tools

Development wisdom and workflow rules. NOT for project-specific conventions (those live in CLAUDE.md, edit via wisdom).

The `BUGS.md` open-issues queue — entry format, lifecycle, pruning to diary. NOT for the record-don't-fix policy (that's CLAUDE.md Bug Triage Protocol), NOT for resolved-bug history (use /diary), NOT for feature backlog (use TODO.md/specs).

Terminal demo GIF + MP4 recordings for READMEs and social (asciinema + agg + ffmpeg). NOT for general Makefile targets (use software) or GUI screenshots (use visual).

Go development. NOT for non-Go code (use rs, py, ts, tsx, or sh).

Humanize text: strip AI-isms and add real voice. NOT for drafting new copy (use writing).

Python development. NOT for shell scripting (use sh) or non-Python code.

Router for engineering knowledge — the language-agnostic code baseline (naming, style, boring-code, design) plus deep runbooks (Docker images, CI Makefiles, deploys, observability, Python tool distribution). NOT for language-specific idioms (use go/rs/py/ts/sh/sql) or terse Docker/systemd rules (ops keeps those hot).

Write high-retention X/Twitter posts and threads. NOT for product copy or long-form marketing prose (use writing).

Install (or update) the Kronael toolkit into ~/.claude/ and bridge it into Codex. Copies skills, agents, hook scripts; merges Claude hook wiring; installs Codex hook wiring; installs the wisdom skill body as ~/.claude/CLAUDE.md; offers the standalone CLI tools (rig, udfix, clp, dockbox). First-time installs get an explained questionnaire. USE when the user says "install kronael", "install kronael tools", "install" (in this repo), or runs /kronael:install.

/learn — extract patterns into skills from session history, and evaluate the session for memory-worthy content. NOT for writing skills from scratch (use wisdom). NOT auto-triggered by the word "learn" in a prompt — invoke explicitly or via the low-frequency memory_nudge hook (PreCompact/Stop).

Install/update Kronael (bundle + CLI tools rig/udfix/clp/dockbox); bridge global/project CLAUDE.md, skills, and hooks into Codex.

Write or edit SKILL.md, CLAUDE.md, AGENTS.md. NOT for general code (use go/rs/py/ts), mining history (use learn), or researching/codifying public best practice into a skill (use scavenge).

Faithful into-Go transcode: port any source language to Go preserving behavior exactly, proven by differential traces. NOT for idiomatic Go rewrites (use go) or broad engineering runbooks (use software).

Draft a PR description. NOT for commit messages (use commit).

Router for creative artifact generation — CVs, web pages, videos, generative art, diagrams. NOT for code (use language skills), Unicode box diagrams in docs (use diagrams), de-AI-ing prose (use humanize), or generating eval skills (use create-eval).

Data visualization report scripts. NOT for general ETL or scraping (use data).

Hiring evaluation from artifacts, repos, demos, resumes, or interview evidence. Use for "would you hire this engineer?", "top notch?", senior/staff calibration, HFT/systems candidate assessment, and what evidence would change the decision. NOT for production adoption of a project (use cto-eval) or business/demo ROI (use ceo-eval).

Ingest media from a URL (YouTube + most sites) with `yt-dlp`/`ffmpeg` — transcript, audio, video, subtitles, formats. NOT for converting a local file to Markdown (use markdown-converter) or writing video scripts (use create).

Convert existing files (PDF, Office, HTML, data, images, audio, EPub, ZIP, YouTube) to Markdown via `uvx markitdown`. NOT for authoring new prose (use writing) or syncing project docs (use readme).

Code refinement orchestrator. NOT for targeted fixes (use improve).

/continue — goal mode. Recall every interrupted, paused, or abandoned task/plan/goal from this session and resume each. NOT for just pushing the single current in-flight task without stopping (use fin).

Ask the pi coding-agent CLI for a second opinion. NOT for routine lookups (use grep/read/recall-memories). NOT a Claude Agent and NOT codex — this is badlogic's `pi` CLI (a distinct model, complements /codex).

/fin — drive the current goal to completion, no confirmation stops. NOT for a single command (just run it), NOT for resuming interrupted/paused agents or tasks from earlier in the session (use con, the multi-goal recall mode).

Dispatch a background audit for a PROBLEM CATEGORY across the whole codebase, filing each real instance in BUGS.md. NOT for a single known bug (use /bugs), NOT for fixing anything found (record-only, see CLAUDE.md Bug Triage Protocol), NOT for foreground/inline audits (use /dispatch directly).

Makefiles. NOT for shell scripts (use sh) or build configs (use language skill).

Give or take a code review — produce findings (give) or apply them (take); local by default, or a GitHub PR with `gh`. NOT for posting arbitrary PR comments (use gh-comment) or filing issues (use gh-issue).

Rust development. NOT for non-Rust code (use go, py, ts, tsx, or sh).

Bash/shell scripting. NOT for Python utilities (use py) or persistent CLI tools (use cli).

SQL queries and schemas. NOT for ORM-only code (use go, py, ts, or rs).

TypeScript/Node.js. NOT for .tsx (use tsx).

Frontend React/Next.js. NOT for plain TypeScript without JSX (use ts).

Assemble and file a GitHub issue on a repo, with a mandatory approval gate — never posts without showing the exact title+body first. Derived from gh-comment.

Server-rendered HTML with htmx attributes. USE when editing .html templates with hx-* attributes (Jinja, Go html/template, Tera, Rails ERB, Phoenix HEEx) or when the project lists htmx in deps. NOT for React/JSX (use tsx). NOT for plain TypeScript (use ts).

Browser automation. NOT for plain HTTP fetches (use sh or py).

/ans — conversational answer-only mode toggle. NEVER edit files or run shell. /ans end to exit. NOT for stopping mid-task to ask (just answer inline).

/next — park a discovered bug or TODO for later without stopping current work. NOT for filing bugs found during a code audit (use /bugs for that).

/dispatch — launch a background subagent at default model. NOT for tasks the main thread needs results from inline; NOT for model-specific work (use /haiku, /sonnet, /opus, /fable).

Browser automation via the agent-browser CLI. USE when a real browser is needed (login flows, screenshots, page interaction, DOM extraction). NOT for plain HTTP fetches. ALWAYS invoke via Bash tool — this is NOT an Agent subagent type, never pass it to Agent(subagent_type=...).

Generate project-specific service-eval skill. NOT for running existing evals or adversarial audits.

CTO evaluation — technical adoption by default; SLA/code audit on request. NOT for exploit/attack-surface security audit (use hacker-eval).