Brand DNA
Build the operating brief from the company itself.
brand/profile.ymlA portable set of agent contracts for researching a company, developing creative, producing video, and preparing a Meta campaign without importing assumptions from the last run.
Add the company URL and any context the site will not reveal. The prompt tells a new agent to read the source contracts, start clean, and stop at the right gates.
Use the canonical product or homepage URL.
Keep private keys and credentials out of this field.
Clean-room rule: each run creates a fresh brand workspace and run ID. Prior company facts, assets, and performance data stay behind.
Run the complete Kiko Meta Ads Pipeline for a new company. INPUTS - Company URL: [PASTE COMPANY URL] - Founder context: No additional founder context supplied. Infer carefully from public evidence and mark uncertainty. - Reference kit: [PASTE THIS PAGE URL] OPERATING INSTRUCTIONS 1. Open the reference kit and read all seven original skill contracts in the Skill Explorer before doing any work. 2. Use the installed meta-ads skills when available. If they are unavailable, follow the exact contracts exposed in the reference kit. 3. Create a fresh company workspace and a fresh timestamped run. Do not carry brand facts, audience assumptions, creative assets, campaign IDs, or performance data from any prior company. 4. Execute in order: Brand DNA, Research, Concepts, Scripts, Produce, Publish, Track. Treat artifact existence as pipeline state so the run is resumable and idempotent. 5. Stop at Gate 1 after concepts. Ask a human to approve angles and hooks. Never fabricate the approval artifact. 6. After approval, build one claim-checked ad kit for each approved angle and hook combination. Every product claim must trace to source evidence. 7. In production, open with a human reaction and spoken hook from frame zero, keep demos free of generated text, keep edits tight, and judge every final. Allow no more than two automatic regeneration loops per failed asset before escalating. 8. Stop at Gate 2 before publishing. Ask a human to approve final creative. Never fabricate the approval artifact. 9. If Meta credentials or business IDs are missing, produce a complete dry-run payload and setup checklist. Never claim that ads were posted. 10. If publishing is authorized, create all campaigns, ad sets, and ads in PAUSED state. Never activate spend. 11. Before any paid API stage, show the expected scope, estimated cost, required credentials, and what will be created. Wait for approval. 12. Maintain a pipeline hub where every named image and video is directly viewable or playable. START NOW First inspect the company URL and return: - detected company name and canonical domain - evidence-backed product summary - workspace path and run ID - credential and cost preflight - any blocking questions Then begin Stage 0. Do not skip a stage or human gate.
Every stage produces a durable output that the next stage consumes. If a run stops, the agent resumes from the last verified artifact instead of starting over.
Build the operating brief from the company itself.
brand/profile.ymlFind paid patterns, organic signals, and audience language.
research/research.htmlTranslate evidence into angles and hooks worth testing.
concepts/concepts.jsonCreate claim-checked production kits for approved ideas.
scripts/ad-kit-*.jsonGenerate, assemble, judge, and repair the creative.
videos/final/*.mp4Create the campaign structure with every ad paused.
meta/meta-ads.jsonRead performance and feed the next creative cycle.
tracking/report.htmlA human selects the angles and hooks before scripts or paid generation begin.
A human reviews judged finals before any Meta objects are created.
The agent can prepare campaign infrastructure. It cannot activate spend.
This explorer exposes the complete operating contracts behind the prompt. An agent can inspect them. A human can audit exactly what the agent was told to do.
Extract the brand profile, ICP, claims, and design tokens. · Original snapshot
Loading source contract...New folders, new evidence, and new assumptions for every company.
Every spoken or visual product claim traces back to a source.
Two automated repair loops, then the agent asks for human judgment.
No credentials means a dry run and setup list, never a fictional publish.