skill-craft

Design system architecture and make technology decisions. Uses a structured inline design or an available independent reviewer for comprehensive work. AUTOMATICALLY INVOKE when: - "design architecture", "what tech stack", "system design" - "how should I structure", "architecture for", "design system" - "technology recommendation", "compare frameworks" NOT for: Direct implementation or a routine task breakdown.

A delegation skill, not an agent type. Ask native agents to work in the background, continue useful work in the main conversation, and incorporate their results when they return. Use for "ask an agent", named agent roles, parallel delegation, or launch-and-notify work.

Use when an implementation task needs a dependency-aware plan before coding: backward planner / backchain / precondition-first planning, dependency DAGs, elaborating incomplete plans, or scheduler-ready step graphs. Turns a natural-language coding request into a forward draft then backward-chaining enriched DAG with explicit unresolved risks, then directly calls the selected Until Loop to repeat dependency review until two consecutive trivial/no-change reviews.

Resolve ambiguous user prompts by choosing whether to answer now, answer with assumptions, ask 1–2 high-value clarification questions, replan, or stop. Use when the best response depends on hidden intent, audience, scope, constraints, risk, output format, or desired depth.

Compare two prompt versions (A vs B) by running both against a directory of test input files, then evaluating results on three dimensions in priority order: quality > tokens > time. **AUTOMATICALLY INVOKE** when user mentions: - "compare prompts", "which prompt is better", "prompt efficiency" - "A/B test prompts", "evaluate prompts", "test these prompts" - Multiple prompt variations to choose between **STRONGLY RECOMMENDED** for: - Optimizing prompt quality - Reducing token usage - Comparing alternative approaches - Before finalizing agent/skill prompts Position bias mitigated via randomized ordering per test case — judge sees A/B in random order, results remapped before aggregation.

Use when a repository candidate needs a deliberate review-and-improvement loop: use recent Git history, make warranted changes, run meaningful checks, and repeat until two consecutive passes make no changes, or the first pass completes with no change. Supports a read-only interpretation preview; not a one-off code review.

Use when Improve should run in a fresh native agent instead of this conversation: start one agent that runs /improve in its own context, keep useful parent work going, then verify and relay its result. For Improve in this conversation, use improve.

Benchmark and compare system prompt variants (V2/V2a/V2b/V2c) for Sheets Chat by running test scenarios through the real GAS-side ClaudeConversation pipeline. Tests both system-placement and user-placement, then evaluates with heuristic scoring (ABTestHarness) and LLM-as-judge. Product-specific: requires an authorized Sheets Chat/GAS execution integration. **AUTOMATICALLY INVOKE** when: - User says "benchmark system prompts", "compare prompt variants" - User wants to evaluate placement (system vs user message) - User says "which prompt variant is best", "run prompt benchmark" **NOT for:** General prompt engineering, non-GAS prompts, one-off prompt writing. Use an ordinary prompt-editing workflow for editing/refining an active prompt.

Use when executing or resuming an agreed dependency graph with one main dispatcher and parallel native workers: claim ready steps, preserve launch state, collect result evidence, verify outcomes, and identify successors. Planning belongs to Backchain or the caller; this skill executes the plan.

Generate comprehensive tests for code. Uses an inline strategy or an available independent test specialist for complex components. AUTOMATICALLY INVOKE when: - "write tests", "test this", "generate tests", "add test coverage" - "test plan", "what should I test" - After feature implementation is complete NOT for: Only running an existing suite without designing or changing coverage.

Compare an agent or skill prompt against its test harness skill for phase-model, skip-condition, and wiring consistency. Reports mismatches and identifies which file is authoritative.

Audit an agent or skill prompt file for internal inconsistencies (phase numbering, behavioral contracts, terminology, stale references). Produces a Q&A with info-gain scores, a learnings section, and a remediation list. Use before any prompt migration.

TDD-based prompt migration — given a target agent/skill prompt and a remediation list, writes failing tests first, then updates the prompt to make them pass. Commits remain user-controlled.

Full prompt-improvement workflow — runs prompt-audit to find inconsistencies, presents a remediation plan, then runs prompt-migrate to apply fixes and prompt-align to verify harness consistency. Use for any agent or skill prompt that needs structural repair.

Add a post-ship improve-to-exhaustion directive to a plan, or run that directive after implementation. Invoke like any skill: /review-coverage, "review-coverage on this plan", residual×2 after ship, or ensure the plan has a /goal-ready ## Review Coverage section. Not for pre-exit plan quality (use review-plan) or raw residual×2 engine mechanics alone (use review-converge under /goal).

Compare two code-review prompt versions against supplied fixture ground truth using an explicitly configured external benchmark runner. Reports an F1-based verdict only when the runner completes both evaluations and comparison.

Markdown-authoritative delivery harness. Start or resume once, follow the script's current action packet, and submit its exact completion call until the script reports completion with an HTML achievement report. Use when the user says shiploop, ship the project, or requests a durable delivery loop.

Run the ShipLoop test harness and audit its retained graph, review, test, product and incremental-change evidence. Use for ShipLoop mock checks, live one-shot E2E smoke/full campaigns, or review of existing trial output. Full Google Apps Script and Salesforce game cases require an authorized test deployment and hosted behavior evidence. Includes its harness for source and marketplace installs; tests a separately selected ShipLoop.

Use when authoring or reviewing a portable multi-host agent skill (Grok, Claude Code, Codex, Hermes): scaffold a prompt-only skill, make a skill host-agnostic, create skill layout, review skill for interop, or install across hosts. Covers prompt-first design, host matrix, anti-patterns (divergent copies, silent mode fallback, abs symlinks), and script-backed CLI contracts.