skills

Single entry point for structured requirements → architecture alignment. Phase 1 grills the product owner to extract BDD scenarios (tabular, template-locked). Phase 2 takes all collected BDD + SWE initial thoughts, grills the SWE on architecture, and records the result as an SDD in document/docs/specs/. Both artifacts are HTML filled from bundled templates so downstream can convert to docx via pandoc. Invoke with no args to start Phase 1, or paste BDD artifact(s) + your initial plan to go straight to Phase 2.

Turn a valuable working session into a polished, bilingual (zh-TW + en) public technical blog post and stage it as a draft in the lyeh-infra Hugo blog. Manually triggered with /blog when the user feels the current session produced an insight worth publishing. Use when the user says "/blog", "write this up", "blog this", "把這個寫成文章", or wants to capture the value/lessons of a session as an article.

Relentless one-question-at-a-time design grilling — the same interrogation core as grill-me — that also preserves the decision's evolution as a graph: rejected branches and why, back-references to earlier decisions, and re-assumptions. Each resolved decision is held in context, then the whole graph is written and rendered once, at convergence, so nothing ever slows a question. Use this instead of grill-me whenever a design discussion is likely to branch or to reopen earlier decisions and you want the decision topology preserved, then converged into spec.md. For a quick throwaway linear stress-test use grill-me; for domain-language / glossary alignment use grill-with-docs.

Learning mentor for coding-focused domains (programming languages, frameworks, algorithms, systems, ML/theory papers). Drives true mastery instead of completing the task for you: decomposes a learning goal into stages and runs each through gap-surfacing → guided paper reading → hands-on implementation → teach-back, withholding exactly the skill the goal targets. Must be invoked explicitly by the user — does not auto-trigger.

Create polished dark-themed architecture diagrams as self-contained HTML+SVG files. Use when the user asks for system, infrastructure, cloud, security, or network topology diagrams.

Scan all staged/unstaged changes, combine with session context and plan, generate an ordered conventional-commit plan within a token budget. Use when the user says "split commits", "commit plan", wants to organize changes into separate commits before a PR, asks how to structure pending work into git history, or mentions staging files with multiple logical changes.

Scan the current session context and the project's adr/ directory to create or update an Architecture Decision Record. Use after any decision point (/grill-me, /plan, or design discussions) when the user invokes /record-adr. Maintains adr/README.md as an auto-updated index.

Bootstrap Claude Code framework config for a new project. Generates a CLAUDE.md framework section and .claude/settings.json with standard harness. Use when starting a new project session and want to configure Claude Code settings interactively.

Save current session state to a private checkpoints repo so work resumes seamlessly after a Pro quota reset, context window limit, or machine switch. Use when approaching usage limits, switching machines, or when the user says "checkpoint", "save state", "save progress", or "I need to switch".

Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.

Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.