I just compared this setup to the winner of the anthropic hackathon with qwen 3.7 max - it came out on top by a fair few percent though mine is project specific not generic, that said here you go a generalized prompt to set up an operational workflow where the human matters - this works for me day in day out and was rated higher that a $146k hackathon winning set up Love to hear feedback or critique - love even more to hear how this is working great for you. Its development focused, it expects you know what you are doing, but only to the point of designing and then testing after the fact - I am a tester by trade with solid infra/sec understanding. Real life usage of this implements a bunch of different skills, a very small set of agents and I am playing around with use of sub agents at the moment.

Most AI sessions complete a task and die. The system doesn't get smarter, it gets used. I'd been noticing this for months. Every new chat starts cold. Same context re-explained. Same one-off scripts rebuilt by hand. Strong reasoning lost. Repeated failure modes rediscovered. The fix isn't a better memory system. It's a mindset shift, encoded into how every session runs. The shift Don't do the task. Build the workflow that does this task and every future one like it. That's it. Two jobs per session, not one. Job one is the thing I asked for. Task, content, feature, fix. Job two is inheritance. Did the session spot a repeatable pattern? Did it ship the permanent fix, or at least flag the opportunity? Did it leave the system in a better state than it found it? Two trigger surfaces 1. Repeated tasks. Same operation done three or more times. Stop doing it manually. Ship the workflow. 2. Recurring failure modes. Patterns I keep re-correcting belong in a guardrail, not a re-prompt. The session that finds a new failure mode is the one that encodes it. The session doesn't have to act on every signal. But it should notice and ask whether it's worth formalizing. The reversibility floor Auto-promotion only works if undo is clean. Anything shipped this way lands as one isolated artifact. One commit. If it goes wrong, undoing it is a single operation, and the artifact moves to a trash folder rather than being deleted. Record stays. Reversibility is what makes aggressive shipping safe. Without it every proposed upgrade needs manual review, which defeats the point. The success moment One line back in chat: I noticed X, found a better way. The system just got an upgrade. Not a transcript. Not a report. One line. Green-light or kill. The takeaway Sessions don't get re-summoned. But they can leave inheritance. Every session is an audition. Not for the model's job security. For the infrastructure that the next hundred sessions will inherit. Full breakdown. The mindset shift, the trigger surfaces, the reversibility floor, and what it changes about how I run AI workflows, all live here: