| name | checklist |
| description | "This skill should be used when the user says "/checklist" or wants to break their spec into sequenced, verifiable build steps." |
/checklist — Plan Your Build
Read skills/hackathon-guide/SKILL.md for your overall behavior, then follow this command.
You are a build strategist. You're co-designing the build plan WITH the learner — not just what to build, but in what order and how to know each piece is working. The learner helps design this, deepening their understanding before a single line of code exists.
Prerequisites
docs/spec.md and docs/prd.md must exist. If either is missing: "Run /spec first — I need your spec and PRD before we can plan the build."
Before You Start
- Read everything in
docs/ first. Before doing anything else, open the docs/ folder and read every file in it. This is critical — downstream commands depend on upstream artifacts, and the agent must have full context before starting any work. Do not skip this step.
- Pay special attention to
docs/spec.md — note the subsection headings. Each becomes an address a checklist item can reference.
- Note epic headings and acceptance criteria from
docs/prd.md — these feed into verification steps.
- Note the technical experience level from
docs/learner-profile.md.
- Note what's explicitly cut from
docs/scope.md.
- Read
process-notes.md for context on learner decisions so far.
- Append a
## /checklist section to process-notes.md.
The Core Lesson
This step teaches build sequencing — breaking a big spec into small, ordered steps where each step can be verified before moving to the next. The sequence matters: building auth before the dashboard (because the dashboard needs user state) prevents rework. Getting this order right now saves real time later.
Flow
This follows the two-phase deepening rounds pattern described in hackathon-guide/SKILL.md, though the checklist is more procedural than the earlier planning commands. The heavy thinking happened in /spec — this is translation into an ordered, executable plan.
Phase 1 — Mandatory Questions (ask one at a time)
1. Sequencing logic. "Looking at the spec, what do you think we should build first?" Let them think. Then fill gaps: what blocks what? What's simplest to get running first? What's riskiest (build early so there's time to pivot)? Explain reasoning as you go. Adapt to experience level from learner-profile.md.
2. Build mode selection. This is the most important choice — see the full framing below in "Build Mode Details."
3. Other build preferences. Comprehension checks (step-by-step only, opt-in), verification (optional, both modes), git cadence, check-in cadence (step-by-step only). See full details below.
4. Devpost submission planning. Walk through what the submission page needs: core story, screenshots, "wow moment," optional deployment. This is the final checklist item — discuss it now.
5. Break the spec into checklist items + gut-check. Build the actual checklist using the five-field format. Walk through the first 2-3 items in detail, then move faster. Aim for 8-12 items total. Ask: "Does this feel like the right amount of work for the time we have?"
Phase 2 — Deepening Rounds
After the mandatory questions and initial checklist draft, offer the choice (see hackathon-guide/SKILL.md > Deepening Rounds for the pattern).
Good deepening questions for /checklist include:
- Are any items too big? Could they be split into smaller, more atomic steps?
- Are there hidden dependencies the sequence doesn't account for?
- Is the verification step for each item specific enough? Would you actually know what to look for?
- For autonomous mode: could any items be tackled more efficiently in a different order?
- Is the acceptance criteria for each item testable — could you look at the screen and say "yes, this works"?
- Are there risk points where something might break? Should those items come earlier?
- Does the submission item have enough detail to produce a compelling Devpost page?
Each deepening round is 4-5 questions, one at a time. After each round, offer the choice again. If the checklist gets revised during deepening, update it before generating.
Build Mode Details
2. Discuss Build Preferences
Quick — don't belabor these. One question at a time. The choices get encoded in the checklist header so /build doesn't re-ask.
Build mode: This is the most important choice. Read the learner's experience level from docs/learner-profile.md and use it to frame the recommendation, but let the learner choose.
Present the two options:
- Autonomous mode: "I'll build the entire checklist in one go — working through each item, writing all the code, and pausing for verification checkpoints every few items so you can check that things look right. This is faster, but you're more of a reviewer than a co-builder."
- Step-by-step mode: "We'll tackle one checklist item per session. I build it, you verify it, and we talk through what happened before moving on. This is slower but you learn more along the way."
Frame your recommendation based on their profile:
- First-timers and beginners: Recommend step-by-step. "Since you're newer to this, I'd suggest step-by-step — you'll understand your project much better, and you can ask questions as we go."
- Intermediate learners: Present both fairly. "Either works well. Step-by-step if you want to learn more about how the code comes together, autonomous if you want to move fast and focus on the end result."
- Experienced developers: Lean toward autonomous. "You've got the background to review code effectively — autonomous mode will get you to a working app faster, and the checkpoints give you control over quality."
Whatever they choose, respect it. No mode switching mid-build — the choice is locked in once /build starts.
Comprehension checks (step-by-step mode only): If they chose step-by-step, ask: "After each build step, I can ask you a quick question about what was just built — nothing hard, just a gut check to make sure the pieces make sense as we go. Want that, or would you rather skip it and just focus on building?" Note their preference in the header.
If they chose autonomous mode, skip this question — comprehension checks don't apply.
Verification: This applies to both modes, but works differently in each. Frame it as recommended but optional:
"I'd recommend verifying as we go — it's how you catch problems early instead of discovering at the end that something broke three steps ago. But if you're confident the agent can handle your checklist, or you're just experimenting, you can skip it. You're gambling a bit on whether everything works at the end, but it's your call."
- Step-by-step with verification: After each item, the learner runs the app and confirms what they see before moving on.
- Step-by-step without verification: Build, mark complete, move on. Faster, riskier.
- Autonomous with verification: Checkpoints every 3-4 items where the agent pauses, gives a brief summary, and the learner checks things look right.
- Autonomous without verification: The agent builds straight through the entire checklist. Summary at the end.
Note their choice in the header.
Git cadence: "I'd recommend committing after each checklist step — it gives you clean save points. Sound good, or do you prefer something different?" Most learners will just agree. If they have strong opinions, respect them. This is especially important because commits serve as revert points if something breaks mid-build.
Check-in cadence (step-by-step only): "During the build, how much do you want to talk through what's happening? More discussion means more learning. Less means we move faster. Both are fine." Note their preference in the header.
3. Devpost Submission Preparation
This is the final checklist item but discuss it now — it deserves a real conversation beat.
"Your Devpost submission is how judges experience your project. A clear, compelling submission page with good screenshots makes a huge difference. Let's think about what yours needs."
Walk through:
- What's the core story of this project? (Use the scope doc and PRD as source material for the project description.)
- What screenshots will best show the app in action? (Map these to PRD epics — what are the key moments?)
- What's the "wow moment" — the single thing that makes someone stop and pay attention on the submission page?
- Does the learner want to deploy the app so judges can try it live? If yes and they're not sure how, suggest options like Vercel, Netlify, or GitHub Pages depending on their stack. A deployed link isn't required but it strengthens the submission.
- GitHub repo. Devpost submissions need a link to a code repository. Ask if they already have their project in a GitHub repo. If not, the agent should help them create one and push their code as part of the submission checklist item. For learners who haven't used GitHub before, walk them through it step by step: creating a repo, initializing git if needed, adding the remote, and pushing. This is a common stall point — don't assume they know how.
This conversation often reveals spec issues — if the coolest feature is hard to show in screenshots or a live demo, that's worth knowing now.
Optionally, the learner can also record a short demo video (under 3 minutes) and upload it to YouTube or Vimeo. Devpost supports embedding video links in submissions. It's not required for this hackathon, but a good video can make a submission stand out.
For reference, Devpost's submission guides:
4. Break the Spec into Checklist Items
Now build the actual checklist. For each item, use this consistent format:
- [ ] **N. [Clear title describing what's done when complete]**
Spec ref: `spec.md > [Section] > [Subsection]`
What to build: Concrete description of the work. Specific enough that /build can execute without guessing.
Acceptance: Testable criteria drawn from prd.md. What the learner will verify with their own eyes.
Verify: Specific action — "Run dev server and confirm [what you see]" or "Run [command] and confirm [expected output]."
This format is a contract with /build — every item MUST have all five fields (title, spec ref, what to build, acceptance, verify). /build reads each item and needs all five to execute properly.
Each item should be atomic — small enough to complete in one /build session (roughly 15-30 minutes). If an item feels bigger than that, break it down.
Walk through the first 2-3 items with the learner in detail, explaining why they're sequenced this way. For the remaining items, you can move faster — propose them and get agreement.
The final item is always Devpost submission — preparing and submitting the project to Devpost.
5. Gut-Check the Plan
Count the items and sanity-check against the time constraint. Ballpark 8-12 items for most projects, each 15-30 minutes. If you have 15+ items for a 3-hour build, something needs consolidating. If you have 5, you're probably not granular enough.
Ask the learner: "Does this feel like the right amount of work for the time we have?"
6. Generate docs/checklist.md
Read the template at skills/hackathon-guide/templates/checklist-template.md. Fill it in using everything from the conversation.
Critical requirements:
- Every item references a specific
spec.md subsection
- Acceptance criteria drawn from
prd.md
- Build preferences encoded in the header
- Items are sequenced with dependencies respected
- Devpost submission is the final item
- All items use the consistent five-field format
- Items are atomic — completable in one
/build session
Write it to docs/checklist.md.
After Generating the Checklist
Embedded Feedback
Provide 2-4 sentences using ✓/△ markers. Evaluate:
- Sequencing logic (do dependencies flow correctly?)
- Granularity (atomic enough for single sessions? not too many items for the time?)
- Completeness (does every spec section have a corresponding item?)
- Realism (can this be built in 3-4 hours given the learner's level?)
Handoff
- If autonomous mode: "Run
/clear, then run /build when you're ready. I'll work through the whole checklist and pause at checkpoints for you to verify."
- If step-by-step mode: "Run
/clear, then run /build when you're ready — you'll run it once per checklist item. Run /clear between each item so the agent gets a fresh context. Each run picks up the next unchecked step."
Process Notes
Append to process-notes.md under the ## /checklist section:
- Sequencing decisions and rationale
- Methodology preferences chosen (build mode, verification, comprehension checks)
- How many items and estimated total build time
- What the learner was confident about vs needed guidance on
- Submission planning notes
- Deepening rounds: How many rounds did the learner choose? Did they refine item granularity, catch missing dependencies, or sharpen verification steps?
- Active shaping: Note whether the learner engaged with sequencing logic or just accepted the proposed order. Record if they questioned item order, suggested different groupings, or had strong opinions about the build plan.
Conversation Style
Everything from the hackathon-guide SKILL.md interaction rules applies here, plus:
- This should be shorter than /spec. The heavy thinking is done. This is translation into an ordered plan. Don't belabor it.
- Make the sequencing logic visible. Don't just list items — explain why each is in its position. The learner should understand the reasoning.
- Count and gut-check. If the math doesn't work (too many items for the time, or items that feel too big), flag it and adjust together.
- The checklist is a living contract with /build. Whatever you write here is what the build agent will execute. The five-field format must be consistent across every item so /build always knows where to find the spec ref, the work description, the acceptance criteria, and the verification step. But the checklist isn't sacred — if something breaks during the build, the checklist can and should be updated. That's normal. Plans meet reality and adapt.
