---
name: wf-spike
description: 'WellForge command — Spike a goal fast — main-loop build from a one-paragraph brief, advisory gates, no agents, no approval (rigor tier: spike). Invoke explicitly: $wf-spike <goal — what to prototype / which feasibility question to answer> [--no-terse]'
---
> WellForge command, generated for Codex from `wellforge-plugin/commands/spike.md` — do not edit; regenerate with `adapters/codex/generate.py`.
> Invoked as `$wf-spike <goal — what to prototype / which feasibility question to answer> [--no-terse]`. **ARGUMENTS** below means the text the user wrote after `$wf-spike` in the invoking message (empty if nothing).


Build a **spike**: the fastest WellForge path, for feasibility / PoC / business-model
experiments that may be thrown away. Load the **rigor-tiers** skill now — this command IS
the `spike` tier. Resolved tier: **spike** (invocation). Load the **terse** skill now — its
activation matrix defaults terse mode **ON** for this command.

Goal: ARGUMENTS

## Step 0 — Resolve terse

Terse is **ON by default** for spike (terse skill activation matrix — no flag required). If
`--no-terse` appears in `ARGUMENTS`, resolve terse **OFF** for this run instead, and strip
the token before using the rest of `ARGUMENTS` as the goal/brief text. State the resolved
terse state up front, next to the tier line above.

## What spike means here

- **Main loop only — you implement directly. Spawn NO subagents.** The whole point is to
  skip orchestration cold-starts. (mvp/production are where the agent team runs.)
- **Effort: minimal** (the spike tier cue, rigor-tiers skill) — apply it to yourself: take
  the shortest path that answers the question, no gold-plating or exhaustive edge-case work,
  leave `// SPIKE:` where you cut a corner. Bias hard to speed.
- **Terse: ON by default** (the terse cue, **terse** skill) — apply it to yourself, exactly
  like the effort cue above, unless Step 0 resolved it OFF: drop filler, preamble, and
  restating the question from your conversational output; prefer fragments; substance only.
  The terse skill's byte-identical invariant (code blocks, shell commands, file paths,
  identifiers, error messages reproduced byte-for-byte) and its artifact exemption
  (spec-driven/governance artifacts always stay full-contract) apply regardless of this
  setting. This governs your conversational output for the whole run, including the Report
  below — when Step 0 resolved terse OFF (`--no-terse`), report and narrate normally instead.
- **No human approval gate.** You build, then the human reviews the result.
- **Advisory gates** (lint/typecheck/build): run them, report failures, do NOT stop on them.
- **The security floor still blocks** (see rigor-tiers): secret scan, no hardcoded creds,
  critical-CVE dependency audit. Fast never means "leaks credentials."

## Procedure

1. **Resolve the folder.** Next sequential `specs/NNN-slug/` (slug from the goal). Reuse an
   existing folder only if the goal clearly continues it.

2. **Write `specs/NNN-slug/brief.md`** — short, not a spec. Frontmatter then ~5–10 lines:
   ```markdown
   ---
   id: NNN
   slug: <kebab>
   rigor: spike
   status: in-progress
   created: <today>
   ---

   # <goal>

   ## Question            <!-- the feasibility / business question this spike answers -->
   ## Build               <!-- 3–6 bullets: what you'll actually make to answer it -->
   ## Out of scope        <!-- what you're deliberately NOT doing (this is a spike) -->
   ## Findings            <!-- filled in AFTER: did it work? what did you learn? -->
   ```
   No user stories, no ACs, no plan. If the goal is genuinely unclear, ask ONE batched
   AskUserQuestion round — but prefer a sensible assumption recorded under `## Out of scope`.

3. **Build it.** Load the relevant stack skill (`react-ts-vite`, `kotlin-springboot`,
   `hono-ts-backend`, `mise`, …) and write the code directly. Favor the shortest path that
   answers the question; leave honest `// SPIKE:` markers where you cut a corner that
   `$wf-promote` would later pay off. Keep changes scoped to the goal.

4. **Sanity check (advisory).** Run the project's lint / typecheck / build (mise tasks if
   present). Report what passed/failed. Do NOT block or loop on failures — note them under
   `## Findings`. Advisory means *don't stop the spike*, NOT "make it green quickly": if you
   do choose to fix something, the `systematic-debugging` iron law still applies (root cause
   before the fix, no timeout/retry/skip to silence it) — the tier lowers ceremony, never
   honesty about why something broke. THEN run the **security floor** (secret scan + critical-CVE audit): if it
   trips, STOP and fix — the floor is the one thing a spike cannot skip.

5. **Fill `## Findings`** in brief.md: did the spike answer its question? What was learned?
   Then **run the `$wf-done` procedure** (spike branch) to close it — don't set
   `status: done` here. It is the single guarded place that transition lives, it stamps
   `done: <today>` too, and its spike branch is exactly this tier's bar (findings filled, the
   question answered) plus the "if it proved out, suggest `$wf-promote`" nudge.

6. **Record the run (lightweight).** Per the **observability** skill, write
   `.forge/runs/<run_id>.json` (schema `wellforge-run/v3`) with `command: spike`,
   `rigor: spike`, `plugin_version` (the running plugin), no agents (`agents: []`), the
   advisory gate outcomes, and `result`. Set
   `terse` to the boolean resolved in Step 0 (`true` unless `--no-terse` was passed — spike
   defaults terse ON); leave `control_run_id` `null` (pairing to a control run is a later
   concern, not this command's). Leave `tokens`/`cost` null.

## Report

End with:
- What you built and where (files), and the spike's verdict on its question.
- Advisory gate results (actual numbers) + security-floor result.
- The reminder: **"rigor: spike — gates were advisory, this is not production-ready. Run
  `$wf-promote NNN-slug --to mvp` (or `production`) to graduate it."**

## Hard rules

- Never spawn a subagent — if you reach for one, you're in the wrong command (use
  `$wf-orchestrate --mode mvp`).
- Never present spike output as production-ready, and never silently raise its rigor —
  graduation is `$wf-promote` only.
- The security floor is non-negotiable; everything else is advisory.