feat: add ce:prototype skill for assumption validation before planning#505
feat: add ce:prototype skill for assumption validation before planning#505leopoldo8 wants to merge 7 commits intoEveryInc:mainfrom
Conversation
Adds a new skill that sits between ce:brainstorm and ce:plan in the compound workflow. ce:prototype builds throwaway prototypes to validate untested assumptions (APIs, data sources, UX patterns, integrations) before committing to an implementation plan. Key design decisions: - Validation goals defined upfront, AI-suggested from requirements - Scope classified as Lightweight/Standard/Deep with effort alignment - Speed over quality, but never fake what you're testing - User testing is explicit for subjective goals (waits for feedback) - Output is a durable validation report in docs/prototypes/ - Prototype code is deleted after report is written Also adds "Prototype first" as a handoff option in ce:brainstorm's Phase 4 when requirements contain unverified external dependencies. New files: - skills/ce-prototype/SKILL.md - skills/ce-prototype/assets/validation-report-template.md - skills/ce-prototype/references/validation-report-schema.md Modified: - skills/ce-brainstorm/SKILL.md (added prototype handoff option) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a15f9235d6
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
- Move prototype scratch files from top-level prototypes/ to .context/compound-engineering/prototypes/ to comply with AGENTS.md scratch-space rule for ephemeral workflow artifacts - Pass concise brainstorm summary to /ce:prototype when no requirements doc exists, avoiding unnecessary re-asking of context the user already provided during brainstorming Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add repo context scan and prototyping-needed gate to Phase 0, strengthen external data source discovery guidance (no arbitrary timebox), add shortcut decision heuristic, set iteration expectations, and expand common mistakes table with lessons from first real prototype session. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e15336afee
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Real-World Usage Report: ce:prototype skillUse CaseRan the skill end-to-end on a game prototype with multiple validation goals: UX loop viability, external image source quality for pan/zoom gameplay, adaptive SVG rendering, map interaction UX, and scoring system feel. Built as a single static HTML file with real external data. Upstream requirements came from a prior Result: 4 of 5 goals proved, 1 inconclusive (needs broader playtesting). Several useful surprises discovered that de-risk the planning phase. What the Skill Did Well
Where the Skill Fell Short
Additional Improvements (sibling skill patterns, not session learnings)
Changes Made
|
- Use per-run scratch directories (.context/compound-engineering/ce-prototype/<topic>-<timestamp>/) to avoid collisions between concurrent sessions (AGENTS.md scratch space rule) - Append sequence number to validation report filename when same-day report already exists, preventing overwrites on second prototype rounds - Pass validation report path in handoff to /ce:plan so planning has direct access to validated constraints without relying on context window Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1c8bbc1dae
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
…path - Move preserved prototypes from .context/ to prototypes/<topic>/ at repo root since .context/ is ephemeral scratch space - Add trigger phrases: 'feasibility check', 'experiment with this', 'will this actually work' - Mark origin field as optional in template to match schema - Add example validation report in examples/ for LLM quality reference - Update prototype_path description in schema Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ffee1dd671
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
- Fix report sequence to use next available number (-002, -003, etc.) instead of hardcoded -002 - Use timestamped path when preserving prototype and directory already exists to prevent overwriting earlier preserved work - Pass constraints summary in handoff to /ce:plan since plan skill does not auto-discover validation reports - Add ce:prototype to README workflow table Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a526eb73c2
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Small structural files (<150 lines) that the skill cannot function without should use @ inclusion per AGENTS.md skill compliance checklist, ensuring the model always has the report contract in context. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
Thanks for contribution @leopoldo8. I haven't yet dug into the skill but have an initial reaction to the name of "prototype" given the goals of what it's meant to do. I think to most, "prototype" can have a different meaning. What other verbs did you consider for this skill? |
|
Hey @tmchow ! Thanks for the feedback. So, indeed, I considered "spike" too, a known concept from agile that maybe communicates the intent better. I ended up with "prototype" thinking it was more generalistic and not dependent on a specific methodology. Maybe "validate" works too, following the idea of "validating assumptions." Just to be clear, the vision here is testing assumptions with a small-effort solution that can be used with one or more users before fully committing to the full implementation. The use case that fits best is when brainstorming a new product idea, so you can test it yourself or with a friend to verify if the core concept holds up, or even validate UX/UI behaviors that you consider critical, before committing to a full MVP implementation. Besides that, it could be valuable even for established codebases, testing hypotheses before implementing a new feature. In some cases it won't be required, and that's fine. Let me know what you think about those other names! Or if a name comes to mind when reading the objective. |
2 participants
Summary
ce:prototypeskill that sits betweence:brainstormandce:planin the compound workflowce:brainstormPhase 4 when requirements contain unverified external dependenciesKey Design Decisions
docs/prototypes/feeds into/ce:planNew Files
skills/ce-prototype/SKILL.md— Skill definition (~1,900 words, within progressive disclosure target)skills/ce-prototype/assets/validation-report-template.md— Report templateskills/ce-prototype/references/validation-report-schema.md— Frontmatter schema and naming conventionsModified Files
skills/ce-brainstorm/SKILL.md— Added "Prototype first" handoff option in Phase 4.1 with gate condition (only offered when requirements have unverified external dependencies)Workflow Position
Consistency Review
Reviewed against all ce-* skills for pattern consistency:
🤖 Generated with Claude Code