Guide · Agent skill
How to use the BeCreative Idea Generator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
BeCreative is an agent skill, a folder of instructions and a small script that an AI agent such as Claude can load on demand. Once it is installed, asking the agent for creative ideas makes it generate several unusual, distinct candidates instead of the most obvious answer, then compare them and recommend the strongest one. This guide explains what the skill does, how to install it, how to trigger it, what it needs, and where it is the wrong tool.
BeCreative Idea Generatordanielmiessler/LifeOS · MIT Download the skill Quick start
- 1Download the ZIP. Download the BeCreative skill ZIP from the download link on this page.
- 2Unzip into a skills folder. Unzip the folder into ~/.claude/skills/ for personal use, or into your project's .claude/skills/ directory for one project.
- 3Or upload in the Claude apps. If you use the Claude apps, go to Settings → Capabilities → Skills and upload the skill there.
- 4Install the Bun runtime. Make sure the Bun runtime is available, because the bundled Diversity.ts script needs it.
- 5Ask for creative ideas. Make a request with words like 'brainstorm' or 'radically different', and include your brief's constraints.
- 6Review the recommendation. Read the recommended candidate, the runner-up and the reason given, then judge the idea yourself.
- 7Check synthetic data. For dataset expansion, compare the diversity index of the output with the seed set using Tools/Diversity.ts index before using it.
What the BeCreative skill makes the agent do
Asking a model for a creative idea usually returns the most probable answer, which is safe and already common. BeCreative pushes past that using a technique called Verbalized Sampling: the agent generates several low-probability candidates that rest on different assumptions, then compares them against each other and names the strongest for your brief.
The skill has a second mode. It can take a small set of example texts (a seed corpus) and expand it into a more varied set of synthetic examples, for example test cases for an evaluation. A bundled script, Tools/Diversity.ts, measures how similar the examples are to each other so you can check that the expanded set is more diverse than the seed.
- Single-pass ideation: several distinct candidates, compared, one recommended.
- Synthetic data: grow a seed set into a more diverse corpus or test set.
- Ideas evolved over several rounds are not handled here; they route to the separate HillClimb Ideate skill.
Install the BeCreative skill
Download the BeCreative ZIP from the download link on this page. Then choose one of the two installation routes below. Use the personal folder if you want the skill available in all your projects, or a project folder if you want it limited to one codebase.
In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip the folder into ~/.claude/skills/ so the skill sits at ~/.claude/skills/BeCreative/.
- Project install: unzip the folder into your project's .claude/skills/ directory.
- Claude apps: go to Settings → Capabilities → Skills and upload the skill.
Trigger the skill with natural requests
You do not need a special command. The skill is triggered by the kinds of request you make. Phrases such as 'be creative', 'brainstorm', 'radically different', 'name this', 'creative angle', 'technical creativity' and 'novel ideas' route to single-pass creativity. Phrases such as 'expand this corpus', 'synthetic data', 'generate diverse examples' and 'create test set' route to the dataset mode.
The skill picks an intent based on your wording. Each intent has its own approach: StandardCreativity for a polished single answer, MaximumCreativity for the most radical option, IdeaGeneration for solving a problem, TreeOfThoughts for complex multi-factor problems, DomainSpecific for artistic or business work, TechnicalCreativity for algorithm or architecture ideas, and SyntheticDataExpansion for datasets.
- Be specific about the brief: audience, constraints and what a good result looks like.
- Say 'maximum creativity' or 'radically different' if the first safe answer is not enough.
- Say 'several rounds' or 'evolve these ideas' if you want iteration; that goes to HillClimb Ideate.
Examples of requests and what you get back
The examples below show the routing the skill is designed for. The result is a single recommended idea with the reason it beats the conventional answer, not a long list of options.
- 'Brainstorm products we could build from a patent portfolio' routes to IdeaGeneration and returns the strongest of five approaches built on different assumptions, with the reason it beats the conventional answer.
- 'Divergent scene concepts for a blog header illustration' routes to DomainSpecific and returns a concept that breaks a real editorial-illustration convention, justified against that domain's criteria.
- 'Evolve ideas for a new featured content collection over a few rounds' routes to HillClimb Ideate, not to this skill.
Build a more diverse synthetic dataset
Give the agent a small set of example texts and ask it to expand the corpus or create a test set from them. The skill is designed to return a corpus that is measurably more diverse than the seed set. The bundled script reports a diversity index, and you can compare the index of the expanded set with that of the seed.
The skill's usage reference for the script is in Tools/Diversity.help.md. The index command named in the skill is shown below; run it with the Bun runtime.
- Provide the seed examples in the request, one clearly separated example per item.
- Ask for the number of new examples you need.
- Compare the diversity index of the output with the seed before using it as a test set.
Tools/Diversity.ts indexRequirements, network use and safety notes
The skill requires the Bun runtime to execute the Diversity.ts script. It also runs scripts and uses the network, so review the bundled files before installing if that matters for your environment.
Two parts are optional. TechnicalCreativity can use a second vendor, which requires an OpenAI login as described in Setup.md. Cross-vendor generation also uses LIFEOS/TOOLS/Inference.ts. Without these, the single-pass modes still work with the agent you already use.
The skill itself warns that a model's self-reported probabilities are not evidence that an idea is novel. A human reviewer must judge the output.
- Runtime: Bun, for Tools/Diversity.ts.
- Network: yes.
- Scripts: yes, including Tools/Diversity.ts.
- Optional: an OpenAI login for the second-vendor step in TechnicalCreativity (see Setup.md).
- Optional: LIFEOS/TOOLS/Inference.ts for cross-vendor generation.
- Licence: MIT.
Tips for getting strong results
The skill works best where many answers are valid, such as names, campaign angles, product concepts and design alternatives. It adds little where there is one correct answer, such as arithmetic or a single bug fix. Judge the output yourself, because the candidates are compared to each other, not checked against a ground truth.
For technical work where a single model's ideas may share its defaults, the skill's own guidance is to use a second vendor. Taste-based briefs such as voice, naming and visual concepts suit the Anthropic lanes; correctness-based briefs such as algorithms and architectures suit the OpenAI lanes.
- Give the brief's constraints up front so the candidates are judged against them.
- Ask for the runner-up too, so you can see what was rejected and why.
- Do not treat a candidate as proven novel because the agent calls it novel.
- If you have your own customizations, they are read from <configRoot>/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/BeCreative/ when that directory exists.
What people use it for
- Brainstorming names, product concepts or campaign angles and getting more than the first obvious answer.
- Choosing between technical design alternatives that rest on different assumptions.
- Expanding a small set of example texts into a more varied set of synthetic examples.
- Building diverse test cases from a seed set for an evaluation.
- Generating illustration or creative concepts scoped to a specific domain.
Frequently asked questions
Is the BeCreative skill free?
The skill is released under the MIT licence, as stated on this page. Some optional parts, such as the second-vendor step for TechnicalCreativity, require an OpenAI login, so check Setup.md for what that involves.
What do I need to run BeCreative?
You need the Bun runtime to execute the bundled Diversity.ts script, and an agent that can load skills. The skill also uses the network and runs scripts. An OpenAI login and LIFEOS/TOOLS/Inference.ts are optional and only needed for cross-vendor steps.
Where do I install the BeCreative skill?
Unzip the downloaded folder into ~/.claude/skills/ for personal use or into a project's .claude/skills/ directory. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Why did the skill not give me several ideas?
The skill is designed for briefs with many valid answers. For a problem with one correct answer, such as arithmetic or a single bug fix, the skill's own notes say the technique adds nothing. Try rephrasing the request with the constraints of your brief.
How is BeCreative different from HillClimb Ideate?
BeCreative runs a single pass of divergent ideation and can expand a seed corpus. HillClimb Ideate handles ideas that evolve over several rounds, and the skill routes any multi-round or 'ideate', 'id8' or 'evolve ideas' request there.
Can I trust that the ideas are novel?
Not on the agent's word alone. The skill warns that model probabilities are not evidence of novelty, so a human reviewer needs to judge whether the candidates are distinct, on-brief and useful.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
BeCreative suits people who want more than the first obvious answer for names, concepts, campaign angles or technical alternatives, and developers who need diverse synthetic examples from a seed set. Pick something else if your problem has one correct answer, if you need ideas to evolve over several rounds (use HillClimb Ideate), or if you cannot review the output yourself.
Download the skill