URGENT: Claude Code expert: make my /design-loop impossible to skip
min $50 USD / hour
Posted 1 minute ago
•
Ends in 6 days
min $50 USD / hour
I make films for brands like Coke, Nike and LEGO with Claude Code. Every shot runs through one skill, the /design-loop: it builds the shot, fresh critic agents judge it, and it only reaches me once it passes. When it runs, it's magic. The trouble is that it gets skipped. Shots no critic has seen slip through, with extra hands and twisted bodies, and finished films slide back to old work that never went through the loop. Right now I have a LEGO film and a Nike film that are basically done, and every final fix reverts them. Two freelancers have already tried and failed to solve this. I need someone who can make the loop unbreakable, so nothing reaches me unless the critics passed it, and every new version starts from the last one I approved. Show me it working right now, on LEGO and Nike, and keep it working for two days. Do that and you have $2,000 a week, ongoing. Keep it going for three months and I'll double it. It only goes up from there. Paid trial or on spec, your call. Only apply if you know Claude Code inside out: skills, subagents and hooks.
Project ID: 40830644
About the project
15 proposals
Open for bidding
Remote project
Active 8 secs ago
Place your bid
Bid amount Email addressBenefits of bidding on Freelancer
Set your budget and timeframe
It's free to sign up and bid on jobs
15 freelancers are bidding on average $180 USD/hour for this job


Hi, I'll make your /design-loop an enforced quality gate rather than an optional step that Claude Code can bypass. I'll inspect your existing skills, subagents, hooks, and versioning workflow to identify where unreviewed shots enter the pipeline and why final fixes revert approved work. I'll implement a fail-closed validation process using Claude Code hooks, isolated critic agents, approval records, and version checks. Every shot must have a passing critic result tied to its exact version before delivery, and every revision must start from the latest approved version. I'll test the workflow against your existing LEGO and Nike projects, demonstrate both successful and blocked submissions, and monitor it over the two-day trial. I can work with your existing setup rather than rebuilding the entire pipeline. I am looking forward to hearing from you. Best regards, Pavle


Hello. The issue isn't the critics themselves. It's that /design-loop is optional, and the workflow can reuse unapproved shots or fall back to older versions. I'd enforce approval through Claude Code hooks, keep critic subagents isolated, and track approved shot versions with immutable manifests. Every export would require a valid critic pass, while failed checks would block delivery. I'd verify this against the current LEGO and Nike projects, including deliberate bypass and rollback attempts. Q1: Is /design-loop currently enforced by hooks or only skill instructions? Q2: Where are approved shot versions and critic results stored? I'd start with the existing workflow rather than rebuild it. Happy to demonstrate the safeguards on both films.


Your Claude Code workflow has a governance problem, not a prompting problem. The /design-loop isn't being enforced as the only path to production, so unvalidated shots and stale versions are slipping through. That means putting guardrails in the workflow itself, not trusting the process to be followed. I'd start by auditing how shots currently enter your pipeline, where the critic step sits, and exactly how that LEGO and Nike revision path is reverting you to pre-loop work. With the film assets and current skill definitions, I can trace the break, add hooks that block unvalidated output before it reaches you, and pin version starts to your last approved state. I've built and hardened multi-agent Claude Code pipelines for GenPark and Applied Labs, both with enforced validation gates, which is the same pattern your system needs. What's the first concrete proof you want, the hook firing to block a skip, or a completed LEGO and Nike run that survives repeated fixes?


Hello, I’m Kristián Lauko, with 10+ years of experience in AI Automation, AI Development, AI Integration, and advanced workflow engineering. I understand the issue: your /design-loop is being bypassed, allowing unreviewed shot versions and regressions to slip into final films. I’d harden the Claude Code flow by tightening skills, subagents, and hooks so every shot must pass the critic checks before reaching you, while also ensuring each new version is built from the last approved state. I have extensive experience building reliable AI-driven pipelines, guardrails, and agent workflows where failures must be caught before output moves forward. I’d ensure clear communication throughout the project and would be glad to jump in and prove the fix on your LEGO and Nike films. Looking forward to hearing from you. Thanks, Kristián Lauko


Hello, This is exactly the kind of Claude Code workflow hardening I enjoy working on. I’ve spent a lot of time building AI automation and AI integration systems where the main challenge isn’t “making it work once,” but making sure it cannot quietly break later. Your /design-loop issue makes sense: if critic approval is not enforced at the right layer, or if state/versioning can fall back to an earlier branch, skipped shots and reverts will keep happening. I’d approach this by tracing the full Claude Code flow end-to-end, then tightening the skill, subagent, and hook logic so every shot must pass the critic gate before it can move forward. I recently worked on a similar AI automation setup where generated outputs were occasionally bypassing validation steps; I fixed it by making the approval layer state-aware and forcing every new run to inherit only from the latest approved version. A couple of quick questions: 1) Are your critics enforced through hooks, subagents, or both? 2) Do you already have version checkpoints for LEGO and Nike, or should I add a protected approval state? If you’d like, I can jump on a call and walk through the exact failure points with you. Best regards, Istkhar


Hello, Your /design-loop works when it actually runs, but right now nothing forces a shot through the critics, and nothing ties the next edit to the last version you already approved. That is why LEGO and Nike keep sliding back into weird hands, twisted bodies, and old frames that never saw a judge. I would turn the loop into a hard gate. With Claude Code hooks on the skill path, a finish attempt only goes through if a fresh critic pass is on record. No pass, it never reaches you. I would also pin each approved shot as the only legal starting point for the next round, so a final fix cannot quietly reopen unjudged work. For the trial I would wire this on LEGO and Nike first, show you a live run where a bad shot gets blocked and a clean one comes through, then leave it running for two days so you can stress it. I can put in 40 hours each week while we lock it down. If it holds, we keep going from there. Paid trial or on spec is fine by me.
内容来自 Freelancer News,版权归原作者所有。
阅读原文 ↗