Nw SpikeSAFE

AI agents that guide you from idea to working code, with you in control at every step.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: nw-spike
description: "Runs a timeboxed PROBE to validate one core assumption, then optionally PROMOTES the probe into a walking skeleton — the first e2e thin slice of the feature, committed and demo-able. Use after DISCUSS when the feature involves a new mechanism, performance requirement, or external integration."
user-invocable: true
argument-hint: "[feature-description] - Example: \"wave-matrix -- derive feature status from pytest + filesystem\""
---

# NW-SPIKE: Probe → Walking Skeleton

**Wave**: SPIKE (between DISCUSS and DESIGN) | **Agent**: Attila (nw-software-crafter) | **Command**: `/nw-spike`

## Overview

Execute a two-phase wave that turns a risky assumption into visible, iterable value as fast as possible:

1. **PROBE** — quick throwaway validation of one core assumption (30-60 min, code in `/tmp/`)
2. **PROMOTION GATE** (interactive) — ask the user whether to promote the probe
3. **WALKING SKELETON** — refactor the probe into an end-to-end thin slice committed to the repository (1-3 h, code in `src/` + 1 acceptance test)

The PROBE answers "does the mechanism work?". The WALKING SKELETON answers "can a user see it working end-to-end?". You never throw away working validated code — you promote it and iterate.

## When to use

The spike is needed when the feature introduces:
- A new mechanism never tried in this codebase
- A performance requirement that cannot be validated by reasoning alone
- An external integration with unknown behaviour

If none of the above apply, skip SPIKE and go to DESIGN.

## Prior Wave Consultation

1. **DISCUSS artifacts**: Read `docs/feature/{feature-id}/discuss/` (required)
   - `user-stories.md` — scope and acceptance criteria
   - `wave-decisions.md` — constraints and assumptions to test
2. **DIVERGE artifacts**: Read `docs/feature/{feature-id}/diverge/recommendation.md` (if present)

## Interactive decision points (Phase 1 entry)

### Decision 1: Probe scope
**Question**: What is the ONE assumption you need to validate?
**Examples**:
- "Can we parse pytest output reliably in <5 seconds?"
- "Can the CEL library evaluate 100 expressions in <1 second?"
- "Can we write to `.git/hooks/` from a subprocess without corruption?"

### Decision 2: Performance budget
**Question**: What is the timing constraint? (Enter "none" if mechanism validation only)
**Examples**:
- "<5 seconds end-to-end"
- "<100ms per operation"
- "Handle 10K items without OOM"

### Decision 3: End-to-end path preview
**Question**: If this probe works, what would the thinnest end-to-end slice look like?
Capture the rough path: `user-facing entry → business logic → persistence/services → user-visible output`. This is **not** a commitment — it's context for the promotion gate later.

## Phase 1 — PROBE

Throwaway validation of the assumption.

### Agent invocation

@nw-software-crafter

Execute PROBE for "{feature-description}".

**Probe question**: {Decision 1 answer}
**Performance budget**: {Decision 2 answer}
**Target e2e path (for later)**: {Decision 3 answer}

**Rules**:
- Code goes in `/tmp/spike_{feature_id}/`. Never in `src/`.
- Max 1 hour. No tests, no types, no error handling, no abstractions.
- One file preferred. Two files maximum.
- Use `time.perf_counter()` for timing.
- Print results to stdout.

**After probe completes**:
1. Write findings to `docs/feature/{feature-id}/spike/findings.md` — binary verdict (WORKS / DOESN'T WORK), timing, edge cases, design implications.
2. Do **not** delete the probe code yet — wait for the promotion gate.
3. Report verdict and ask the orchestrator to run the promotion gate.

## Phase 2 — PROMOTION GATE (interactive)

Run this gate **only after** the probe completes and findings.md is written.

Present the user with three choices:

| Choice | When to pick | Outcome |
|---|---|---|
| **PROMOTE** | Probe verdict is WORKS and the mechanism is worth building on | Proceed to Phase 3 — walking skeleton |
| **DISCARD** | Probe verdict is WORKS but not worth pursuing (findings are enough) | Delete `/tmp/spike_{feature_id}/`. Commit `findings.md`. Hand off to DESIGN. |
| **PIVOT** | Probe verdict is DOESN'T WORK or revealed a better approach | Delete probe code. Annotate `findings.md` with the pivot. Either loop back to DISCUSS or run a second probe. |

**Default**: if the probe verdict is WORKS and no reason to stop, recommend PROMOTE but let the user override.

Record the promotion decision in `docs/feature/{feature-id}/spike/wave-decisions.md` as an explicit wave decision with rationale.

## Phase 3 — WALKING SKELETON (only if PROMOTE)

Refactor the probe into the thinnest end-to-end slice that is committed, tested, and demo-able.

### Definition of Done (walking skeleton)

1. **End-to-end path**: the slice enters from a real user-facing entry point (CLI command, HTTP endpoint, UI action, hook) and exits at a real user-visible output (stdout, HTTP response, rendered screen, persisted file). Every layer in between is exercised — **no layer is mocked** unless that layer is an external paid service classified as costly in DISTILL's Walking Skeleton Strategy (then use the fake/contract test pattern).
2. **One acceptance test**: a `@walking_skeleton @driving_port` tagged scenario in `tests/{test-type-path}/{feature-id}/acceptance/walking-skeleton.feature`. The scenario MUST be green before hand-off.
3. **Production location**: code lives under `src/{production-path}/`, not in `/tmp/`. Minimal module skeleton is fine — no premature abstractions, no features beyond the walking skeleton.
4. **Committed**: the walking skeleton commit message is `feat({feature-id}): walking skeleton — {one-line description}`.
5. **Demo-able**: running the single acceptance test (or the real entry-point command) produces visible output that matches the user story from DISCUSS.
6. **Back-propagation**: if building the skeleton reveals a contradiction with DISCUSS or DESIGN, write the contradiction to `docs/feature/{feature-id}/spike/upstream-issues.md` and stop 

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.