메뉴
Time-boxed throwaway investigation to answer one hard question. Use when facing technical uncertainty before a slice — the output is knowledge, not production code. Retires risk by producing a spike verdict with clear recommendations.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Read and write a Petri net (SDCPN) document by Automerge URL. Use when creating, editing, or querying Petri nets — adding or removing places, transitions, arcs, color types, differential equations, and parameters.
Create and maintain an Assumption Ledger — a persistent record of assumptions, their confidence, and validation status. Use when starting a new slice, resuming work in a new context window, or when implicit assumptions risk causing drift. Tracks requirements, architecture, and implementation assumptions.
Create or update a concept capsule — the conceptual anchor for a project or feature area. Use before writing code on a new project or feature, or when terms and boundaries feel unclear. Defines glossary, invariants, happy-path scenario, and non-goals.
Write a tracer-bullet card — a precise specification for one thin end-to-end slice of work. Use when scoping a new slice, defining what to build next, or breaking a feature into provable increments. Covers target behavior, boundary crossings, risks, and definition of done.
Create characterization tests (Golden Master) for existing code so you can refactor or replace safely. Use before refactoring, strangler-replacing, or modifying code with unclear behavior. Captures observable behavior with a minimal harness, producing tests, fixtures, and a coverage report.
Methodology triage consultant for tracer-bullet development. Use when unsure which pragma skill to run next, when starting a new project, or when the current approach feels stuck. Interviews the user, assesses state, and recommends the next pragma skill.
| name | spike |
| description | Time-boxed throwaway investigation to answer one hard question. Use when facing technical uncertainty before a slice — the output is knowledge, not production code. Retires risk by producing a spike verdict with clear recommendations. |
| argument-hint | [question to answer and what you'll try] |
Run a throwaway experiment to answer one specific, falsifiable technical question. Spikes produce knowledge, not code.
A spike is NOT a prototype (prototypes try to be kept). Spike code is explicitly discarded.
The question and approach: $ARGUMENTS
Before starting, confirm:
If the question is vague, sharpen it before proceeding.
Write the exact falsifiable question. Examples:
Bun.file() API read a 500MB file without loading it all into memory?"Single throwaway file. Hardcode everything. Print to stdout. Let crashes be information.
Do not modify project source files. Install dependencies in isolation if needed.
The verdict is the deliverable. Capture: did it work, any surprises, any conditional constraints, performance (if relevant).
Before returning, classify what changed and route durable outcomes:
/pragma:capsule update/pragma:roadmap update/pragma:milestone update/pragma:assumptions update/pragma:contractUse explicit none for lanes with no change.
/pragma:card or /pragma:slicenone | /pragma:capsule update — [reason]none | /pragma:roadmap update — [reason]none | /pragma:milestone update — [reason]none | /pragma:assumptions update — [reason]none | /pragma:contract — [reason]/pragma:assumptions update (mandatory after every spike) to record which assumption was validated/invalidatedState: discovery, Next: first non-'none' authority delta command (otherwise /pragma:consult or /pragma:card when uncertainty is retired and next behavior is clear), Loop: /pragma:consult (default unless user explicitly continues directly)