- Home
- Skills
- Agents & Orchestration
- thinktank
Works with the AI Agents you already use
thinktank
Evidence-led engineering framework for coding agents featuring multi-role delegation and independent verification.
$10
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
thinktank
Example session with this skill installed
thinktank implement and verify a new idempotent stripe webhook handler in the backend domain, including retry logic and a failing test case to start.
- Read your context and instructions
- Compiled the thinktank
I have initiated the ThinkTank engineering cycle for the Stripe webhook implementation. The task has been framed, domain contracts have been generated, and the independent checker has been assigned.
Engineering Verdict: VERIFIED
| Phase | Activity | Result | Evidence |
|---|---|---|---|
| Frame | Detected Node.js/Express stack | grounded | package.json line 12 |
| Plan | Generated thinktank-backend contract | valid | scripts/validate-plan.mjs |
| Act | Created src/webhooks/stripe.ts | success | Git diff +142 lines |
| Verify | Executed npm test tests/webhooks.spec.ts | pass | 4 tests passed, 0 failed |
| Check | Independent Checker rerun of test suite | ACCEPT | Terminal log run_id: 882f |
Verified Deliverables
// src/webhooks/stripe.ts
import { Request, Response } from 'express';
import db from '../db';
export const handleStripeWebhook = async (req: Request, res: Response) => {
const event = req.body;
// Idempotency check: Ensure we haven't processed this event ID
const existingEvent = await db.processedEvents.findUnique({
where: { id: event.id }
});
if (existingEvent) {
return res.status(200).json({ received: true, duplicate: true });
}
// Domain logic execution...
await db.$transaction(async (tx) => {
await tx.processedEvents.create({ data: { id: event.id } });
// ... update subscription status
});
return res.status(200).json({ received: true });
};
Next steps
- Review the generated
tests/webhooks.spec.tsfor edge case coverage. - Approve the PR to merge the verified implementation into the
mainbranch. - Observe the deployment logs for the new idempotency table migration.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Large-scale software engineering tasks often fail when agents make trivial edits without verifying the broader system impact. ThinkTank solves this by enforcing evidence-led engineering, requiring executed proof and independent verification for every significant change. It moves beyond simple prompting by framing tasks within the live repository, delegating domain-specific work through read-only contracts, and utilizing a multi-role architecture where the maker never issues their own acceptance verdict.
What it does
- Multi-role orchestration splits work between Parent, Prompter, Domain Worker, and Independent Checker roles.
- Evidence-led verification maintains a ledger of claims, file-line citations, and test outputs for every modification.
- Cognitive discipline follows a structured cycle of perception, situation modeling, action, and verification.
- Dynamic stack detection identifies the repository's language and framework to load relevant engineering skills.
- Safety-gated delivery enforces human-in-the-loop checkpoints for merges, deployments, and high-risk AI touchpoints.
How it works
- Perception and framing starts by reading live files and runtime probes to build a grounded situation model.
- Domain delegation uses the Prompter to generate strict, read-only contracts for specific backend, frontend, or security tasks.
- Execution and verification runs the implementation within isolated worktrees, followed by automated test rungs.
- Independent check passes the final artifact to a separate Checker role that reruns probes to issue a final verdict.
Frameworks & tools
This skill integrates with Claude Code (using Agent and Bash tools) and Hermes. It supports optional local Qdrant memory and includes deterministic validation scripts for Node.js.
Why this beats prompting it yourself
Standard prompts lack the structural "maker-checker" separation needed for production reliability. ThinkTank prevents regression by freezing acceptance criteria before code is written and requiring external proof that matches the original brief.
Use cases
- Complex migrations involving cross-file changes that require verified dependency mapping.
- Security hardening using dedicated security domain contracts to identify and fix abuse cases.
- Performance optimization where changes must be backed by simulation evidence and runtime benchmarks.
- Architectural refactoring that needs isolated worktrees and independent verification of API contracts.
Known limitations
Does not support arbitrary shell commands or external network grants without host-specific adapters. Memory features require local Qdrant setup or host-provided tools. Standard loops are limited to three correction rounds.
Requires host-native tools (Bash, Read/Write). Memory requires local Qdrant or host support. Limited to three correction rounds by default. No automatic deployment without human approval.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 27 days ago
Requires