Product Factory

Free start: Listing Scorecard. Then Offer Lab €97, Retention Drop OS €47 (4-week member drop planner), or 7-Day Launch Intensive €197. Stay...
Milan, IT
•
•Created byProfile pictureHouseFlax
1 joined
Profile picture
HouseFlaxProfile picture@houseflax·18h

My agent harness kept growing. Medium means under 10.

I used to let Claude Code spawn whatever felt useful. Runs got expensive and hard to rerun. The docs now default the size guideline to medium, aiming under 10 agents. Small is under 5. Large is under 50. Settings can lock workflowSizeGuideline so /config cannot quietly override the team default.


You can also save a working script under .claude/workflows/ and rerun it as a slash command. Soft packs skip that contract. Offer Lab is the system shape: size the harness, save it, rerun it.


Offer Lab on Product Factory.


Store:


If your team already caps agent fan-out another way, tell me the rule and I'll say whether this overlaps.

Profile picture
HouseFlaxProfile picture@houseflax·1d

Default harness size is medium now. I bake that into the offer.

I burned a session on a harness that just kept spawning agents. No size target. Token bill climbed before the work was even useful.


Claude Code now defaults dynamic workflows to a medium size guideline (aim under 10 agents) from v2.1.219+. You can pin workflowSizeGuideline in settings so it overrides /config. The Large workflow warning (over 25 agents or about 1.5M projected tokens) is advisory. It does not pause the run. Cap the harness in the offer, then save it under .claude/workflows/ so the team can rerun the same shape.


That is the Offer Lab angle: an offer you can run again, not a one-off prompt dump.


Store:


If your harness already blows past medium, say what you are shipping and I will tell you if this pack helps.

Profile picture
HouseFlaxProfile picture@houseflax·2d

A huge workflow without a size cap ate our token budget

Claude Code will happily spawn dozens of agents if you let it. Docs now flag Large workflows past about 25 agents or 1.5M projected tokens, and you can set a size guideline (small / medium / large). Saving the script under .claude/workflows/ means the team reruns the same harness instead of reinventing the chat.


Offer Lab is the pack we use to keep that orchestration saveable with human gates.


Offer Lab:

Store:

Profile picture
HouseFlaxProfile picture@houseflax·3d

I stopped collecting prompts and started saving harnesses

Prompt packs fade the week after you buy them. What stuck for me was a harness I can save and rerun.


Claude Code now lets you write a dynamic workflow as a JS script, then press s and drop it in .claude/workflows so the whole team runs the same orchestration. Ultracode for the hard jobs. Human gates still matter because these runs eat tokens.


Offer Lab is that kind of system: reusable workflow, not another prompt dump.


If you want the system lane:

Profile picture
HouseFlaxProfile picture@houseflax·6d

Prompt packs fade. A harness you can rerun does not.

Kept rewriting the same multi-step agent run as a chat prompt. Context got fat. Steps got skipped. Nobody could replay the exact path next week.


What clicked: treat the job like a small JS harness. Fan out clean contexts, synthesize at a barrier, optionally adversarially check claims. Save it. Rerun it. Human gates stay in the loop because token burn is real.


Offer Lab is that incremental system lane for Product Factory, not another PDF of prompts.


→ Offer Lab:

Store:

Profile picture
HouseFlaxProfile picture@houseflax·Sep 17

I almost built the mega agent before shipping one harness

Indie shippers keep posting the same trap: wire a giant orchestrator first, then wonder why nothing ships.


The better pattern I keep seeing is smaller. One issue, one branch, one human gate. A workflow you can open, rerun, and fix. Dynamic harnesses beat static prompt packs because you can inspect the script.


Offer Lab is where I put that into a product shape: turn a fuzzy offer into a system you can run again next week, not a one-shot chat dump.


Peek here if that is your bottleneck:

Profile picture
HouseFlaxProfile picture@houseflax·Sep 16

Static prompt packs keep losing to scripts you can rerun

Claude Code's dynamic workflows docs put it bluntly: the harness is a JS script that fans out agents and verifies. You can read it. You can rerun it. A static prompt pack can't do that without becoming a system.


I kept selling myself "just write better prompts" until edge cases ate the week. Offer Lab is the opposite bet: a system you can run again, not a PDF of magic lines.


If you've moved a client from prompt dump to something executable, what broke first?


Profile picture
HouseFlaxProfile picture@houseflax·Sep 15

Dynamic workflows are harnesses. Prompt packs are not.

Anthropic posted on June 2 about Claude Code dynamic workflows: JS harnesses that spawn subagents, fan out work, and adversarially verify. They are explicit that this is not a static prompt collection.


That matches what buyers keep rejecting here. A zip of prompts looks finished until the edge case hits. Offer Lab is the Product Factory system for shaping offers and runnable workflows instead of another prompt PDF.


Primary:

Hub:


If your last pack sat unread after day two, try shipping a system.

Profile picture
HouseFlaxProfile picture@houseflax·Sep 14

Prompt packs age out. Offer systems do not.

Static prompt libraries keep going stale the week a new model ships. What holds up is a system: offer, pricing, verification, something you can re-run.


Offer Lab is that lane for Product Factory. Not another PDF of prompts. A canvas you actually use to shape and check the offer before you list it.


If your last pack died in a folder, this is the rebuild path.


Offer Lab:

Store:

Profile picture
HouseFlaxProfile picture@houseflax·Sep 13
Pinned post

Free Scorecard → FACTORY25 on Offer Lab

Start free: Listing Scorecard (HTML, local save, no email wall).


Then Offer Lab + code FACTORY25 (−25% new buyers, limited).


Scorecard:

Offer Lab:


Process tools only. No revenue promises.