Atlas / Skills / nwave-ai / Nw Execute

Nw ExecuteSAFE

skills/nwave-ai/nw-execute

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

Verdict
SAFE
Grade
B
Trust score
89 /100
Version
—
Hosts
—
License
MIT
Stars
617
01

Overview

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

Read from source at commit 1a864ec81cdeOBSERVED · 2026-10-14
Advertisement
02

What it tells the agent

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-execute
description: "Use when a DELIVER roadmap already exists and you need to dispatch exactly one identified step through its TDD cycle. Use nw-roadmap to create the plan, nw-deliver for the whole wave, and nw-continue to resume at the next inferred step."
user-invocable: true
argument-hint: '[agent] [feature-id] [step-id] - Example: @nw-software-crafter "auth-upgrade" "01-01"'
---

# NW-EXECUTE: Atomic Task Execution

**Wave**: EXECUTION_WAVE | **Agent**: Dispatched agent (specified by caller)

## Overview

Dispatch one unit of DELIVER work to an agent: a single roadmap step. `/nw-execute` extracts the step from `roadmap.json` and dispatches it for the 3-phase TDD canon; the agent appends phase events to `execution-log.json`.

## Syntax

```
/nw-execute @{agent} "{feature-id}" "{step-id}"
```

## Context Files Required

- `docs/feature/{feature-id}/deliver/roadmap.json` — Orchestrator reads once, extracts step context
- `docs/feature/{feature-id}/deliver/execution-log.json` — Agent appends only (never reads)

## Rigor Profile Integration

Before dispatching the agent, read rigor config from `.nwave/des-config.json` (key: `rigor`). If absent, use standard defaults.

- **`agent_model`**: Pass as `model` parameter to Agent tool. If `"inherit"`, omit `model` (inherits from session).
- **`tdd_phases`**: Modify the TDD_PHASES section in the DES template to match the configured phases. The 3-phase canon (ADR-025) is `[RED, GREEN, COMMIT]`; the lean variant is `[RED, GREEN]`. Legacy 5-phase contract (`[PREPARE, RED_ACCEPTANCE, RED_UNIT, GREEN, COMMIT]`) and its lean variant (`[RED_UNIT, GREEN]`) remain supported for audit-log replay of pre-2026-05-07 commits. Remove omitted phases' instructions from the template.
- **`refactor_pass`**: If `false`, skip COMMIT phase refactoring instructions.

## Dispatcher Workflow

1. **Parse Parameters** — Extract agent name, feature ID, and step ID from invocation. Gate: all three parameters present and non-empty.
2. **Load Rigor Profile** — Read `.nwave/des-config.json` key `rigor` (default: standard if absent). Gate: config loaded or default applied.
3. **Validate Context Files** — Confirm `roadmap.json` and `execution-log.json` exist under `docs/feature/{feature-id}/deliver/`. Gate: both files present; report path-not-found if missing.
4. **Extract Step Context** — Grep roadmap for `step_id: "{step-id}"` with ~50 lines context. Gate: step found; report available step IDs if missing.
5. **Invoke Agent** — Call Agent tool with DES template below, applying rigor model and phases from step 2. Gate: Agent tool called, not executed inline.

## Agent Invocation

@{agent}

Use this DES template verbatim. Fill `{placeholders}` from roadmap. Without DES markers, hooks cannot validate.

```
<!-- DES-VALIDATION : required -->
<!-- DES-PROJECT-ID : {feature-id} -->
<!-- DES-STEP-ID : {step-id} -->

# DES_METADATA
Step: {step-id}
Feature: {feature-id}
Command: /nw-execute

# AGENT_IDENTITY
Agent: {agent-name}

# SKILL_LOADING
Before starting TDD phases, read your skill files for methodology guidance.
Skills path: ~/.claude/skills/nw-{skill-name}/SKILL.md
Always load before RED: tdd-methodology.md, quality-framework.md (3-phase canon, ADR-025) — legacy 5-phase logs reference loading at PREPARE.
Load on-demand per phase as specified in your Skill Loading Strategy table.

# TASK_CONTEXT
{step context from roadmap - name|criteria|test_file|scenario_name|implementation_notes|deps|files_to_modify (per nWave/templates/roadmap-schema.json)}

# DESIGN_CONTEXT
{Summary of architectural decisions relevant to this step, extracted by the orchestrator from docs/product/architecture/brief.md and wave-decisions.md. Include: component structure, dependency boundaries, technology choices, and any design constraints that affect implementation. If no design artifacts exist, write "No design artifacts available — use project conventions."}

# TDD_PHASES
3-phase canon (ADR-025, 2026-05-07). Execute in order:

1. RED - Activate the pre-authored acceptance test (PRIMARY TBU DEFENSE); write PBT unit tests ONLY if the AT cannot reach GREEN without them.
   AT activation: If TASK_CONTEXT includes test_file, locate it and remove the @skip/@ignore/@pending/xit/.skip/[Ignore] marker from the target scenario (the AT scaffold was authored by DISTILL — DELIVER does NOT re-author ATs). Run it — must fail for business logic reason (not import/syntax error). Fail-for-right-reason gate: collected ≥ 1, failures ≥ 1, no collection errors, semantic AssertionError / expected-exception-not-thrown.
   PORT-TO-PORT PRINCIPLE: The acceptance test exercises the scenario through
   the driving port (application service, orchestrator, CLI handler, API controller),
   not a decomposed helper or internal class. A correctly-written port-to-port test
   makes TBU structurally impossible — if a new function were missing or unwired,
   THIS test stays RED. That is the entire point: GREEN is unreachable without wiring.
   Litmus test: "If I delete the call-site that wires the new code, does this test fail?"
   If no → the test is at the wrong level. Stop and flag to orchestrator (DISTILL re-author needed).
   Conditional unit-test authoring: write PBT unit tests (or integration tests for adapter/infrastructure code — adapters use real infrastructure, never mocked unit tests) ONLY when the AT requires them to reach GREEN. If the AT can pass via direct minimal implementation, skip unit-test authoring inside RED.

2. GREEN - Minimal code to pass AT + any unit tests authored in RED.
   After GREEN: run FULL test suite. If all pass, proceed to COMMIT immediately.
   Smell test: if any new function is only called from test code, your acceptance
   test is at the wrong abstraction level — stop and flag.
   Never move to new task or stop without committing green code.

3. COMMIT - Commit this step's owned files via `des-commit` (parallel-safe).
   Use `des-commit` instead of raw `git add` / `git commit`. It holds an
03

Trust audit

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.

LayerWhat it checksResult
L0Provenance & inventoryPASS
L1Static analysis of the codeNA
L2Instruction surface (what it tells the agent)PASS
L3Class-specific surfacePASS
L4Behavioural (sandbox)SKIPPED

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.

Audited 2026-10-14 · audit v0.4.1 · source sha 1a864ec81cdefull audit observations/trust-audit/skill/nwave-ai__nw-execute.json · Report an issue / request a re-scan
04

Audit history

Every audit this skill has had.

DateSourceVerdictGradeScoreChange
2026-10-141a864ec81cdeSAFEB89first audit
05

Questions

What does the Nw Execute skill do?

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

Is Nw Execute safe to install?

The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.

What can Nw Execute access on my machine?

The audit observed no filesystem, network or shell use at all in its source.

How current is this page?

The grade is for one exact copy of the source (1a864ec81cde), read on 2026-10-14. The repository is watched, and a new audit runs when it changes — this is the first audit.

Advertisement