runneth-apps

Helps an admin decide who can change what Runneth knows and how Runneth behaves. Out of the box, anyone the team trusts to chat with Runneth can also edit anything in the team's shared brain. That means whoever you let chat with Runneth can change the context it references AND the rules for how it behaves for everyone on the team. Most teams do not realize this until something gets accidentally changed. This skill runs a short consultative conversation, then writes a small set of edit rules that protect the areas you choose. Idempotent. Safe to re-run.

The full creative-strategy framework as one installable stack. Orchestrates 15 skills across three layers (foundation, strategy, execution) to go from understanding a brand all the way to finished hooks, concepts, and format-ready briefs. Use when someone wants to build creative strategy for a brand end to end, set up a brand's creative foundation, map messaging angles, or generate strategically grounded ads. Triggers on "build creative strategy", "creative strategy for [brand]", "set up the creative engine", "map messaging angles", "run the creative strategy framework", "go from brand to ads". Each of the 15 underlying skills is also independently callable for narrower tasks (run brand intake, score this hook, what format should this be).

Build this org's Runneth brain from the pre-enrichment package the CSM uploaded, plus live pulls against this workspace and any wired customer integrations. Synthesizes everything across the 10 brain domains with source pills on every line, then renders the welcome card. Triggers on phrases like: "build my brain", "set up my Runneth", "run brain-onboard", "set up the brain from the uploaded package", "synthesize the brain", "process the brain package", "build the org brain", "do the brain onboarding".

Weekly competitive intelligence scan. Pulls each competitor's full active and recently-killed ad portfolio, runs survival analysis, diffs against last week's baseline to surface what actually changed, and delivers a delta report to a configured Slack channel. Triggers on the weekly schedule set at install, or explicitly on "competitor scan", "what are competitors doing", "competitive intel", "competitor watch", "run competitor watch", "what changed in the market".

The deterministic creative-strategy orchestrator. Routes every creative-strategy turn through a 12-theme classifier into a per-theme skill chain. Enforces the Synthesis Gate, surgical-edit precision, persona-conditioned generation, slice integrity, and the saved winner-definition. Use when the workspace has runneth-classic installed and the user asks for performance analysis, hooks, concepts, briefs, creative feedback, audience research, competitor teardowns, or any creative-strategy work. The chain reads brand-audit, watched-brand state, customer voice, and conversation anchors before generating output.

First-run setup for the runneth-classic pack. Does a deep silent pre-discovery across Motion brand-context, workspace-goal, saved reports, recent ad performance, custom conversions, and the public web, then runs a short conversational onboarding that confirms what was discovered and enriches the gaps. Hands off to brand-audit to deepen the brand foundation. Triggered automatically by runneth-classic's post-install step. Can be re-invoked any time with "set up runneth", "reconfigure runneth", or "rerun runneth setup".

Use to audit an existing Runneth org for misplaced routine data and one-DB-per-app violations. Triggers on "audit my routines", "audit the brain", "are my routines storing data correctly", "check my org's storage layout", "clean up the brain", "which routines should be writing to the app database", "are any apps sharing a database", "do my apps and routines match", "find misplaced data", "review where my routines write to". Inventories every Buildeth app and every routine, infers each routine's write target, classifies routines against the storage rules, and produces a migration plan file. Plan only; never modifies state.

Runs the foundation + strategy layer of the creative-strategy framework to produce a durable brand-context bundle. Brand-intake, product-catalog, review-audit, brand-relevant-keywords, competitor-analysis, and creative-strategy-engine, run in order and saved as per-skill markdown files under /agent/brain/brand-audit/<workspace-slug>/. Every creative-deep-dive, every weekly deck, every concept brief reads this bundle as upstream context — it does not get re-run on every creative review. Refreshes once a week to catch new reviews, competitor moves, and persona shifts. Triggers on first install via missing brand-context, explicit invocation ("audit the brand", "build the brand context", "run brand-audit", "refresh the brand"), or the Monday refresh routine.

Orchestrates the execution + diagnostic creative-strategy skills (9 skills total) to produce a mechanism-level deep dive on a single creative or every creative in a test set. Reads the workspace's brand-audit bundle as upstream strategic context — never re-runs foundation work. Single-creative mode produces one teardown with executable iteration recommendations; test-set mode runs the diagnostic on every creative in the set plus a cross-creative comparison. Triggers on "deep dive on", "analyze this creative", "break down ad", "what's working in", "deep dive the test", "analyze every creative in test", "iteration recs for". Each of the 9 underlying skills is also independently callable for narrower tasks ("evaluate this hook", "score this hook", "diagnose this body").

Import saved memories, custom instructions, projects, and recurring preferences from ChatGPT, Claude, or Gemini into Runneth's brain. Hand the user a provider-specific extraction prompt, parse the resulting upload, triage every item inline against the existing brain (flagging conflicts and raise candidates), render a read-only HTML review, parse chat-based approval (ranges, categories, negatives, raises), confirm explicitly, then write approved items to the user's imports directory and roll them up into a per-user imports profile. Items the user explicitly raises go to their main profile, org brain, or user.md (admin-gated). All imports are indexed so they surface at session-open. Default scope is per-user imports only — nothing else gets touched without an explicit raise. Triggers include "import my ChatGPT context", "bring over my Claude memory", "import from Gemini", or a new user mentioning substantial history in another AI provider.

Build a complete brand kit (identity, visual system, voice and messaging, application patterns, anti-patterns) comprehensive enough that another agent can produce on-brand work (ads, landing pages, decks, social, emails) without ever seeing the original brand. Enriches a landing-page-summary with brand-system inference, or builds from existing files, or conversationally from scratch. Always uses computed DOM styles (not raw stylesheet extraction) for visual system values. Produces both a markdown source-of-truth and an HTML deliverable styled in the brand's own system. Use when the user asks to "build a brand kit", "extract brand guidelines", "create a brand guide", "what's their brand DNA", "set up brand", or "give me a brand-system spec for X".

Generate A/B test backlogs and experiment ideas for a landing page. Outputs a scored, prioritized backlog of 10-20 tests grouped by element (hero, proof, form, CTA, pricing) and hypothesis type (clarity, trust, friction, urgency, specificity), each with a verbatim current state, specific variant copy, primary metric, guardrail, expected lift band, win-rate prior, min sample size estimate, effort estimate, and "why this might lose" failure mode. Always works against a fresh landing-page-summary. Reads prior CRO audit history so it never suggests fixes already shipped. Reads the brand kit if present so variant copy stays on-brand. Use when the user asks to "give me A/B tests", "what should we test", "test ideas for this page", "experiment backlog", "next sprint tests", or "after shipping the audit, what's next".

Exhaustively extract a landing page into a structured spec dense enough that another agent could nearly recreate the page from the summary alone. Every section, every link, every image, every CTA, every component, the full design system, and the full voice. Uses a headless Playwright browser to handle JS-rendered pages, redirects, and gated content that WebFetch alone cannot reach. Output is consumed by brand-kit and optimize-landing-page, but can be run standalone. Always re-fetches by default; archives prior versions as an in-system changelog. Use when the user asks to "summarize this landing page", "extract from this URL", "break this page down", "give me a landing page spec", "deep-dive this page", or pastes a URL with "what's on this".

CRO audit of a landing page. Diagnoses message-match, awareness-stage alignment, friction, proof, and CTA architecture, then returns the top 3 fixes ranked by expected lift x ease plus an A/B test queue. Enriches a landing-page-summary with conversion-rate intelligence. Designed for the workflow: creatives point to a landing page, audit the landing page to make sure it's optimized for the ads. Always runs against a fresh landing-page-summary, never a stale saved one. Use when the user asks to "optimize this landing page", "CRO review", "why isn't this converting", "what should I A/B test", "audit my landing page", "improve this page", or "what's wrong with my hero".

Builds and maintains a durable strategic brief for each connected paid channel (Meta, TikTok, Pinterest, AppLovin, Snapchat, or any other ad platform integration). Produces channel-specific strategy briefs that every future performance question reads first, an alignment-check visualization the user reacts to, an optional cross-channel synthesis when 2+ channels are connected, and a Friday drift+ping routine that catches strategy shifts and rewrites briefs only on confirmation. Triggers on first run via missing brief; can be invoked explicitly with "audit my paid account", "build the strategy brief", "what's our paid strategy", "set up strategy audit", or any variation of strategic-frame setup for paid acquisition. Channel-agnostic — auto-detects connected platforms and researches unknown ones before producing their brief.

Use when someone is building a process, app, or recurring workflow that meets the conditions for a self-improvement loop — recurrence plus taste or stakes. Guides Runneth through collaboratively designing the right feedback loop for that specific process: extracting the goal, identifying the assertion being made, routing corrections to the right storage layer, finding the natural intercept moment, and wiring in the capture mechanism. Always collaborative — Runneth proposes, the user validates the consequential decisions.

Produces an on-brand weekly performance deck as an openable HTML page at a dated app route, framed against last week's deck so the narrative continues over time. Walks each funnel stage with stage overview, new launches and active tests, deep dives on top creatives or creatives that shifted, and concrete recommendations for next week. Composes paid-strategy-audit (funnel frame), creative-deep-dive (per-creative diagnoses), and brand-kit (visual style + brand screenshots). Runs every Friday by default. Triggers on schedule, on explicit request ("build the weekly deck", "Friday deck", "performance deck", "generate this week's deck"), or as a follow-up after a fresh audit.

Monitors all connected integrations AND all active routines. Writes results to two state files and sends Slack alerts to the admin channel when anything is degraded or broken. Integration health: 5-layer checks per integration (credential, token, scope, capability, functional read). Alert on broken or degraded. Routine health: checks active status, firing schedule (overdue detection), dependency chains, and execution record freshness. Alert on broken chains, overdue routines, and execution failures. Alert types for both: first alert on status change, silence for {{SILENCE_HOURS}}h, one reminder after silence window, restored when healthy. All settings in /agent/brain/integration-health/health-config.json.

Adds Step 8b to the integration-onboarding skill. This step fires after every integration connection and notifies the admin of integration health monitoring. If an admin channel is already configured: tells them where alerts will go (Path 1). If no admin channel is configured: starts the setup flow (Path 2). This file documents the patch to apply to the existing integration-onboarding skill. Add Step 8b immediately before the existing "Step 8 — Hand off to integration-activation" section.

One-time setup for brand-audit. Asks for the brand site URL, the review sources to mine, the competitor shortlist, and the Slack channels for the Monday refresh ping. Runs automatically after install and can be re-invoked any time with "set up brand-audit", "reconfigure brand-audit", or "personalize brand-audit".

One-time setup for paid-strategy-audit. Asks the user where Runneth should post the Friday drift-check pings, which users (if any) to tag inside those posts, and which campaigns to exclude from efficiency math. Runs automatically after install and can be re-invoked any time with "set up paid-strategy-audit", "reconfigure paid-strategy-audit", or "personalize paid-strategy-audit".

First-time setup and reconfiguration skill for the competitor-intel use case. Walks through selecting competitor brands, resolving them in Motion's ad library, configuring Slack delivery, and scheduling the weekly routine. Safe to re-run at any time.