MergeProof
Generates client-ready, risk-annotated merge reports from AI coding agent worktrees so dev agencies can prove quality and justify their work.
One product, one founder, thirty days. Everything below is built to be executed inside the contest window — the official rules still decide what counts.
Decide in 20 seconds
The facts that decide whether this is worth thirty days of yours.
Founder or lead developer at a 2-10 person dev agency in North America or Europe that delivers client projects using Claude Code, Cursor or Codex.
Market demand 4.3/10
Narrow the buyer to a segment with an urgent, already-budgeted problem and reference the specific trigger event that creates willingness to pay.
Why this could work in 30 days: one founder can ship the MVP in 4 days; the plan targets a first sale on day 8; 11 sales at $99 clears $1,000 of collected revenue; strongest signal is speed to first revenue at 7.8/10.
Score and projections come from the generated report and are not guarantees. The $100K Hackathon is run by Ship or Die — read the official rules before you build.
Your first 24 hours
15 concrete actions. Tick them off as you go — nothing is saved or sent.
Product opportunity and positioning
Target customer
Founder or lead developer at a 2-10 person dev agency in North America or Europe that delivers client projects using Claude Code, Cursor or Codex.
Why now
AI coding agents are now standard in small agencies, but the client-facing documentation layer was never built; agencies already pay for agents and already write these summaries by hand.
Problem
After agents write a large share of the diff, the agency must explain what changed to a non-technical client before a demo or milestone, which means an hour or more of manual diff reading and an informal summary that omits risk.
Solution
Connect a GitHub repo, pick base and head commits, and MergeProof returns a structured report: executive summary, per-file change log, risk flags on auth/database/config/infra paths, merge readiness score, exported as a branded PDF and share link.
MergeProof sits downstream of the AI coding workflow, not inside it. Agent orchestration tools run the work; MergeProof sells the artifact that follows — a client-facing merge report an agency can attach to a milestone email or an invoice.
The buyer. A 2-10 person dev agency in North America or Europe that bills clients for web and app work and now lets Claude Code, Cursor or Codex write a large share of the diff. The buyer is the founder or lead developer: the same person who runs the agents, reviews the branches and writes the client update at 11pm. No procurement, no security review, one card.
The trigger. A branch that mixed agent output from three worktrees must be explained to a non-technical client before a Friday demo or a milestone payment. Today that means an hour of reading diffs and a hand-written summary that quietly omits auth and database changes. That trigger is recurring and budgeted, because the milestone money depends on the client understanding the work.
Why $99 clears. Two to three hours of founder review at a blended $75-150/hour is $150-450 of unpriced labour. A $99 report that takes five minutes to generate and ten to review is an easy yes, and it does not need a committee.
Differentiation. Generic AI code review outputs comments for developers. PR summarizers output engineer-facing changelogs. Neither produces a branded, client-safe PDF with risk flags and a merge readiness score that an agency can put its own logo on. MergeProof intentionally does not orchestrate agents, does not apply fixes, does not offer kanban, comments or assignments, and does not support non-Git repos.
Honest score. JEV rates this concept 60.8/100 (band C) with confidence 0.72. The two weakest criteria are market demand (4.27) and operational simplicity (4.78). The 30-day plan answers both: the wedge is anchored on a billable client deadline rather than a vague productivity claim, and the stack stays stateless on Vercel, the GitHub API, OpenAI and Stripe with no human review step inside delivery. The strongest criteria — speed to first revenue (7.84) and solo build feasibility (7.50) — are exactly what a 30-day window rewards.
Wedge, not platform. One repo, one commit range, one report, one price. Agencies that need it twice convert themselves to $199/month. Everything else waits until there is verified revenue.
Smallest sellable MVP
A paid, self-serve flow where an agency connects GitHub, selects a commit range, enters client name and logo, pays $99, and receives a branded PDF plus shareable link.
Must-have features
- GitHub OAuth and repo picker
- Base/head commit range selection
- Diff fetch via GitHub compare API with per-file patches
- LLM analysis into a fixed JSON schema (summary, file log, risks, score)
- Deterministic risk rules for auth, database, config and infra paths
- Branded PDF renderer with agency name, logo and colour
- Public share link page at /r/{id}
- Stripe Checkout for a $99 report credit
- Dashboard listing past reports with download
Explicitly out of scope
- Agent orchestration or kanban board
- IDE or MCP integration
- Automated code fixes
- Non-Git repositories
- Team collaboration, comments or assignments
- Self-hosted version
- Jira or Linear sync
- Human QA review as a paid add-on
Demo flow (what the buyer sees)
- Buyer signs in with GitHub
- Selects repo and base/head commit
- Enters client name and uploads logo
- Pays $99 via Stripe Checkout
- Backend fetches the diff and runs rule layer plus LLM pass
- Report JSON rendered to PDF and share page
- Buyer downloads PDF or sends the link to the client
Technical implementation plan
Stack
Architecture
Next.js app on Vercel handles auth, selection and rendering. GitHub OAuth stores a token per user; a server action calls the compare API for the chosen base/head and stores the per-file patches in Postgres. A deterministic rule pass scans paths and hunks for auth, database, config and infra changes, then an LLM pass fills the summary, per-file prose and score against a strict JSON schema. The merged JSON is rendered to PDF by React-PDF and to a share page at /r/{id}. Stripe Checkout grants one report credit, which is consumed at generation time.
Build steps
- Scaffold Next.js repo, deploy placeholder to Vercel
- Register GitHub OAuth app and Stripe account, connect TrustMRR
- Implement compare-diff fetch and store one diff row
- Define report JSON schema and prompt v1
- Add per-file chunking with token budget
- Build rule layer for risky paths and hunks
- Render HTML report, then PDF via React-PDF
- Add share link page and dashboard
- Gate generation behind a Stripe credit
- End-to-end paid test on a real public repo
AI components
- LLM generates executive summary and per-file change descriptions
- LLM assigns merge readiness score against a fixed rubric
- Rule layer pre-flags risky paths before the LLM writes prose
- Optional LLM pass to compress the risk list into client-safe language
Pricing and monetization
Single Report
Entry offerFor Agencies that need an occasional client report
- One branded PDF report
- Shareable link
- Risk flags and merge readiness score
Agency Monthly
For Agencies shipping multiple client projects per month
- Unlimited reports
- Custom branding
- Priority support
Model: One-time $99 per report with an optional $199/month unlimited plan for agencies producing multiple reports.
First revenue offer: I'll generate one free MergeProof report from a public repo of yours; if it's usable for a client, the next one is $99.
Expected ACV: $99
Upsell path: Pitch $199/month to every buyer after their second report, and ask each buyer for one agency introduction.
First-customer acquisition
Direct DMs to agency founders on X and LinkedIn who publicly mention using AI coding agents.
Where to find them
- X search: "AI coding agent" plus "agency"
- X search: agency founders posting Claude Code or Cursor screenshots
- X replies under AI coding agent launch threads
- LinkedIn search: "dev agency" founders posting about AI-generated code
- GitHub orgs with small public repos containing .claude or .cursor rule files
- Freelance-to-agency developers listing client work on their profile
Exact outreach motions
- DM 25 agencies on day 7 offering one free report on a public repo
- Follow up 48 hours later with an actual sample PDF
- DM 35 more on day 8 with the client-deadline angle
- DM 50 LinkedIn founders on day 11 referencing their posted AI workflow
- Reply to public posts asking about reviewing AI-generated PRs
- Day 27 deadline DM: reports before Dec 16
Message template
Saw you ship client work with AI agents. I built MergeProof: it turns a commit range into a branded PDF with an executive summary, per-file changes and risk flags, so you can send it to the client instead of writing the summary yourself. Want a free report on one of your public repos?
30-day execution roadmap
Week 1 · Build & validate · 7 days
Accounts and skeleton
- Register domain and scaffold Next.js repo on Vercel.
- Create GitHub OAuth app and Stripe account; connect TrustMRR.
- Draft landing headline and $99 offer copy.
Deliverable: Live placeholder page with payment rails ready.
Commit range to stored diff
- Build GitHub OAuth flow and repo/branch picker.
- Fetch compare diff with per-file patches via GitHub API.
- Persist the diff row in Postgres.
Deliverable: Selecting a commit range returns and stores a real diff.
Structured report JSON
- Define JSON schema: summary, file log, risks, score.
- Write system prompt with risk taxonomy and few-shot example.
- Add per-file chunking and result merge.
Deliverable: API returns valid report JSON for a test repo.
PDF and share link
- Render HTML report with agency name, logo and colour.
- Generate PDF with React-PDF.
- Publish share page at /r/{id}.
Deliverable: One commit range produces a branded PDF and link.
Self-serve paid flow
- Add Stripe Checkout for a $99 report credit.
- Gate generation behind credit consumption.
- Run one end-to-end paid test.
Deliverable: MVP live; a stranger can pay $99 and get a report.
Samples and announcement
- Generate 3 sample reports from public OSS repos.
- Publish sample PDF and share link on the landing page.
- Post a 120-word explanation of the artifact on X.
Deliverable: Public samples plus launch post.
150 qualified prospects
- Build 150-agency list from X and LinkedIn searches.
- Log handle, agent used and client-facing work.
- Send first 25 DMs with the free-report offer.
Deliverable: 150-row outreach sheet; 25 DMs sent.
Week 2 · Launch & first sale · 7 days
Convert replies
- Send 35 more DMs.
- Reply to every response within 2 hours.
- Send $99 Stripe links to warm replies.
Deliverable: First paid report.
Deliver free reports fast
- Deliver every free report within 24 hours.
- Ask 3 recipients what they would change.
- Fix the single biggest blocker found.
Deliverable: 3 delivered reports and written feedback.
Unattended self-serve
- Add branded public share page.
- Add read-only token path for private repos.
- Add clear error states for unsupported diffs.
Deliverable: Report generation without founder intervention.
LinkedIn volume
- Send 50 LinkedIn DMs to agency founders.
- Follow up day-7 non-responders with a sample.
- Post the sample in 2 relevant X threads.
Deliverable: 75 cumulative DMs; reply data captured.
Explain the artifact
- Publish a thread on what a client-ready merge report contains.
- Ship the pricing page with both tiers.
- Email free-report users the $99 link.
Deliverable: Thread live and pricing page published.
Evidence-backed flags
- Add deterministic rules for auth, db, config and infra paths.
- Constrain LLM scoring with a rubric and evidence lines.
- Re-run 3 samples and diff outputs.
Deliverable: Risk flags cite file and line evidence.
Week-2 scorecard
- Count DMs, replies, reports and revenue.
- Cut the worst-performing message angle.
- Set the week-3 revenue target.
Deliverable: Written scorecard and adjusted script.
Week 3 · Repeat & systematize · 7 days
Recurring revenue
- Add $199/mo Stripe subscription with unlimited reports.
- Add saved branding and logo upload.
- Grandfather early buyers into their first month.
Deliverable: Subscription live and purchasable.
New segments
- Add 100 agencies from GitHub orgs shipping AI-assisted repos.
- Personalise one line per prospect on their stack.
- Send 40 DMs.
Deliverable: 250 cumulative prospects in the sheet.
Zero-question first report
- Add a short text walkthrough on the dashboard.
- Auto-email the report link on completion.
- Add invoice and receipt fields.
Deliverable: First report completed with no support message.
Template v2
- Add a human verification checklist section.
- Add a disclaimer line about automated analysis.
- Test on 5 repos across different languages.
Deliverable: Report template v2 shipped.
Objection handling
- Publish an anonymized before/after sample.
- Ask buyers for written permission to quote feedback.
- Add FAQ answering 'is this just ChatGPT?'.
Deliverable: Landing page v2 with FAQ.
Repeat buyers
- Pitch the monthly plan to every one-time buyer.
- Follow up all free-report users.
- Ask each buyer for one agency introduction.
Deliverable: 2 upgrade or referral conversations started.
Unit economics
- Recompute per-report cost from API usage.
- Remove features no buyer used.
- Set the final-week revenue target.
Deliverable: Unit economics sheet.
Week 4 · Scale what works · 7 days
PDF v3
- Improve PDF typography and page breaks.
- Add a cover page with client and project name.
- Add a capped per-file appendix.
Deliverable: PDF v3 used on the next paid report.
Public signals
- Reply to X posts mentioning AI agent PRs.
- Send 40 DMs to agencies with public client case studies.
- Offer a same-day sample report.
Deliverable: 120 cumulative DMs this week.
No failed paid report
- Add retry and queue for LLM calls.
- Add a per-report token budget guard.
- Add error logging with alerting.
Deliverable: Zero failed paid reports after changes.
Two indexable pages
- Publish 'MergeProof vs manual PR summary'.
- Publish the risk taxonomy page.
- Submit to 2 launch directories.
Deliverable: 2 live pages plus directory listings.
Monthly pitches
- Pitch $199/mo to every one-time buyer again.
- Ask each buyer for one agency intro.
- Log objections verbatim.
Deliverable: Logged objections and upgrade attempts.
Deadline framing
- Send 50 DMs with a Dec 16 deadline angle.
- Follow up every unanswered thread.
- Post a final X thread with two samples.
Deliverable: 200 cumulative DMs; final thread live.
Keep support under 10 min/day
- Write a 5-question FAQ from real messages.
- Add a refund policy page.
- Confirm receipts and invoice emails work.
Deliverable: Self-serve FAQ and refund page.
Days 29-30 · Consolidate · 2 days
Reconcile the ledger
- Chase every unpaid Stripe link.
- Confirm TrustMRR shows all payments.
- Check for pending refunds or chargebacks.
Deliverable: Payment ledger reconciled against TrustMRR.
File the entry
- Capture dashboard and revenue screenshots.
- Submit before 13:00 UTC on Dec 16.
- Write a retrospective and next-30-day plan.
Deliverable: Submission filed before the deadline.
Revenue and profit scenarios
How the numbers are defined
Hackathon profit — Collected revenue − refunds − chargebacks − payment processing fees − paid marketing spend.
Business profit — Hackathon profit − operating costs (hosting, domains, tooling, AI/API delivery costs).
The contest ranks entrants on hackathon profit only. Hosting, tooling and AI/API delivery costs are not deducted by the organisers, but each sale still has to cover them — so we show both. Definitions verified on 2026-11-15 against the official rules.
Stated assumptions
- Conservative: 150 agencies DM'd at roughly 2% paid conversion yields 3 one-time $99 reports; no refunds.
- Base: 250 prospects with 8% requesting a free report and 10 paid reports at $99.
- Upside: 400 prospects plus one $150 sponsored newsletter placement yields 25 paid reports.
- Payment fees assume Stripe at 2.9% + $0.30 per charge, about 3.5-4% blended on these volumes.
- Operating costs cover Vercel Pro ($20/mo), domain ($12) and OpenAI tokens at $0.15-0.40 per report.
- No revenue is counted outside Nov 16 - Dec 16, 2026, 13:00 UTC.
- All figures are projections from stated assumptions, not verified revenue or existing customers.
Conservative
3 customers- Refunds
- $0
- Chargebacks
- $0
- Payment fees
- $12
- Paid marketing
- $0
- Hackathon profit
- $285
- Operating costs
- $25
- Business profit
- $260
Base case
10 customers- Refunds
- $0
- Chargebacks
- $0
- Payment fees
- $39
- Paid marketing
- $0
- Hackathon profit
- $951
- Operating costs
- $45
- Business profit
- $906
Optimistic
25 customers- Refunds
- $50
- Chargebacks
- $0
- Payment fees
- $97
- Paid marketing
- $150
- Hackathon profit
- $2,178
- Operating costs
- $90
- Business profit
- $2,088
Model your own numbers
Runs in your browser · nothing is saved- Collected revenue
- $990
- Payment fees (3.9%)
- $39
- Paid marketing
- $0
- Hackathon profit
- $951
- Business profit
- $906
Hypothetical arithmetic, not a forecast. The fee rate is a flat 3.9% assumption, refunds and chargebacks are left at zero here, and none of this is tracked data.
Scenario models from stated assumptions, not verified revenue.
Main risks and validation experiments
Risk 1Agencies will not trust an AI-written report in front of a client.
Show file-and-line evidence for every risk flag, add a human verification checklist and an automated-analysis disclaimer.
Send 10 sample reports generated from public repos and ask directly: would you send this to a client?
If fewer than 2 of 10 say yes after two template revisions, reposition from client-facing deliverable to internal QA summary.
Risk 2LLM risk flags are generic or wrong, creating liability.
Deterministic rule layer for auth, database, config and infra paths before the LLM writes prose.
Hand-label 3 real diffs and measure flag recall against the labels.
If recall on auth/db/config changes stays under 50% after the rule layer, stop selling risk scoring and sell the summary only.
Risk 3Cold DM conversion is too low to reach a first sale by day 8.
Two message angles tested, one free report as the entry offer, one follow-up with a real sample attached.
Send 50 DMs by day 8 and measure reply rate and free-report requests.
Below 5% reply rate, abandon cold DMs and switch to public sample threads under AI agent posts.
Risk 4Large diffs break unit economics or delivery cost.
Per-report token budget, per-file chunking and a capped appendix.
Measure token cost on 5 repositories of different sizes.
If average cost exceeds $2 per report after capping, restrict the plan to commit ranges under a fixed file count.
Risk 5Private repository access blocks onboarding.
Provide a read-only fine-grained token fallback alongside GitHub OAuth.
Run the flow against 3 private repos with a scoped token.
If private repos fail twice in testing, launch public-repo only and state the limit.
Hour-by-hour: the first 24 hours
Hours 1-2
- Read the official rules page and note the Dec 16 13:00 UTC cutoff.
- Register the domain, create Vercel, GitHub OAuth and Stripe accounts.
- Connect Stripe to TrustMRR before any customer exists.
Output: Working accounts and payment rails with zero customers.
Hours 3-5
- Ship the landing page with the $99 offer and a sample placeholder.
- Write the positioning line: we produce the document you send the client.
- Add a single email capture form.
Output: Live landing page with offer copy.
Hours 5-8
- Prototype GitHub compare-diff fetch in a script.
- Store one diff from a public repo.
- Check patch sizes and token counts.
Output: Raw diff data and a token cost estimate.
Hours 8-11
- Write prompt v1 with the report JSON schema.
- Run it against the stored diff.
- Iterate twice on summary quality.
Output: First structured report JSON.
Hours 11-14
- Render the JSON to a PDF with placeholder branding.
- Publish it as the first public sample.
- Draft a 20-agency outreach list with handles and agent used.
Output: Sample PDF plus first outreach list.
Full 30-day launch report
Read the full generated report
MergeProof — 30-Day Launch Report
Concept: MergeProof v1 · JEV: 60.8/100 (band C, confidence 0.72) · Window: Nov 16 – Dec 16, 2026, 13:00 UTC · Founder: solo, no team, no funding.
1. Product opportunity and positioning
MergeProof sells one artifact: a branded, risk-annotated merge report for a Git commit range produced by AI coding agents. It does not orchestrate agents, does not touch a kanban board, does not fix code. The customer is a 2-10 person dev agency that already bills clients and already lets Claude Code, Cursor or Codex write part of the diff.
The buying moment is not “I want AI tooling”. It is “the client demo is Friday and I have to explain what changed”. Today the founder reads diffs at night and writes a summary from memory. MergeProof returns a PDF with an executive summary, a per-file change log, risk flags on auth, database, config and infra paths, and a merge readiness score, in minutes, for $99.
Positioning line: we do not review your code, we produce the document you send the client. The generic alternative is a PR summarizer or ChatGPT pasted into a document: engineer-facing, unbranded, no risk structure. JEV's weakest criteria are market demand (4.27) and operational simplicity (4.78). This plan answers both by anchoring on a billable client deadline and by keeping delivery stateless on managed services with no human review step.
2. Smallest sellable MVP
An agency signs in with GitHub, picks a repo and a base/head commit range, enters a client name and uploads a logo, pays $99 and receives a PDF plus a share link. Four capabilities only: commit-range diff, LLM analysis into a fixed JSON schema, branded PDF and share-link renderer, Stripe paywall. Teams, comments, Jira sync, agent orchestration and non-Git repos are out. Definition of done on day 5: a stranger pays $99 and gets a report without the founder touching it.
3. Technical implementation plan
Next.js on Vercel; GitHub REST compare API for diffs; Supabase Postgres for report rows; an OpenAI model for analysis against a strict JSON schema; React-PDF for rendering; Stripe Checkout; share pages at /r/{id}. Pipeline: fetch compare, split per-file patches, run the deterministic risk-rule pass, run the LLM pass for summary and per-file prose, merge and score, render. Chunk diffs above roughly 8k tokens per file and cap total tokens per report so delivery stays under $0.50. The most likely breakage is PDF layout on long reports; mitigate by limiting the per-file appendix to the first 200 changed lines and paginating explicitly.
4. Pricing and monetization
$99 per single report (one branded PDF, share link, risk flags, score) and $199 per month for unlimited reports with custom branding. Delivery cost is cents in tokens plus about $1 of hosting, so every sale is gross-margin positive by construction. The week-one offer is one free report from a public repo, then a Stripe link for $99. The upsell path is the monthly plan, offered only after a second report, because repeat purchase is the only honest buying signal.
5. First-customer acquisition
One channel: direct DMs to agency founders on X and LinkedIn who publicly mention AI coding agents. Build a 150-row list on day 7 from searches combining “AI coding agent” with “agency”, “Claude Code” with “client work”, and founders posting Cursor screenshots. Log the agent each one uses. Send a 45-word DM offering a free report on their repo, then one follow-up after 48 hours with a real sample PDF attached. Week-one target: 60 DMs, 8 free reports delivered, 3 paid reports.
6. Detailed 30-day execution roadmap
Days 1-5 build and ship the MVP: accounts, diff fetch, LLM schema, PDF, Stripe, live by day 5. Days 6-7: three public sample reports and 150 prospects. Days 8-14: 100+ DMs, free reports delivered by hand, first paid report targeted on day 8, friction fixes and a deterministic risk layer. Days 15-21: $199/month subscription, third outreach round, report template v2, upgrade asks. Days 22-28: white-label polish, reliability and queueing, two comparison pages, 150 more DMs, expansion pitches. Days 29-30: reconcile the payment ledger against TrustMRR and file the submission before 13:00 UTC on Dec 16. Every day in the roadmap carries a named deliverable; nothing counts as done without one.
7. Revenue and profit scenarios
Projections only, not verified revenue.
| Scenario | Customers | ARPU | Gross | Refunds | Fees | Ads | Hackathon profit | Operating | Business profit |
|---|---|---|---|---|---|---|---|---|---|
| Conservative | 3 | $99 | $297 | $0 | $12 | $0 | $285 | $25 | $260 |
| Base | 10 | $99 | $990 | $0 | $39 | $0 | $951 | $45 | $906 |
| Upside | 25 | $99 | $2,475 | $50 | $97 | $150 | $2,178 | $90 | $2,088 |
Fees assume Stripe at 2.9% + $0.30 per charge. Operating costs are Vercel, domain and OpenAI tokens: not deducted from hackathon profit, deducted from business profit.
8. Main risks and validation experiments
Trust. An agency may not dare send an AI-written report to a paying client. Test: ten sample reports from public repos, ask “would you send this?”; if fewer than 2 of 10 say yes after two template revisions, reposition to internal QA. Accuracy. Risk flags can be generic. Test: hand-label three diffs and measure recall on auth, database and config changes; if the rule layer cannot reach 50%, stop selling risk scoring. Channel. Cold DMs may not convert. Test: 50 DMs by day 8 and measure reply rate; below 5%, switch to public sample threads under AI agent posts. Cost. Very large diffs can break unit economics. Test: token cost on five repos; cap per report so no report exceeds $2. A fifth risk, private-repo access, is tested with a read-only scoped token on three repos before day 5.
9. First 24-hour action plan
Hours 1-2: read the rules, register the domain, create Vercel, GitHub OAuth and Stripe accounts, connect TrustMRR. Hours 3-5: ship the landing page with the $99 offer. Hours 5-8: prototype compare-diff fetching and store one diff. Hours 8-11: prompt v1 producing the report JSON. Hours 11-14: render one PDF, publish it as the first sample, and draft the 20-agency outreach list.
This brief was generated from public data about established startups, then narrowed and scored for a solo 30-day launch. Revenue and profit figures are scenario models, not verified results, guarantees or evidence of existing customers. Validate demand yourself before you spend a month on it — and check the contest rules for what counts toward your score.
Related ideas
All ideas →Proofline Deal Room
Turns a sales rep's own call notes and product facts into a single hosted, branded deal page with a personalized ROI math block, so buyers get one link instead of a long email.
FlowQuote Cost Compiler
Turns an n8n or Make workflow export into a client-ready cost quote — per-node token estimates, cost per 1,000 runs, monthly projection and agency margin — so automation freelancers can price AI builds without guessing.
ClaimLeak Teardown
Turns a de-identified claims CSV into a white-label Revenue Leakage Audit PDF that independent medical billing companies use to win small-clinic clients.