Back to Skills

ce-ideate

Generate and evaluate grounded ideas. Use when the user asks for ideas, improvements, surprising options, or AI-generated directions before choosing one to develop; use ce-brainstorm to refine the user's own idea.

21,992stars1,623forksUpdated 6/25/2026

Security Assessment

Safe(100/100)
Security Score100/100

About ce-ideate

ce-ideate generates and critically evaluates grounded ideas about a topic, producing a ranked ideation artifact rather than requirements, plans, or code. It is the first stage of a compound-engineering workflow: ce-ideate answers "what are the strongest ideas worth exploring?", ce-brainstorm then defines a chosen idea precisely, and ce-plan determines how to build it. Typical triggers include requests like "what should I improve", "give me ideas", "ideate on X", or "surprise me" -- cases where the user wants AI-generated suggestions instead of refinement of an existing idea.

Its quality mechanism is explicit rejection with reasons: generate many candidates, critique all of them, then explain only the survivors. A core principle is to ground before ideating by scanning the actual codebase first, so output is tied to the repository rather than abstract product advice. An optional argument supplies context -- a concept, a path, a research artifact to draw on, a constraint such as "low-complexity quick wins", or a volume hint such as "top 3" -- and with no argument it proceeds open-ended. The skill asks questions through the platform's blocking question tool (AskUserQuestion in Claude Code, with equivalents in Codex, Gemini, and Pi), one at a time, falling back to numbered chat options only when no blocking tool exists.

Sub-agent dispatch is tiered by task shape rather than a hardcoded model name: an extraction tier (cheapest capable model) for evidence scouting, a generation tier (mid-tier) for evidence-driven ideation, and a ceiling tier (the orchestrator's model) for cross-cutting synthesis and final arbitration; surprise-me mode and a "go deep" override raise the whole fleet to the ceiling tier. Output format is resolved early and is exclusive -- the artifact is written as either HTML or markdown, with HTML the default since ideation docs are read mainly by humans weighing directions, and a pipeline override forcing markdown. The artifact is written to docs/ideation/ when present, else a temporary path.

FAQ

How does ce-ideate differ from ce-brainstorm and ce-plan?

ce-ideate identifies the strongest ideas worth exploring, ce-brainstorm defines one chosen idea precisely enough for planning, and ce-plan determines how it should be built. ce-ideate routes its action into brainstorming and does not skip to planning.

What output format does it produce?

It writes a single ranked ideation artifact as either HTML or markdown, never both. HTML is the default because the doc is read mainly by humans, but pipeline or disable-model-invocation contexts force markdown.

Does it need an argument to run?

No. An argument is optional context -- a concept, a path, a research artifact, a constraint, or a volume hint. With no argument it proceeds with open-ended ideation.

How does it keep ideas grounded rather than generic?

It scans the actual codebase first and grounds ideas in the repository. Its quality mechanism is to generate many candidates, critique all of them, and then explain only the survivors with reasons for rejection.

Where is the ideation artifact saved?

It is written to docs/ideation/ when that directory is present, otherwise to a compound-engineering temporary path. The skill does not produce requirements, plans, or code.

All Files

8 files
references/html-rendering.md31.4 KB
View
references/markdown-rendering.md7.9 KB
View
references/universal-ideation.md14.7 KB
View
SKILL.md47.5 KB
View
references/divergent-ideation.md13.2 KB
View
references/ideation-sections.md8.5 KB
View
references/post-ideation-workflow.md17.0 KB
View
references/web-research-cache.md4.0 KB
View

Install ce-ideate

Download and extract the skill files to your .claude/skills/ directory.

Quick Setup:

  1. Copy the skill folder to .claude/skills/
  2. Claude will automatically detect and use the skill