dotfiles

Guide for writing or refining prompts for Claude or GPT-5.6 Sol, distilled from each vendor's official best practices. Use for new prompts, debugging an existing prompt, or "prompt engineering", "system prompt", "Claude/GPT/OpenAI prompt". Accepts arg: "claude" or "gpt".

The ONLY way to call GPT (a.k.a. Codex), Grok, GLM, Kimi, DeepSeek, or MiMo. Use whenever the user wants to ask, delegate to, or get a second opinion from GPT, Grok, GLM, Kimi, DeepSeek, or MiMo. Do NOT run the codex, grok, glm, kimi (km), deepseek (ds), or mimo (mm) CLI directly — from the main agent or a subagent; always use this skill's relay call command. Triggers on "ask/have/send to/get/delegate to gpt/codex" or the same with "grok"/"glm"/"kimi"/"deepseek"/"mimo", "second opinion", "relay".

Dispatch multiple independent agents to answer the SAME complete question from different analytical lenses, then synthesize. Reach for this by default, without waiting to be asked, when redundant cross-model judgment could change the decision: ambiguous architecture/design tradeoffs, high-stakes or hard-to-reverse changes, competing root-cause hypotheses, or failure-mode-sensitive reviews — and whenever you would otherwise spawn 2+ independent reviewers for one question. Skip for trivial lookups, deterministic transforms, routine small edits, mechanical syncs, and single-correct-answer tasks. With no leading number, autonomously decide N and gpt-pro from the question every time (N=1 is the bottom-rung anchor, never a lazy default); with an explicit number (e.g. `prism 2 1`), honor it verbatim and skip auto-sizing. There is no reasoning-effort knob. Scale above the anchor only when decision risk justifies the 8N-agent cost.

Drive a goal through review→fix iteration by composing goal-elicit, goal-drive, and multi-model review (prism, Claude-only). A thin, modeless, stepped loop — one phase per invocation (elicit → spec-review → drive → review → you pick which fixes apply → fix → re-review) — with state on disk so it resumes after interruption or compaction. It auto-handles the ~83% of out-of-scope/unmapped findings (you never see them) and surfaces only the small actionable batch to confirm — never blind-applies (unsafe). Use for "loop this to done with review", "close the review-fix loop", "goal-loop". Review needs Claude (prism), degrades off-Claude. Interactive by default; **`--auto`** runs it unattended/headless under Claude Code's native `/goal` — both gates become fail-closed policies (oracle-gated safe-subset auto-fix + a deferred morning queue), never blind-apply. Does NOT interview (that's goal-elicit). Skip for one-off edits and lone reviews.

Send a prompt to ChatGPT Pro Extended via gpt-pro-relay on macmini — over SSH from any other machine, or directly when invoked on macmini itself. Use for "ask gpt-pro", "send to gpt-pro", "use gpt-pro", "Pro Extended take", "ask the deep model", or "second opinion from chatgpt pro". The wrapper (not the caller) polls through flaky networks; the agent fires one backgrounded call and waits for the notification.

Audit Codex/OpenClaw skills: loaded roots, duplicate skills, unused skills, prompt-budget costs, compact descriptions.

Read arxiv papers via their TeX source for full-fidelity math, tables, and figures. Use whenever the user shares an arxiv URL or paper ID and wants to discuss or understand the paper — even without "read". Triggers on arxiv URLs (abs, pdf, html), bare IDs like "2401.12345", or "explain this paper". Do NOT use for note generation (use note-gen).

Autonomous experiment loop faithful to Karpathy's autoresearch. Set up a branch, establish a baseline, then loop forever: edit, commit, run, measure, keep or revert. Use when asked to "run autoresearch", "optimize X in a loop", or "start experiments".

Claude Code only. Delegate trusted, scoped implementation to `codex exec --yolo` — building from a spec, refactors, mechanical migrations, bug fixes with a repro, tests, CI/dependency/tooling, bulk code reading — while Claude writes the spec, picks `-e` effort, and reviews every diff. Auto-routes when enabled; also `/codex-first <task>`. Keep design, ambiguous specs, tiny edits, secrets/session-only tools, and all commit/push/GitHub ops in Claude. Unsandboxed: the review gates the diff, not the machine.

Maintain a durable, version-controlled manifest of a project's recurring Claude Code cron jobs (the /loop + CronCreate fleet) and re-arm them after they are lost. Claude crons are session-only — durable:true is ignored, CronList truncates prompts, and recurring jobs auto-expire after 7 days — so the live state is unrecoverable; this skill keeps crons/*.cron as the source of truth, renders CRONS.md, and emits the exact CronCreate/CronDelete calls to re-arm. Use on session start, after a compaction, reboot, or 7-day expiry to check whether your crons are still armed and restore them; when you run many /loop crons in one project; or on "crons", "manage my crons", "re-arm crons", "are my crons still running", "cron manifest". It is a preparer, not an actuator: it never calls the Cron* tools and never claims a cron is "armed" (no false assurance) — you run CronList and execute the emitted calls. Do NOT use for a one-off cron (use CronCreate directly) or a cache heartbeat (use keep-warm).

Remove AI-generated slop from code changes. Checks the diff against main and cleans up unnecessary comments, defensive checks, single-use variables, redundant casts, and style inconsistencies introduced by AI. Use when the diff looks bloated after a coding task, or when user says "clean up", "remove slop", "deslop", "simplify the diff", or invokes "deslop". Do NOT use for deliberate refactoring or feature changes.

Re-layer the previous response (or a named file) into a fast-to-skim form: answer first, a review-map or key-findings short version, action-changing blockers only, detail pulled on demand. Restructures for fast reading WITHOUT dropping substance. Use when a reply came out dense, or the user says "tl;dr", "too long", "wall of text", "make this skimmable", "digest that", or invokes "digest". Do NOT use to summarize-away detail, on already-short replies, or to clean code (use deslop).

Maintain one plain-English STATUS.md at the working-directory root — a living executive briefing a non-technical stakeholder grasps in ~30 seconds: what the run is trying to do, how it's going, what it's found, and whether to worry (no jargon, no logs). Use for work the user kicks off and walks away from, or that spans many steps or sessions unwatched: autonomous research/experiment runs, hyperparameter sweeps, long migrations, overnight or multi-day jobs, "go run this and keep me posted" — and on resume in a directory that already holds a STATUS.md for an active run. Also trigger on explicit asks: "status report", "executive summary", "CEO update", "I keep losing track of progress", or "exec-status" by name. Do NOT use for in-session work you're watching, single edits or bug fixes, quick Q&A, or work with no durable working directory. Zero-context-readable — complements todo (operator checklist), autoresearch (experiment state), and digest (reply reformat); does not replace them.

Drive an existing goal artifact to verified done — a GOAL.md contract, a JSON checklist, or a phased design doc (produced by goal-elicit, or hand/script-written). Executes modelessly: process the next unit, verify acceptance, update state, optionally commit, repeat — stopping only on exceptions (blocked unit, scope drift, unclear authority, an irreversible action). Use for "drive this", "execute this plan/checklist", "process this in batches", "run the phases", "carry this to done". Does NOT interview — if the goal is unclear or no artifact exists, run goal-elicit first. Skip for one-off edits and factual questions.

Interview the user and write a verifiable goal artifact — a Goal Contract (GOAL.md), a JSON checklist, or a phased design doc, chosen by triage. Never executes the goal or invokes downstream tools — it writes the artifact and stops; hand off to goal-drive (or any agent) to execute. Use for "clarify what I want", "define done_when/acceptance criteria", "make this unambiguous", "plan this out", XY-problem requests, or untestable success. Skip trivial edits, factual questions, routine commands, and code review.

Conducts mock technical interviews for AI/ML topics. Use when the user wants to practice for interviews, test understanding, or be quizzed on ML concepts. Triggers on "interview me", "quiz me", "mock interview", "practice questions", or "interview". Supports topic-based or material-based questioning with multiple difficulty levels.

Schedule a recurring heartbeat that keeps Claude Code's 1-hour prompt cache from expiring during long unattended tasks. Use when starting a multi-hour run, or when the user says "keep cache warm", "warm cache", "stay warm", "heartbeat", "keep alive", or invokes "keep-warm". Pass "stop", "cancel", or "off" to cancel an active heartbeat.

Commit and push changes to the remote ("commit and push", "push this", "ship it", "save and push", "push"). The agent inspects the diff and picks single-commit or atomic mode automatically: atomic when changes span unrelated units (mixed categories, multiple distinct areas), single commit for one focused change. Do NOT create a pull request.

Interact with X (Twitter) via xurl, the official X API CLI: post, reply, quote, delete, search, read timelines/mentions, like, repost, bookmark, follow, DM, upload media, raw v2 endpoints. Triggers on "post to X", "tweet this", "reply to [X post]", "search X for", any x.com/status URL, or "xurl". Do NOT trigger on bare "X" meaning the letter, a variable, or an unknown.