spike · v1.0.0 · 2026-09-25 · sha256 ca3dfe5c2eb0369f
spike v1.0.0A
Immutable. This exact content never changes and is served at /api/v1/blob/ca3dfe5c2eb0369f.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_oej2sncwuhc7nqr6/ca3dfe5c2eb0369f5e90a0fe7966b302460630e39a06253f672380945596e2f4.json
Then /plugin install spike@markdownregistry-oej2sncwuhc7nqr6 in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
---
name: spike
description: Throwaway exploration to validate an idea before committing to a real implementation. Time-boxed; isolated under the workspace's .ethos-work/spikes/ so it never pollutes the project's source. Use when the question is "is this even feasible?".
version: 1.0.0
author: ethosagent
tags: [coding, planning, prototyping]
required_tools: [read_file, write_file, terminal]
ethos:
category: planning-and-process
default_personalities: [engineer, coordinator]
prerequisites:
external_cli: []
auth: []
env_vars: []
optional_tools: [process]
integrates_with:
- tool: process
role: long-running spikes (e.g. a server or watcher) run as a background process so they can be killed cleanly
surface_metadata:
invocation_trigger: "user says 'let's just see if X works' / 'quick experiment' / 'prototype'; agent self-invokes when the question is feasibility"
estimated_turns: "3-10 (small, by design)"
---
# Spike
Throwaway exploration. The output is "yes / no / maybe with these caveats", not production code.
## When to use this skill
- The question is "is this even feasible?" — you do not yet know whether the right approach exists.
- The user said "prototype this", "just see if X works", "quick experiment".
- A spike is cheaper than reading docs (e.g. you'd rather find out how a library actually behaves than guess from its README).
When the answer is already known, write real code. Do not spike.
## What this skill writes
Spikes live under `.ethos-work/spikes/<slug>/` in the workspace. They never go inside the project's source tree: `.ethos-work/` is scratch space, kept out of git by adding `.ethos-work/` to `.git/info/exclude`. (Not under `~/.ethos`: the terminal guard refuses any command that names the Ethos state dir.) This is deliberate — spikes are throwaway, and putting them in the project would invite "but it works in the spike" arguments.
Nothing expires or deletes spikes automatically: a spike directory stays until someone removes it. Clean up by hand (see "Outcome" below).
## The procedure
1. **Acknowledge that this is throwaway.** The first line of any spike file is `# Spike: <question>`. Set the user's expectation that nothing here is going to ship as-is.
2. **Pick the smallest scope that answers the feasibility question.** A spike is not a feature. If you find yourself adding a second concern, stop — that's a sign the spike has scope-crept and the answer is already "yes, but with these constraints".
3. **Set up the spike directory:**
```
.ethos-work/spikes/<slug>/
├── README.md # the question, the approach, the result
└── <code, scripts, fixtures>
```
4. **Run / measure / report.** Whatever the question demanded — execute it, capture the output, write the result into `README.md`.
5. **Outcome.** End with a one-line recommendation:
- **Keep** — leave it where it is, may revisit. It stays until someone deletes it.
- **Promote** — worth turning into real code. There is no promote command: the user decides where it belongs in the project. Do not copy spike code into the source tree yourself; say which files are worth keeping and let the user bring them over, or write the real implementation fresh once they agree.
- **Discard** — delete `.ethos-work/spikes/<slug>/` now (`rm -r .ethos-work/spikes/<slug>`).
## Hard rules
- Spikes never touch the project's source files. If you want to spike with the project's code, copy the relevant files into the spike directory first.
- A spike that grows past ~200 lines is no longer a spike — it has become a proto-implementation. Stop, write a plan, then start over.
- Output a recommendation. A spike with no recommendation is not a finished spike.
## Integrates with
- The `process` tool, when the spike needs to run a long-lived process (a server, a watcher). Start it with `process_start`, leave it running, kill with `process_stop` when done. Do not block the chat on a long-running spike.