gata 0.1.1

Automated multi-agent pipeline that transforms news into satirical cartoons starring Gata

Gata Newsroom

An automated multi-agent pipeline that transforms daily topics into a recurring satirical cartoon series starring Gata, a serious investigative calico cat who views all geopolitics through the lens of feline priorities.

Status

Stage	Name	Status
1	Core pipeline (Satirist/Critic creative loop + image generation)	✅ Complete
2	Community config + model fallback chains	✅ Complete
3	Cultural Strategist (Framer + Resonator)	✅ Complete
4	Text Output Bundle (logs, HTML explanations, prompt card)	✅ Complete
5	Housekeeping + rules redefinition	✅ Complete
6	Trend Scout — automated topic discovery via NewsAPI.org + Gemini	✅ Complete
7	Comedy style configuration via humor.yaml	✅ Complete
8	Free-text community mode — --community accepts any description	✅ Complete
9	Multi-panel cartoon format — --panels and --layout flags	✅ Complete
10	Agent personality — inconvenience level (0–100) + dual satirist mode	✅ Complete
11	Dynamic news search — infers country + category from free-text; --community + --topic mode	✅ Complete

How it works

1. Trend Scout fetches today's top headlines for the community and ranks them by satirical potential. For free-text communities, it infers the appropriate country and news category in a single Gemini call (infer_community_profile). In --community + --topic mode, Trend Scout is bypassed entirely.
2. Cultural Strategist (Framer + Resonator loop) negotiates a cultural angle and audience-specific references for the chosen topic
3. Satirist + Critic loop iterates on a satirical cartoon concept until approved (up to 5 iterations)
4. Image Generator renders the approved concept into a PNG via a fallback chain of Gemini image models
5. Explainer produces two HTML explanation pages: one in the target language, one in English for operators
6. Bundle writer saves the full output package: image, conversation logs, HTML files, and prompt card

All agents use prioritised model fallback chains. Both dual-loop pairs (Framer/Resonator and Satirist/Critic) include a 3-pass self-review injected into both personas.

Agents

Agent	Sub-agents	LLMs	What it does
Trend Scout	—	Gemini	Fetches today's headlines from NewsAPI.org and picks the top 3 ranked by satirical potential for the community
Cultural Strategist	Framer, Resonator	Claude (Framer) · Gemini (Resonator)	Framer proposes a cultural angle and audience references; Resonator approves or challenges until the angle is specific and sharp
Creative Loop	Satirist, Critic	Claude (Satirist) · Gemini (Critic)	Satirist writes the cartoon concept and image prompt; Critic evaluates it against six quality rules and rejects with a concrete fix if any fail; loops up to 5 iterations
Image Generator	—	Gemini image models	Renders the approved image prompt into a PNG; tries up to 5 models in order before failing
Explainer	Writer, Editor	Claude (Writer) · Gemini (Editor)	Writer drafts two HTML pages (in-language for end users, English for operators); Editor approves or requests revision

Setup

python -m venv .venv && source .venv/bin/activate
pip install -e .

Required secrets

Three API keys are required:

Variable	Where to get it	Used by
ANTHROPIC_API_KEY	console.anthropic.com	Cultural Strategist, Satirist, Explainer
GEMINI_API_KEY	aistudio.google.com	Trend Scout, Critic, Image Generator, Resonator
NEWSAPI_ORG_KEY	newsapi.org	Trend Scout (headline fetching)

Option A — .env file (recommended for local development)

Create a .env file in the project root. It is git-ignored and never committed.

ANTHROPIC_API_KEY=your_anthropic_key_here
GEMINI_API_KEY=your_gemini_key_here
NEWSAPI_ORG_KEY=your_newsapi_key_here

The pipeline loads this file automatically on startup and logs credentials loaded from .env file to confirm.

Option B — environment variables (recommended for CI/CD and servers)

Export the variables in your shell before running the pipeline. The pipeline detects that no .env file is present and logs reading credentials from environment variables.

export ANTHROPIC_API_KEY=your_anthropic_key_here
export GEMINI_API_KEY=your_gemini_key_here
export NEWSAPI_ORG_KEY=your_newsapi_key_here
python pipeline.py --community uk-politics

In GitHub Actions, add the three values as repository secrets (Settings → Secrets and variables → Actions) and reference them in your workflow:

env:
  ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
  GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
  NEWSAPI_ORG_KEY: ${{ secrets.NEWSAPI_ORG_KEY }}

Usage

# Named community (exact match in communities.yaml; topic selected by Trend Scout)
python pipeline.py --community uk-politics

# Free-text community (no entry required in communities.yaml; language and tone inferred)
python pipeline.py --community "US community that dislikes Trump"
python pipeline.py --community "Communauté française qui critique Macron"

# Community + topic mode (brief from community, topic supplied directly — no Trend Scout)
python pipeline.py --community uk-politics --topic "Number 10 is becoming available for rent, again."
python pipeline.py --community "Adeptos portugueses de futebol" --topic "O Ronaldo vai levar Portugal ao mundial"

# Random community and topic
python pipeline.py

# Manual mode (bypasses communities.yaml entirely)
python pipeline.py --topic "AI hype" --audience "developers" --language "English" --tone "dry wit"

# Multi-panel cartoon (add --panels N and --layout direction to any mode above)
python pipeline.py --community uk-politics --panels 3 --layout horizontal
python pipeline.py --community portuguese-adults --panels 2 --layout vertical
python pipeline.py --topic "World Cup result" --audience "football fans" --language "English" --tone "excited" --panels 4 --layout horizontal

Multi-panel flags

Flag	Values	Default	Description
--panels	1–4	1	Number of panels in the cartoon strip
--layout	horizontal, vertical	horizontal	Panel arrangement direction

CLI flags take precedence over communities.yaml panel config, which takes precedence over the defaults. Output filename prefix: {N}{d}_ for multi-panel (e.g. 3h_english_topic.png); no prefix for single-panel.

Output is saved to a bundle folder:

• Community mode: output/{community}/{topic}/
• Manual mode: output/manual/{topic}/

Each bundle contains:

• cartoon.png — the generated image
• agent0_log.txt — Agent 0 negotiation history (cultural strategy)
• bc_log.txt — B/C creative loop history (satirist + critic exchange)
• explanation.html — in-language explanation of the joke for end users
• deep_dive_en.html — English operator deep-dive (news context, cultural references, satirical logic)
• prompt_card.txt — verbatim image prompt for standalone reuse

Communities

Communities are defined in communities.yaml. Each community specifies a target audience, output language, tone, a list of seed topics, and optionally a default panel count and layout direction.

Community	Language	Tone
uk-politics	English	Dry British wit
uk-tech-engineers	English	Dry British wit
portuguese-adults	Portuguese	Sátira política afiada
portuguese-politics	Portuguese	Sátira política afiada
us-startup-crowd	English	Sarcastic Silicon Valley cynicism

To add a new community, add an entry to communities.yaml — no code changes required.

Comedy configuration (humor.yaml)

humor.yaml controls the comedy style and agent personality for every run. All fields default to off so the file is optional.

Section	Field	Type	What it does
framer	wordplay_scan	bool	Framer actively looks for pun/wordplay opportunities
framer	joke_types	list	Menu of joke types the Framer chooses from
framer	language_register	string	Register for wordplay (vernacular, formal, …)
framer	inconvenience	0–100	How aggressively Framer surfaces uncomfortable truths
satirist	preferred_style	string	Tone commitment (deadpan, absurdist, …)
satirist	avoid	list	Joke types/styles to avoid
satirist	subversion	string	Subversion intensity (high, medium, low)
satirist	joke_explanation	bool	Add a <joke_explanation> block after each concept
satirist	inconvenience	0–100	How aggressively Satirist forces uncomfortable truths
critic	evaluate_joke_mechanics	bool	Critic checks that the chosen joke type is executed correctly
critic	flag_if_no_subversion	bool	Critic rejects straight-play concepts with no twist
critic	inconvenience	0–100	How aggressively Critic demands uncomfortable truths
critic	dual_satirist	bool	Replace adversarial Critic with a co-creating Second Satirist

Inconvenience levels: 0 = off; 1–33 = mild nudge ("look beneath the obvious"); 34–66 = medium push ("don't let the target off the hook"); 67–100 = maximum ("if the audience doesn't squirm, it isn't ready").

Dual satirist mode: when critic.dual_satirist: true, the Critic becomes a Second Satirist who builds on the first Satirist's idea rather than evaluating it against rules. The loop still terminates when the Second Satirist responds with APPROVED.

Tech Stack

• Python 3.x
• Framer — Anthropic Claude (claude-sonnet-4-6 → claude-opus-4-7 → claude-haiku-4-5-20251001)
• Resonator — Google Gemini (gemini-2.5-pro → gemini-2.5-flash → gemini-2.0-flash)
• Satirist — Anthropic Claude (same fallback chain as Framer)
• Critic — Google Gemini (gemini-3.1-flash-lite → gemini-2.5-flash → gemini-2.5-pro → gemini-3.1-pro-preview)
• Image Generator — Google Gemini image models (gemini-3.1-flash-image-preview → gemini-3.1-flash-image → gemini-3-pro-image-preview → gemini-3-pro-image → gemini-2.5-flash-image)
• Explainer Writer — Anthropic Claude (claude-sonnet-4-6, max 8192 tokens)
• Explainer Editor — Google Gemini (gemini-2.5-flash → gemini-2.5-pro → gemini-2.0-flash)
• Trend Scout — fetches today's headlines from NewsAPI.org and uses Gemini gemini-2.5-flash to rank them by satirical potential for the community

Upcoming Work

See TODO.md for the full backlog. Key items:

• Post-generation image review — vision model checks for rendering artifacts (duplicate labels, garbled text, Gata integrity failures)
• Error hardening — <verdict> parse retry, content policy → B/C feedback loop, per-run log files
• AWS distribution — upload to S3, static webpage
• Scheduling — daily automated runs via AWS EventBridge
• Voting system — funny / not funny ratings feeding back into the pipeline
• Trend Scout enhancement — multi-source aggregation, deduplication, and quality scoring