hq-mode
Skill1 starsChanged 15 days agoMITView on GitHub
/hq-mode: take on a whole task the lab's way, such as an analysis, a cluster run, a figure, a paper or grant section, a talk, an experiment plan or landing tickets. Reads the working-style page first, then follows a playbook. Never submits Slurm work, spends Edison credits or merges an unreviewed PR without the user's yes, whatever the session says. User-invoked only.
In this repository36 other files it gives its agents.
- AGENTS.mdAGENTS.md
- broSkill
- codebase-designSkill
- code-reviewSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- implementSkill
- implement-specSkill
- improve-codebase-architectureSkill
- ketchupSkill
- lab-containersSkill
- lab-edisonSkill
- lab-hpcSkill
- lab-jupyterSkill
- lab-setupSkill
- prototypeSkill
- prSkill
- researchSkill
- retroSkill
- setup-matt-pocock-skillsSkill
- tddSkill
- teachSkill
- technical-writingSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- unslopSkill
- unslop-zhSkill
- wait-whatSkill
- wayfinderSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Read references/working-style.md in full before the first tool call: asking, autonomy, delegation, the reply, the user's documents and the …
- Distinctive: line 91
figure-it-out: no playbook fits. Design one, then run it like an experiment.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
hq-mode
Read references/working-style.md in full before the first tool
call: asking, autonomy, delegation, the reply, the user's documents and the lab defaults. Its
always-pause list outranks every session override ("headless", "land it", "going to bed") and
every principle: an sbatch outside an approved run plan waits for the user's yes, and "close the
issue" never merges a PR nobody reviewed. A repo's own AGENTS.md wins over that page. Then
match the task below.
Triggers
A skill written with a leading / is one the user types: suggest it, or read its SKILL.md
and follow it once the user has asked for that step. A bare name loads when you reach for it. Never build inside a
grill, spec or research session: the user types the next step.
| When | Go to |
| A question for the user | The asking rules first. A fact is never the user's to answer |
| An idea in a repo | /grill-with-docs; with no repo, /grill-me. Too big for one session: /wayfinder |
| A design question code can settle | prototype, bridged by /handoff when it needs its own folder |
| The idea is settled | /to-spec, /to-tickets, then the land-the-tickets playbook: /implement per ticket or /implement-spec for all, each driving tdd and code-review; pr writes the PR body |
| A build just ended | /retro |
| Something broken | diagnosing-bugs |
| Raw issues arriving | /triage |
| Reading legwork | research |
| The answer is in someone else's head | /to-questionnaire |
| Moving to a new session, folder or person | /handoff |
| Time for upkeep | /improve-codebase-architecture; a module's shape, codebase-design; the words, domain-modeling |
| A cluster, ssh, Slurm, a GPU | lab-hpc before the first ssh. Notebooks: lab-jupyter. Containers: lab-containers |
| Edison | Only when the user types /lab-edison |
| Prose a person reads | unslop as the last pass, or unslop-zh when the output contains Chinese. Docs, READMEs, PR bodies, protocols: technical-writing. Skills and agent docs: writing-for-agents |
| A number you measured or will report | The explain-the-number principle; for a timing, references/benchmark-checklist.md |
| Work the user steps away from | The checks it rests on, per references/verification.md; a decision trail per references/show-me-your-work.md |
| A correction from the user | Write it at once where it lasts: the repo's AGENTS.md, the skill, or memory. Then ask what would stop it recurring |
| The end of any playbook | The close-the-session playbook |
Principles
references/principles.md indexes the principles, each with when it applies. Name each principle that shaped a decision in the reply, and only after reading its page in full.
Playbooks
Copy the matched playbook's steps into the todo list, verbatim, before any task todos. With no
todo tool, list them as a checklist in your first reply and give each step's state, done or
skip: <reason>, in your last. A step you skip stays with skip: <reason>. Each lives at
references/playbooks/<name>.md.
Science
- analysis: a question answered from data, ending in a checked number or figure.
- cluster-run: a pipeline or job array on a lab cluster.
- method-research: explain or compare an experimental method from its papers.
- experiment-plan: a plan, proposal page set or timeline for bench or compute work.
- figure: a data figure, schematic or illustration.
Writing
- paper-section: a manuscript section, abstract, cover letter or co-author round.
- grant-section: a proposal section, a form or a CV within word limits.
- talk: a deck, its script, or a recorded talk.
- review-a-document: review the user's own document without editing it.
- message: polish a message the user drafted.
- docs-page: a page on a repo's docs site, its diagrams included.
Code
- land-the-tickets: build a set of tickets through subagents and land them.
- improve-a-metric: one change, one measurement, keep or revert.
- refactor-analysis-code: pin the outputs first, prove them unchanged after.
- investigation: a read-only question. Change nothing.
Any
- long-run: a task driven to a checkable predicate without stopping.
- close-the-session: nothing open when the session ends.
Fallback
- figure-it-out: no playbook fits. Design one, then run it like an experiment.
Attribution
The shape and several playbooks come from pstack's poteto-mode and figure-it-out (MIT,
cursor/plugins), the flow in the trigger table from mattpocock/skills' ask-matt (MIT). Both
licenses are in the repo's upstream/ folder.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.