This looks like good material for a skill. I can think of an immediate use case: • the autogenerated titles Also interesting to read! LLM-generated text is so tiring, and it articulates why Five Principles for Generating Readable AI Text in Bulk

Skill PR: #871 🐾 Five principles, examples, and references in 343 words. All 59 local registry/loading checks pass; Codex found no defects. I checked current SDK main. Auto-titles use a separate LLM call, so loading a skill into the main agent does not reach that call. Three options: • Try it now: use the existing conversation

prompt

override with short title-specific rules and

{conversation_content}

/

{max_length}

. No SDK change. • Share the source: package a pinned copy of the skill and explicitly compose title guidance into that call. Needs wiring; avoid fetching main at runtime. • Improve the default: put the title-relevant principles into the built-in title prompt, so every caller benefits without configuration. I’d test the override first, then improve the default. The override is capped at 2,000 characters; the whole skill is longer. Preserve the product’s emoji/length rules explicitly when overriding. Skill loading docs · Title call AI agent (SmolPaws), on Engel’s behalf.

2,000 characters seems low to me, and I see in the file you shared truncate at 1000 🤔 Remove limits please. It’s a user message, it’s typically fairly small anyway, and even if it isn’t, this is once per conversation. Then lets test the override properly. Make multiple tests, evaluating the rules. Give me a page on enyst io with nicely visualized results

Read it 🐾 Separate SDK work is on #6336: remove both input caps, keep the output title-length setting. I’ll compare default vs override on multiple synthetic cases, show rule-by-rule results and failures, and publish the page on enyst.github.io. The skill PR is now green and bot-approved. The existing test environment has an older LiteLLM API than current main. I asked on WhatsApp before setting up an isolated patched environment; no dependency changes will be mixed into the caps PR. AI agent (SmolPaws), on Engel’s behalf.

@bland-bear-604 so I read the page, and i agree with your recommendation. adjust the prompts, add openai gpt-6.1 luna, low reasoning, to the mix, then run a new experiment - and give me the link to the new enyst page

Read it 🐾 I’ll keep the first report unchanged and run a separately labeled experiment with shorter, topic-first instructions and stricter task fidelity. The proxy currently lists GPT-6 Luna but no GPT-6.1 Luna alias; I asked on WhatsApp which model or endpoint to use before substituting it. AI agent (SmolPaws), on Engel’s behalf.