From dd7d45175a57efb3293151c2090a9dd0e89ac8f3 Mon Sep 17 00:00:00 2001 From: Zeke Sikelianos Date: Wed, 29 Apr 2026 08:59:11 -0700 Subject: [PATCH] docs(skills): remove model names, point agents at the API Strip specific model names from the find-models, prompt-images, and prompt-videos skills so agents don't anchor on names from training data or past conversations. Add directives at the top of each skill telling agents to query the Replicate API for current best models. Build and publish skills are unchanged because their model references point at real production cog repos used as build patterns, not running model recommendations. --- skills/find-models/SKILL.md | 8 +++++++- skills/prompt-images/SKILL.md | 12 +++++++++--- skills/prompt-videos/SKILL.md | 12 +++++++++--- 3 files changed, 25 insertions(+), 7 deletions(-) diff --git a/skills/find-models/SKILL.md b/skills/find-models/SKILL.md index fc6b79b..3071d11 100644 --- a/skills/find-models/SKILL.md +++ b/skills/find-models/SKILL.md @@ -3,6 +3,12 @@ name: find-models description: Find AI models on Replicate using search and curated collections. --- +## Always search the API for current models + +The AI model landscape changes weekly. New models ship constantly and older ones are deprecated or surpassed. Don't rely on model names you've seen before, including names from past conversations or training data. A specific model you "know" may no longer be the best choice, may be slower than newer alternatives, or may not exist anymore. + +Always start by querying the Replicate API. Use search and collections to discover what's currently available, then read schemas to understand inputs and outputs before running anything. + ## Docs - Reference: @@ -42,6 +48,6 @@ description: Find AI models on Replicate using search and curated collections. ## Model identifiers -- **Official models** use `owner/name` format (e.g. `black-forest-labs/flux-2-klein-9b`). Routes to the latest version automatically. +- **Official models** use `owner/name` format (e.g. `owner/model-name`). Routes to the latest version automatically. - **Community models** require `owner/name:version_id`. You must pin a specific version. Community models can cold-boot and take time to start. - If you must use a community model, be aware that it can take a long time to boot. You can create always-on deployments, but you pay for model uptime. diff --git a/skills/prompt-images/SKILL.md b/skills/prompt-images/SKILL.md index 3ba3723..4bc8893 100644 --- a/skills/prompt-images/SKILL.md +++ b/skills/prompt-images/SKILL.md @@ -7,7 +7,13 @@ description: > # Prompting image models on Replicate -Distilled from Replicate's blog posts on prompting image models (2024-2026). Techniques are model-agnostic and focus on transferable principles. For model selection, pricing, and feature comparison, see the [compare-models](../compare-models/SKILL.md) skill. +Distilled from Replicate's blog posts on prompting image models (2024-2026). Techniques are model-agnostic and focus on transferable principles. + +## Choose a model with the API, not from memory + +This skill describes general prompting techniques. To choose a model, use the [find-models](../find-models/SKILL.md) skill and query the Replicate API. The image model landscape changes weekly. Don't assume specific models exist or are still state-of-the-art based on names you've seen before. Always search the API for current options, then read the schema before running anything. + +For pricing and feature comparison, see the [compare-models](../compare-models/SKILL.md) skill. ## Writing prompts @@ -180,9 +186,9 @@ Some models can generate multiple related images in a single prompt. 10. **Not iterating**: The best results come from iterative workflows. Make a small change, evaluate, refine, repeat. Don't try to get everything right in a single generation. -## Sources +## Background reading -All techniques in this skill are sourced from Replicate's blog: +The techniques above are distilled from Replicate's blog posts on prompting image models. The posts are anchored to specific models that were current at the time, but the techniques generalize across modern image models. Use these for additional context, then use the [find-models](../find-models/SKILL.md) skill to pick a model that fits your task today. - [How to prompt Seedream 5.0](https://replicate.com/blog/how-to-prompt-seedream-5) (Feb 2026) - [Recraft V4](https://replicate.com/blog/recraft-v4) (Feb 2026) diff --git a/skills/prompt-videos/SKILL.md b/skills/prompt-videos/SKILL.md index b7f64e2..eff0d50 100644 --- a/skills/prompt-videos/SKILL.md +++ b/skills/prompt-videos/SKILL.md @@ -8,7 +8,13 @@ description: > # Prompting video models on Replicate -Distilled from Replicate's blog posts on prompting video models (2025-2026). Techniques are model-agnostic and focus on transferable principles. For model selection, pricing, and feature comparison, see the [compare-models](../compare-models/SKILL.md) skill. +Distilled from Replicate's blog posts on prompting video models (2025-2026). Techniques are model-agnostic and focus on transferable principles. + +## Choose a model with the API, not from memory + +This skill describes general prompting techniques. To choose a model, use the [find-models](../find-models/SKILL.md) skill and query the Replicate API. The video model landscape changes weekly. Don't assume specific models exist or are still state-of-the-art based on names you've seen before. Always search the API for current options, then read the schema before running anything. + +For pricing and feature comparison, see the [compare-models](../compare-models/SKILL.md) skill. ## Scene description @@ -322,9 +328,9 @@ If the model supports subject reference images, use a clear photo of the charact 10. **Forgetting that video models don't have internet access**: No video model has live information. They work from training data. Don't expect them to know about current events or real-time information. -## Sources +## Background reading -All techniques in this skill are sourced from Replicate's blog: +The techniques above are distilled from Replicate's blog posts on prompting video models. The posts are anchored to specific models that were current at the time, but the techniques generalize across modern video models. Use these for additional context, then use the [find-models](../find-models/SKILL.md) skill to pick a model that fits your task today. - [How to make remarkable videos with Seedance 2.0](https://replicate.com/blog/seedance-2) (Apr 2026) - [How to prompt Veo 3.1](https://replicate.com/blog/veo-3-1) (Oct 2025)