Twelve methods. None mandatory by default.

Clarifies unsettled product, interaction, or architecture choices until implementation can proceed with confidence.

Use when a real decision is still open.

Carries an approved specification into a complete, traceable implementation plan without shrinking the accepted outcome.

Use when the specification is already authoritative.

Structures settled multi-step work when no adopted specification already owns the complete plan.

Use when sequencing matters before edits begin.

Turns a settled request or plan into working code through coherent, verified slices.

Use when the route is known and the work should move.

Reproduces the failure, traces its causal mechanism, and restores the contract with focused evidence.

Use when behavior is wrong or unexplained.

Uses tests as design and evidence, with strict red-green discipline where it materially sharpens the contract.

Use for rules, state, parsers, regressions, and risky boundaries.

Inspects a diff or implementation for concrete, actionable defects against the real goal and repository truth.

Use when another pair of eyes should challenge the change.

Validates external review feedback against current source before applying a narrow, justified correction.

Use when feedback arrives from a reviewer or bot.

Matches fresh evidence to the exact claim, broadening checks only when the blast radius requires it.

Use when readiness or completion needs proof.

Creates isolation when it reduces real risk while preserving ownership, existing work, and cleanup boundaries.

Use for long, risky, or concurrent repository work.

Splits genuinely independent work into bounded lanes with one writer per surface and an explicit return.

Use when parallelism repays its coordination cost.

Closes the engineering loop with scoped validation, documentation truth, integration choices, and an honest handoff.

Use when the work must become safely resumable or ready to integrate.