mstack help

Answer the user's question about mstack, hand them a prompt they can send, and name the file the answer came from. For a help question, do not start the work. The user asked how, and an mstack run spends real tokens, so let them send the prompt.

A message that asks for work, such as "use mstack to fix this bug", is not a help question. Read the meta-mode skill, do the work under it, and mention once that invoking meta-mode keeps the work in this style.

This file maps questions to the skills and docs that hold the answers. Those files own the details. Read the skill you route to before you quote it, and trust it when it disagrees with this map. Skills may be installed in isolation, so name another skill in plain text. The README.md and docs/ paths below belong to the repository, and the default install does not ship them. Read them when the checkout is at hand. Otherwise give the user the public copy: https://github.com/3metaJun/mstack/blob/main/ followed by the path. The guide artifact installs docs/guide/ locally.

Find out what they need

Infer the need from the message and the conversation. A named situation, such as "which skill reviews a diff?", goes straight to its section. If the need is still unclear, ask one multiple-choice question, then answer only the section they pick: get set up, start a task with meta-mode, pick a skill, fix a run that went wrong, or make mstack their own.

Check the state that changes the answer, and mention it only when it does:

• No ~/.config/mstack/models.json means /setup-mstack has not run for this user, so roles run on inherit-parent. The scripts read that path under the home directory on every platform by default, and their --file option overrides it.
• No verify-* skill or other app harness in the project means agents have no scripted way to drive the app. Mention create-verification-skill when the question is about proving a change works.
• Which harness the user runs decides the command names in the next section.

Get set up

1. Install for the harnesses in use, from README.md:

   npx @3metajun/mstack --harness all
   npx @3metajun/mstack --harness codex,claude
   npx @3metajun/mstack --harness all --skill tdd,diagnosing-bugs

   --dry-run prints the plan. An existing skill directory is kept unless --replace is given, which moves it into a backup directory. --artifact adds the optional agents, meta-mode-tools, guide, or session-context files. --project /path/to/repo copies skills into that repository's own skill directories instead, for harnesses that read project skills only, such as agents started in per-thread worktrees. It is skills only, so it is refused together with --artifact. Node.js 18 or newer is required.

2. Run /setup-mstack. It asks for a reasoning budget, maps a model to each role, and writes ~/.config/mstack/models.json, the portable file every mstack skill reads. New sessions pick it up.
3. Start a real task with /meta-mode, a goal, and a check that can pass or fail.

Installing changes nothing until the user invokes a skill. Offer to word their first prompt with them. The guide page docs/guide/01-setup.md has the walkthrough.

Harness differences that change the answer:

• Claude Code can also load mstack as a plugin. Its skills are namespaced, so the command is /mstack:meta-mode. The installer gives bare names such as /meta-mode. Pick one route per machine, because both loading duplicates the skills.
• Codex, OpenCode, pi, and Grok share one copy in ~/.agents/skills/. Claude Code and Antigravity have their own directories. docs/harness-adapters.md lists them.
• Claude Desktop chat, Cowork, and claude.ai take one uploaded zip per skill from npm run pack-claude-skills. Uploaded skills are isolated, so sibling links become plain text, and meta-mode playbooks and tools are not reachable there.
• T3 Code lists whatever the provider finds, so installing for the provider is enough. See docs/t3code.md.

If cost is the worry, say where the tokens go. mstack spends extra tokens on subagents and review panels. Rerun /setup-mstack with a smaller budget or cheaper models. A role set to inherit-parent runs on the chat's own model. A shorter reviewer list runs fewer subagents, one per entry. Save meta-mode for work that needs rigor.

Start a task with meta-mode

meta-mode matches the task to a playbook, copies the playbook's steps into the todo list, and runs other skills as the steps need them. A skipped step stays in the list as skip: <reason>. A good prompt states the goal and how to tell it is done. It does not list skills, because a hand-written sequence tends to drop steps the playbook would keep.

/meta-mode the export writes duplicate rows when a retry lands mid-run. repro first, then fix and verify.

The mode stays active for the current task. It does not apply to a casual turn or after the user opts out.

Pick a skill

The default answer is meta-mode, which runs most of the others when its steps need them. Name a skill directly when the user wants more or less than the playbook gives. Read the skill before you recommend it, and give one example prompt.

If a skill directory is missing from the table, read its frontmatter and route by its description. principle-* skills are covered below.

Close calls:

• how explains what the code does. why explains the reasons. teach runs both and explains the result plainly.
• arena gives every worker the same brief and merges the best parts. swarm splits work into slices or a race and returns one report.
• interrogate reviews the diff. blast-radius looks for breakage outside it and proves the one fact that makes the change safe.
• figure-it-out designs one rigorous run. The Orchestrate playbook runs a program of many PRs. The Autonomous run playbook drives one task to a finish condition.

Playbooks and principles

Playbooks are step lists inside meta-mode, not skills, so they are not invoked by name. Describing the task picks one. The Playbooks section of the meta-mode skill (skills/meta-mode/SKILL.md) lists every playbook and when it applies. Common phrasings:

• "babysit this pr" or "check on pr 123" runs Babysit. It drives the PR to merge-ready and stops there.
• "land the stack" runs Shipping. "take over this branch" runs Session pickup. "pause safely" runs Pause safely.
• "prototype" or "sketch it to decide" runs Prototype. For a multi-phase plan, ask for a plan and the Multi-phase plan playbook writes it without implementing.

Principles are one-rule skills that meta-mode reads and cites in its replies. The user steers with the names, as in "apply prove it works. show me the real output." The guide page docs/guide/08-principles.md lists them.

Fix a run that went wrong

Make mstack my own

• automate-me drafts a personal mode skill from the user's history. reflect turns a session's lessons into skill edits the user approves. /meta-mode write a skill for <workflow> runs the authoring playbook, and the Eval playbook tests a skill change.
• docs/guide/09-make-it-yours.md covers both. Fix a misbehaving skill in its own PR, not inside the feature work where it went wrong.

Reply

Lead with the answer. Give at most one example prompt in a code block, then name the file it came from. Keep it short unless the user asked for the whole map.