idea-launcher
Use when Codex needs to turn a rough app, tool, game, automation, or product idea into a ready-to-build project plan, scaffold direction, implementation task breakdown, verification checklist, deployment workflow, or first build prompt. Trigger for idea shaping, MVP narrowing, project specification, starter structure generation, planning docs, pre-build interviews, product and technical tradeoff analysis, factory mode, sharpen-the-axe mode, and ship mode.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- idea-launcher
- description
- Use when Codex needs to turn a rough app, tool, game, automation, or product idea into a ready-to-build project plan, scaffold direction, implementation task breakdown, verification checklist, deployment workflow, or first build prompt. Trigger for idea shaping, MVP narrowing, project specification, starter structure generation, planning docs, pre-build interviews, product and technical tradeoff analysis, factory mode, sharpen-the-axe mode, and ship mode.
02 资源文件 42
- 参考文档agents/openai.yaml323 B
- 脚本assets/blueprints/automation/src/main.py1.2 KB
- 脚本assets/blueprints/automation/tests/test_main.py774 B
- 资产文件assets/blueprints/data-tool/data/fixtures/sample.csv40 B
- 脚本assets/blueprints/data-tool/src/main.py1.1 KB
- 脚本assets/blueprints/data-tool/tests/test_main.py368 B
- 资产文件assets/blueprints/game/index.html600 B
- 资产文件assets/blueprints/game/package.json272 B
- 脚本assets/blueprints/game/scripts/build.js281 B
- 脚本assets/blueprints/game/src/game.js1.4 KB
- 资产文件assets/blueprints/game/src/styles.css708 B
- 脚本assets/blueprints/game/tests/smoke-test.js431 B
- 资产文件assets/blueprints/web-app/index.html1.2 KB
- 资产文件assets/blueprints/web-app/package.json272 B
- 脚本assets/blueprints/web-app/scripts/build.js280 B
- 脚本assets/blueprints/web-app/src/main.js888 B
- 资产文件assets/blueprints/web-app/src/styles.css1.9 KB
- 脚本assets/blueprints/web-app/tests/smoke-test.js580 B
- 资产文件assets/starter-packs/automation.json2.9 KB
- 资产文件assets/starter-packs/data-tool.json2.8 KB
- 资产文件assets/starter-packs/game.json2.5 KB
- 资产文件assets/starter-packs/web-app.json2.7 KB
- 资产文件assets/templates/AGENTS-additions.md810 B
- 资产文件assets/templates/DEPLOY-PLAN.md1.0 KB
- 资产文件assets/templates/FIRST-BUILD-PROMPT.md1.0 KB
- 资产文件assets/templates/IMPLEMENTATION-CHECKLIST.md1.2 KB
- 资产文件assets/templates/SPEC.md1.4 KB
- 资产文件assets/templates/STARTER-STRUCTURE.md1.1 KB
- 资产文件assets/templates/TASKS.md1.6 KB
- 资产文件assets/templates/TEST-PLAN.md1.1 KB
- 参考文档package.json539 B
- 参考文档README.md2.7 KB
- 参考文档references/examples.md2.4 KB
- 参考文档references/frontend-design.md2.3 KB
- 参考文档references/full-build-workflow.md2.0 KB
- 参考文档references/interview-workflow.md3.0 KB
- 参考文档references/modes.md3.8 KB
- 参考文档references/project-shaping-rubric.md3.1 KB
- 参考文档references/subagents.md1.7 KB
- 脚本scripts/idea-launcher.mjs1.1 KB
- 脚本scripts/launch_project.py7.5 KB
- 脚本scripts/validate_placeholders.py2.2 KB
03 SKILL.md 正文
Idea Launcher
Turn vague ideas into practical project plans, starter folders, and execution prompts for a one-person software factory. Optimize for fast learning, narrow scope, good taste, verification, and deployability.
Operating Rules
- Prefer a concrete proposal with explicit assumptions over a long questionnaire.
- Ask at most 1-3 clarifying questions per pass. Ask only questions that change the plan materially.
- Always include both the good side and bad side before recommending a direction.
- Treat the first useful shipped version as the default target. Push nice-to-haves into later milestones.
- If the user asks to brainstorm, explore options, tradeoffs, and failure modes before converging.
- If the user asks to build, run the prompt-readiness triage first, shape only enough to prevent rework, then move into implementation.
- Respect local
AGENTS.mdandMEMORY.mdinstructions. If a durable preference or project fact is discovered, update the relevant memory file and mention the change. - Commit and push after major tasks when the active workspace instructions say to do so.
Prompt-Readiness Triage
Classify the user's idea before choosing brainstorm, planning, or build behavior.
Detailed-enough prompt: the prompt names the product or workflow, gives a target user or context, implies a platform or output, and has enough constraints to choose a first build without inventing the core product. Do not skip straight into code. First give a serious pass:
- Summary of what will be built.
- Assumptions you will use unless corrected.
- Pushbacks, risks, and likely scope traps.
- Suggestions that would improve the build without bloating v1.
- The smallest credible v1 and what is explicitly not in v1.
After that pass, proceed when the user already asked to build and no answer is needed to avoid wrong work.
Basic, half-formed, or unclear prompt: the prompt is mostly a theme, name, vibe, feature pile, or unclear request. Use brainstorm mode or the clarifying pattern from references/interview-workflow.md. Ask only the 1-3 questions that change direction, and offer a default assumption path so the user can say "go".
Explicit brainstorm prompt: expand options first, compare tradeoffs, then recommend one direction and one smaller version.
Core Workflow
- Capture the idea: Restate the concept in one sentence, name the target user, and infer the likely project type: app, tool, game, automation, integration, content workflow, data product, or experiment.
- Triage readiness: Decide whether the prompt is detailed enough, basic/unclear, or explicitly brainstorm-oriented.
- Interview lightly: Use
references/interview-workflow.mdto collect only the missing information needed to shape scope, risk, stack, and output. - Run the tradeoff pass: List the good side, bad side, hidden complexity, and the smallest credible version.
- Shape the project: Use
references/project-shaping-rubric.mdacross usefulness, scope, stack, data model, UI, workflows, automation, risks, verification, and deployment. - Select a mode: Use
references/modes.mdfor default launch, factory mode, sharpen-the-axe mode, brainstorm mode, or ship mode. - Produce artifacts: Create or update the needed planning files using
assets/templates/. - Prepare execution: Convert decisions into atomic tasks, verification gates, subagent splits where allowed, and a first build prompt.
- Generate a starter when useful: In factory mode, use
scripts/launch_project.pywith a matching starter pack to create a project folder, then runscripts/validate_placeholders.pyagainst the result. - Build the actual product: In full-build mode, generate the starter, inspect the generated docs and source, then implement the project-specific behavior, UI/design, tests, and deploy config directly in the generated project.
- Reinforce frontend quality: For web apps, tools, dashboards, and games, apply
references/frontend-design.mdbefore coding UI. Favor fast static starters and only add heavier frontend stacks when the workflow justifies them.
Output Contract
For a planning response, provide:
- Idea summary
- Good side
- Bad side
- Key assumptions
- Recommended MVP
- Project shape
- Open questions, capped at 3
- Next artifact or build step
For file generation, write only the artifacts that are useful for the current request. Start with:
SPEC.mdfromassets/templates/SPEC.mdTASKS.mdfromassets/templates/TASKS.mdAGENTS.mdadditions fromassets/templates/AGENTS-additions.md- Implementation checklist from
assets/templates/IMPLEMENTATION-CHECKLIST.md - Test plan from
assets/templates/TEST-PLAN.md - Deploy plan from
assets/templates/DEPLOY-PLAN.md - First build prompt from
assets/templates/FIRST-BUILD-PROMPT.md - Starter structure from
assets/templates/STARTER-STRUCTURE.md
When copying templates, replace placeholders, delete irrelevant sections, and keep open questions explicit instead of hiding uncertainty.
Factory Script
Use the factory script when the user wants a reusable starter structure or a generated project folder:
python scripts/launch_project.py --name "Project Name" --idea "One-line idea" --type web-app --out ./output
python scripts/validate_placeholders.py ./output/project-name
Starter pack choices:
web-appautomationgamedata-tool
Use --force only when intentionally regenerating files. Do not overwrite a user's existing project files without explicit approval.
Full-Build Mode
Use full-build mode when the user wants the actual app, game, automation, or data tool, not just specs. The factory script creates a runnable baseline; Codex must then edit the generated project until it matches the idea.
Full-build sequence:
- Generate the closest starter pack.
- Read the generated
docs/SPEC.md,docs/TASKS.md, and source files. - Replace demo behavior with the real core workflow.
- Shape the interface or output format to the domain. For web apps and games, load
references/frontend-design.mdand treat design as part of implementation, not a later polish task. - Add or update focused tests for the real behavior.
- Run local verification commands from
AGENTS.md. - Start the dev server or provide the runnable command/path.
Do not stop at generated docs when the user asked to build. The generated starter is the starting point, not the deliverable.
Subagents
Use subagents only when the current session and user instructions allow it, and when the split will materially save time. Prefer independent sidecar work: repo discovery, UI surface planning, data model review, risk review, test planning, deployment research, or isolated implementation slices. Do not delegate the immediate blocking decision.
For detailed split patterns and prompt shapes, use references/subagents.md.
Quality Gates
Before finishing a launch plan, confirm:
- The MVP is narrow enough for one person to build.
- Every major feature maps to a user workflow or explicit learning goal.
- The data model is named even if it is intentionally simple.
- Risks have mitigation or scope cuts.
- The test plan verifies the highest-risk behavior first.
- The deploy plan names the target environment, secrets, migration needs, and rollback path.
- The first build prompt is self-contained enough for a fresh Codex thread to execute.
- Generated starters have no unresolved double-brace template placeholders.
- Full-build outputs run locally and demonstrate the real core workflow.
- Frontend outputs are fast to run, domain-specific, responsive, and free of generic placeholder UI.
References
references/interview-workflow.md: Clarifying without over-questioning.references/project-shaping-rubric.md: Good/bad tradeoff and project shaping rubric.references/modes.md: Factory mode, sharpen-the-axe mode, ship mode, and default launch mode.references/full-build-workflow.md: How to turn a generated starter into a real runnable project.references/frontend-design.md: Fast, practical frontend design rules for generated apps and games.references/subagents.md: When and how to split work across agents.references/examples.md: Short examples for web apps, automations, and games.assets/templates/: Copy-ready planning and execution templates.assets/blueprints/: Runnable baseline projects copied by the factory script.assets/starter-packs/: JSON defaults for generated project starters.scripts/launch_project.py: Generate a starter project from templates and a starter pack.scripts/validate_placeholders.py: Check generated artifacts for unresolved placeholders.
04 同仓库更多 Skills
winui-app
Bootstrap, develop, and design modern WinUI 3 desktop applications with C# and the Windows App SDK using official Microsoft guidance, WinUI Gallery patterns, Windows App SDK samples, and CommunityToolkit components. Use when creating a brand new app, preparing a machine for WinUI, reviewing, refactoring, planning, troubleshooting, environment-checking, or setting up WinUI 3 XAML, controls, navigation, windowing, theming, accessibility, responsiveness, performance, deployment, or related Windows app design and development work.
vercel-deploy
Deploy applications and websites to Vercel. Use when the user requests deployment actions like "deploy my app", "deploy and give me the link", "push this live", or "create a preview deployment".
security-best-practices
Perform language and framework specific security best-practice reviews and suggest improvements. Trigger only when the user explicitly requests security best practices guidance, a security review/report, or secure-by-default coding help. Trigger only for supported languages (python, javascript/typescript, go). Do not trigger for general code review, debugging, or non-security tasks.
playwright
Use when the task requires automating a real browser from the terminal (navigation, form filling, snapshots, screenshots, data extraction, UI-flow debugging) via `playwright-cli` or the bundled wrapper script.
openai-docs
Use when the user asks how to build with OpenAI products or APIs and needs up-to-date official documentation with citations, help choosing the latest model for a use case, or model upgrade and prompt-upgrade guidance; prioritize OpenAI docs MCP tools, use bundled references only as helper context, and restrict any fallback browsing to official OpenAI domains.
netlify-deploy
Deploy web projects to Netlify using the Netlify CLI (`npx netlify`). Use when the user asks to deploy, host, publish, or link a site/repo on Netlify, including preview and production deploys.