nicopowers

nicopowers

Nico's personal engineering-process skills, shared across Claude Code and Codex.

The accepted target is a standalone maintained fork: Superpowers is its reviewed primary upstream, selected Matt Pocock skills are secondary upstream inputs, and neither is a runtime dependency. Exact source revisions, mappings, update commands, and deliberate differences live in UPSTREAM.md. See ADR 0004 and the approved replacement design.

The replacement source is complete. The complete 14-skill Superpowers foundation, its Claude Code bootstrap hook, and three complete Matt engineering disciplines are imported with documented Nicopowers adaptations. Grounded architecture review is available only by direct user invocation. The repository now ships standalone Claude Code and Codex bootstrap paths. Isolated installation and behavior evidence is recorded in the replacement verification report. Its linked invocation ledger separates structural checks, executed behavior, and read-only control-flow simulations. Live cutover remains a separate, later operation. The exact rollback-ready cutover checklist records the observed live owners, commands, stop conditions, and rollback paths.

Status

This repository is the installation source, but no live Claude Code or Codex configuration is changed by cloning or validating it. The native package and router passed isolated installation checks on 2026-07-22. The framework is verified but not yet cut over; live execution requires a separate explicit confirmation of the reviewed Live cutover checklist.

Why

These encode one deliberate engineering spine, so an agent doesn't improvise the process on every task: align on a design before coding, plan multi-step work, implement test-first, debug to root cause, prove claims with evidence, take review without sycophancy, and isolate risky work in a worktree. The disciplines fire automatically (see docs/invocation.md); the ceremony scales to the risk, the gates stay strict.

Workflow Depth

Nicopowers automatically chooses the lightest safe path. Quick uses a chat-only brief and a narrow check for clear, small changes. Standard uses an approved specification and implementation plan for meaningful or risky work. Deep covers ambiguous, multi-system, or multi-session work and offers Wayfinder as an opt-in decision map. A direct request such as use Wayfinder counts as consent. Depth controls which process skills apply: a clear Quick behavior or configuration change does not invoke brainstorming or planning just because behavior changes. Standard owns design, specification, and planning when concrete uncertainty appears. An explicit user choice of depth always wins.

Skills

Foundation skills are model-invoked — the agent reaches for them automatically, and you can invoke them by name. Explicit-consent workflows can be user-invoked only. Vocabulary is in CONTEXT.md; the model/user-invoked split is in docs/invocation.md.