maga2010org/hackathon-run/stack-picker

stack-picker

Choosing a stack under pressure is the single highest-leverage decision in the first hour. Wrong stack = 4 hours of yak-shaving instead of building. Right stack = an MVP by midnight.

Input contract

Required:

• team_skills: list of strings ("python" | "javascript" | "typescript" | "react" | "react-native" | "swift" | "kotlin" | "rust" | "go" | "java" | "ruby" | "elixir" | ...)
• time_remaining_minutes: integer >= 0
• demo_format: one of web | mobile | desktop | cli | api | data | ml | hardware

Optional:

• prize_category: free text ("best ai use" | "best sustainability" | "best new developer tool" | ...)
• must_integrate: list of external APIs or services the project must touch
• team_size: integer >= 1 (default 1)

Execution

1. Filter viable stacks

For each entry in the candidate table, the stack is viable iff:

• All required team skills are present in team_skills, OR
• The team has at least one generalist (javascript + python both count)

2. Score each viable stack

score = skill_match * 3 + speed * 2 + prize_fit * 2 - setup_cost

3. Recommend top 1 + runners-up

Output:

• recommendation: the top-scored stack, with a one-line rationale
• runners_up: next 2 stacks, with the tradeoff each one buys
• anti_recommendation: the worst-scored viable stack, with the failure mode

4. Emit the 30-minute setup

For the recommended stack, emit a step-by-step 30-minute bootstrap:

1. Install / clone / scaffold command
2. One minimal "hello world" file path + content
3. The single command that proves it works (browser URL, CLI exit, curl)

Helper

The skill ships scripts/pick.py which implements the scoring algorithm above end-to-end and writes both .hackathon/state/stack.json and .hackathon/artifacts/stack-bootstrap.md. Run it with:

python3 skills/stack-picker/scripts/pick.py \ --team-skills python,javascript \ --time-remaining 240 \ --demo-format web \ --out-dir .hackathon

Output contract

Files written:

• .hackathon/state/stack.json (NEW schema)
• .hackathon/artifacts/stack-bootstrap.md (30-min setup walkthrough)

Acceptance criteria

• [ ] Outputs exactly one recommendation from the viable set.
• [ ] Each recommendation has a one-line rationale.
• [ ] Recommendation stack is reachable inside setup_cost minutes.
• [ ] anti_recommendation is named and the failure mode is concrete.
• [ ] Bootstrap walkthrough is runnable end-to-end in <= 30 minutes.

Failure modes

Trigger phrases (for agent intent matching)

• "what should we build with"
• "what language for a hackathon"
• "we need to pick a stack"
• "best stack for X"
• "what framework"