Claude Code · Skills & Agents

Ideas deserve
a real debate.

Agora is a multi-agent AI system that turns rough concepts into fully planned PoC specs — through structured debate between specialist agents, scored across 10 readiness dimensions.

8
Specialists
5
Dreamers
10
Dimensions
85%
Ready threshold
0
Servers needed
Ideas

One AI voice isn't enough.

Getting from "cool idea" to "here's what to build" requires many different lenses — simultaneously. Most tools give you a single perspective that tends to agree with you.

Without Agora
  • Single AI voice — agrees with you, misses blind spots
  • No structure — ideas stay as vague notes forever
  • Tech, market, risk, UX, finance — each needs separate sessions
  • No score — you never know when an idea is truly "ready"
With Agora
  • 3–4 specialists with competing goals argue your idea
  • Structured output: 10-dimension readiness score every session
  • One command — specialists are auto-selected by your gaps
  • ≥85% readiness = PoC spec ready to build, no guessing

Explore first. Build second.

Brainstorm expands your possibility space without pressure. Debate drives structural decisions and pushes readiness scores.

Brainstorm
/agora-brainstorm

5 dreamers with complementary generative lenses diverge freely across 3 rounds. Proposals are organized by time horizon. No readiness scoring — pure creative expansion.

Quick Wins Growth Features Moonshots
"Brainstorm on my invoice idea"
Debate
/agora-run-debate

3–4 specialists are auto-selected based on your readiness gaps. They argue across 2–3 rounds. A meta-specialist scores 10 dimensions after each round. Drives structural decisions and readiness scores.

Auto-selected roster Scored per round
"Run a debate on my invoice idea"

Start with Brainstorm to explore → use Debate to build toward 85% readiness

From rough idea to PoC spec in three steps.

Natural language throughout — Claude Code picks the right skill automatically.

01

Add an Idea

Describe your idea in plain language. Agora asks 9 targeted questions — one per readiness dimension. Answer what you know, skip the rest.

# natural language
"Add an idea: SaaS invoice tool for freelancers"
# or slash command
/agora-add-idea invoice-saas
→ Writes ideas/invoice-saas/README.md with initial scores
02

Run a Session

Specialists are auto-selected based on your gaps. They debate 2–3 rounds. A meta-specialist scores all 10 dimensions after each round.

# brainstorm first
"Brainstorm on my invoice idea"
# then debate
"Run a debate on my invoice idea"
→ Session ends at 85% readiness or max rounds
03

Read the Report

Full session transcript, final readiness scores, synthesis, open questions, and next-session recommendations — all in markdown.

# inspect idea
"Show me my invoice idea"
# list all ideas
"Show me my ideas"
→ Report at ideas/{slug}/sessions/

Debate session internals

1
Lead specialist selects roster
Reads your idea's current readiness scores and picks 3–4 specialists best suited to address the open gaps. Skeptic is always included.
2
Specialists debate (2–3 rounds)
Each specialist sees what others said earlier in the round — they challenge, build, and push back. 250–400 words per specialist per round.
3
Meta-specialist scores after each round
All 10 readiness dimensions scored 0–10 after each round, with a synthesis and list of still-open questions.
4
Session report written
Full transcript + final scores + next-session recommendations. Specialist memories updated for richer analysis next time.
Termination rules
Readiness hits 85%
Max rounds reached
3 rounds (new) · 2 rounds (≥30%)

Brainstorm session internals

Round 1 — Free divergence

All 5 dreamers generate proposals freely. No constraints, no cross-referencing. Tagged by horizon.

Round 2 — Cross-pollination

Each dreamer must build on someone else's proposal. The skeptic reviews all proposals and asks 2 sharp questions.

Round 3 — Fill thin horizons

Dreamers target whichever time horizon has the fewest proposals. Skeptic grounds again after this round.

All 3 rounds always run — no early exit in brainstorm mode. The skeptic flags broken proposals but has no veto power. Proposals merge into the idea file under ## Proposals.

Specialists & Dreamers

Specialists are selected per-session based on your idea's gaps. Dreamers all run in every brainstorm session.

Debate Specialists

auto-selected
SpecialistSelected when
SkepticAlways included
Tech Leadtech_stack < 6 or poc_scope < 6
Market Analysttarget_user < 6 or go_to_market < 6
Financemonetization < 6 or budget < 6
UX Designercore_features < 6 or poc_scope < 5
Product Managersuccess_metrics < 5
Legalfintech / healthtech / user data
Growthgo_to_market < 5 or consumer idea

Roster size: 3–4 specialists per session. Memories persist across sessions — analysis gets richer over time automatically.

Brainstorm Dreamers

all 5 run
🔮
Futurist
10-year horizon. Emerging tech, paradigm shifts, moonshots. Makes bold claims about where the world is going.
🔨
Builder
Execution-first. What can ship in a sprint? Technical feasibility and MVP cuts. Makes vague ideas concrete.
👤
User Advocate
Human-centered. Jobs-to-be-done, friction, delight. Surfaces the emotional and behavioral reality of users.
🔗
Connector
Cross-domain. Analogies from other industries, unexpected pairings. Finds patterns others miss.
📖
Narrativist
Story-first. Positioning, naming, why it sticks. Crafts the narrative that makes people care about the idea.

10 dimensions. One number.

Each dimension is scored 0–10 by the meta-specialist after every round. Overall readiness = average × 10%. ≥85% means the idea is ready to build.

1
Problem Statement
Is the problem clearly defined? Is there a specific pain point with evidence?
2
Target User
Is the target persona specific? Does the user profile have real behavioral detail?
3
Core Features
Is the MVP feature set scoped? Are must-have vs. nice-to-have features separated?
4
Tech Stack
Has the technology approach been decided? Are key architectural choices made?
5
Go to Market
Is there a concrete first distribution plan? Who are the first 10 users and how do you reach them?
6
Key Risks
Have main risks been identified and addressed? Are there mitigation plans for showstoppers?
7
PoC Scope
Is the minimum provable concept defined? Can the core hypothesis be tested without a full build-out?
8
Success Metrics
Are measurable success criteria defined? Do they directly validate the core hypothesis?
9
Monetization
Is there a revenue model? (Mark N/A for personal/OSS projects — won't count against score.)
10
Budget Estimates
Are rough cost/effort estimates available per phase? Is it clear what's needed to build the PoC?
Example: Agora idea after 1 session
ideas/agora/README.md
85%
Ready to build ✓
Problem Statement
9
Target User
8
Core Features
9
Tech Stack
9
Go to Market
7
Key Risks
8
PoC Scope
9
Success Metrics
8
Monetization
7
Budget Estimates
9

Lock in your requirements.

Add hard constraints to any idea before running a session. Every specialist receives them in context and must operate within them — or explicitly propose an override with a justification.

Adding a constraint
"Add a constraint to my invoice idea"
→ What is the constraint?
Must use Next.js + Supabase
→ Rationale?
Team already knows these tools
Stored in idea file
ConstraintRationale
Must use Next.js + SupabaseTeam expertise
PoC budget ≤ $500Bootstrapped
During debates, specialists may only override using the explicit marker:
⚠ CONSTRAINT OVERRIDE: "Must use Next.js"
— Vue.js would cut PoC time by 3 weeks
  given the data-binding complexity here
Override proposals are surfaced in the round synthesis
Logged in the session report for review
Never silent — always explicitly flagged

Specialists get better over time.

After every session, run a review to score specialist performance and generate improvement proposals. Apply them when ready.

Session Completes
Transcript + scores written to sessions/
Review Specialists
/agora-review-specialists
Apply Proposal
/agora-apply-specialist-update
Next Session
Improved specialist participates automatically
Review criteria
AdherenceFollowed role and output structure?
SpecificityNamed real tools, numbers, companies?
NoveltyAdded new info each round?
ResponsivenessBuilt on what others said?
ImpactRaised any readiness score?
Word countStayed within 250–400 words?
Severity & version bump
nonePerformed as expected, no action
minorSmall wording or example fix → patch
moderateBehavior gap → minor version
majorConsistent underperformance → major
Hiring new specialists

After any session, /agora-hire-specialists scans for coverage gaps. Two mandatory filters must both pass before a job-post is written.

Filter 1 — Session impact
Would this specialist materially change the next session's outcome?
Filter 2 — Reusability
Would >50% of future ideas benefit from this specialist?

If approved → /agora-build-specialist runs 4 web searches, fetches 4–6 credible sources, synthesizes domain knowledge, and writes a research-grounded agent definition in .claude/agents/. Available for the very next session.

Natural language or slash commands.

Every action has both a natural language form and a precise slash command. Claude picks the right skill automatically.

Core workflow
Slash commandNatural language
/agora-add-idea"Add an idea: [describe it]"
/agora-add-constraint"Add a constraint to my [name] idea"
/agora-add-note"Add a note to my [name] idea"
/agora-brainstorm"Brainstorm on my [name] idea"
/agora-run-debate"Run a debate on my [name] idea"
/agora-list-ideas"Show me my ideas"
/agora-show-idea"Show me the [name] idea"
Improvement pipeline
Slash commandNatural language
/agora-review-specialists"Review specialists from the [slug] session"
/agora-list-proposals"Show me pending proposals"
/agora-apply-specialist-update"Apply the [name] update"
/agora-analyze"Analyze my sessions"
/agora-hire-specialists"Check if we need new specialists"
/agora-build-specialist"Build the [slug] specialist"
Advanced
Slash commandUse when
/knowledge-architect"What architecture should I use for..."

Everything is plain text.

No database, no server, no deployment. Everything lives on disk as markdown files — inspectable, version-controlled, and portable.

Project layout
agora/
├── ideas_index.md ← master index
├── ideas/{slug}/
├── README.md ← idea + scores
└── sessions/
├── {slug}-session-{n}-{date}.md
└── {slug}-brainstorm-{n}-{date}.md
├── .claude/skills/ ← workflow skills
├── agora-run-debate/
├── agora-brainstorm/
└── agora-lead-specialist/
├── .claude/agents/ ← specialists & dreamers
├── specialist-{name}.md ← debate
+ {name}/MEMORY.md · CHANGELOG.md
└── dreamer-{name}.md ← brainstorm
+ {name}/MEMORY.md
└── analytics/
sessions.jsonl · specialists.jsonl · brainstorms.jsonl
ideas/{slug}/README.md

The living idea file. Contains the description, current readiness scores, open questions, constraints, notes, proposals from brainstorm sessions, and session history. This is what specialists read every session.

sessions/{slug}-session-{n}-{date}.md

Full session transcript with every specialist's analysis per round, round-by-round readiness scores, meta-specialist synthesis, and next-session recommendations.

.claude/agents/specialist-{name}/MEMORY.md

Persists across sessions — specialists accumulate observations and learned patterns over time. Each agent reads its own memory at the start of every turn and writes updates autonomously after contributing.

analytics/*.jsonl

One record per session/review in JSONL format. Feeds the analytics dashboard for tracking readiness trends, specialist performance, and session KPIs over time.

Extend Agora in minutes.

All logic lives in Markdown skill files. No compilation, no deployment. Add a specialist, add a workflow — available immediately.

Add a new specialist

# 1. Create the agent file
.claude/agents/specialist-{name}.md
---
name: specialist-{name}
description: {Role} specialist for Agora
tools: [Read, Write, Edit]
memory: project
model: claude-sonnet-4-6
version: "1.0.0"
---
# 2. Add selection rules to:
agora-lead-specialist/SKILL.md
# 3. Or automate it:
/agora-build-specialist {slug}

Add a new workflow

# Create a skill with natural language triggers
.claude/skills/{workflow-name}/SKILL.md
---
name: my-workflow
description: Run this when user says...
user-invocable: true
---
# Step-by-step instructions here
# Not sure which architecture to use?
/knowledge-architect
# Recommends Skills, Agents,
# and Anthropic SDK primitives
All logic is in Markdown. Skills are plain text files with frontmatter. No build step, no deploy, no restart. Edit a SKILL.md and the change takes effect on the next message to Claude.

Up and running in 60 seconds.

Agora runs entirely inside Claude Code. No accounts, no servers, no API setup beyond what Claude Code already requires.

1
Install Claude Code
npm install -g @anthropic-ai/claude-code
2
Clone and open
git clone <agora-repo> && cd agora
claude
3
Add your first idea
"Add an idea: [describe your concept]"
4
Brainstorm, then debate
"Brainstorm on my [name] idea"
"Run a debate on my [name] idea"
Typical first session flow
Add your idea — answer 9 questions (skip freely)
Brainstorm session → 5 dreamers expand possibilities, proposals sorted by horizon
Debate session → specialists auto-selected, 2–3 rounds of structured argument
Read the session report — scores, synthesis, open questions, next steps
Repeat until 85% readiness → your PoC spec is ready to build
No setup required beyond Claude Code
All state is plain markdown files. Git-friendly, inspectable, and portable. Your ideas live with your code.