria-ai

# 2. Configure
adaria-ai init

# 3. Run
adaria-ai start
adaria-ai status    # verify 3 launchd jobs loaded
adaria-ai doctor    # health check

CLI Commands

adaria-ai init                # Interactive setup wizard
adaria-ai init slack          # Configure Slack credentials only
adaria-ai init collectors     # Configure data source credentials
adaria-ai init social         # Configure social platform credentials

adaria-ai start               # Install and load 3 launchd plists
adaria-ai stop                # Unload all launchd plists
adaria-ai status              # Check launchd job states
adaria-ai logs                # Tail daemon logs

adaria-ai doctor              # Full health check
adaria-ai analyze             # Run weekly analysis manually
adaria-ai monitor             # Run daily monitor manually

Configuration

All runtime state lives in ~/.adaria/ (override with ADARIA_HOME):

~/.adaria/
  config.yaml          # Written by `adaria-ai init`
  apps.yaml            # Your apps, platforms, keywords, competitors
  sessions.json        # Slack thread <-> Claude session map
  audit.jsonl          # Every action logged
  data/adaria.db       # SQLite database
  logs/                # Daemon, weekly, monitor logs

Secrets are stored in macOS Keychain, never in config files.

Dev Profile

Run a separate dev instance alongside production:

# Dev profile uses ~/.adaria-dev with isolated keychain namespace
npm run init:dev
npm run smoke:collectors:dev

Skills Reference

| Skill | Command | What it does | |-------|---------|-------------| | ASO | aso <app> | Keyword rankings, rank changes, competitor diffs, metadata proposals | | Reviews | review <app> | Sentiment analysis, complaint clustering, reply drafts | | Onboarding | onboarding <app> | Funnel analysis, cohort retention, drop-off hypotheses | | SEO Blog | blog <app> | Blog post generation + Fridgify recipe content | | Short-form | shortform <app> | YouTube Shorts performance + content ideas | | SDK Requests | sdkrequest <app> | SDK event aggregation and analysis | | Content | content <app> | Pinterest pins + trend content ideas | | Social | social <app> | Generate + post to 6 platforms (approval-gated) |

Security

• Slack allowlist — Only configured user IDs can interact with the bot
• Prompt injection defense — All external data (reviews, recipes, competitor metadata) is sanitized and XML-wrapped before reaching Claude
• Read-only MCP tools — Mode B has no write access; skills are never exposed as tools
• Approval gates — Every write action requires human confirmation
• Audit log — Every invocation, skill dispatch, and approval action is logged
• No secrets in npm — Pre-publish scanner blocks tarball if credentials detected
• Keychain storage — All tokens stored in macOS Keychain, not files

Tech Stack

• Runtime: Node.js 20+, TypeScript (strict, ESM)
• AI: Claude CLI (via @anthropic-ai/claude-code)
• Messaging: Slack (Socket Mode + Bolt)
• Database: SQLite (better-sqlite3, WAL mode)
• Process management: macOS launchd
• Testing: Vitest (545 tests)

Requirements

• macOS (launchd-based process management)
• Node.js 20+
• Claude Code CLI installed and authenticated
• Slack workspace with a bot app (Socket Mode enabled)

License

MIT