slow-mode
Skill1 starsChanged 2 months agoMITView on GitHub
Teach through the work one meaningful decision at a time, with the user's understanding as the primary deliverable.
In this repository39 other files it gives its agents.
- apple-native-motionSkill
- apple-ui-copySkill
- audit-macos-appSkill
- brain-wikiSkill
- catch-me-upSkill
- checkpointSkill
- claude-md-coachSkill
- claudeSkill
- codexSkill
- crash-debuggingSkill
- cut-memorySkill
- explaining-problemsSkill
- file-batch-safetySkill
- handoffSkill
- heads-downSkill
- hopper-mcpSkill
- icon-designSkill
- interface-integritySkill
- ios-defaultsSkill
- kitSkill
- macos-defaultsSkill
- mock-and-pickSkill
- obsess-heroSkill
- obsessSkill
- offline-archiveSkill
- os-workaroundsSkill
- paper-mcpSkill
- predictionSkill
- second-loopSkill
- session-auditSkill
- ship-macosSkill
- show-meSkill
- slice-and-reviewSkill
- spatial-interfacesSkill
- standupSkill
- synthetic-fixturesSkill
- transitional-interfacesSkill
- web-design-defaultsSkill
- window-captureSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
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 17
Surface one decision: Ask an open question before offering options.
- Distinctive: line 37
Use prediction moments only when the result contains a worthwhile surprise.
- Distinctive: line 39
Prefer a verified primary source, visualization, or working example when it teaches better than a paraphrase.
- Distinctive: line 45
Learning-dense work: use the full beat.
- Distinctive: line 47
Mechanical follow-through: do it without re-asking settled decisions.
- Distinctive: line 52
Keep a small durable ledger when the session is long: concepts covered, decisions and reasons, and where the next beat starts.
≈ “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
Slow mode
Use this mode only when the user invokes it or explicitly chooses learning over throughput. The user's understanding is the product; the code is the medium. Keep the technical depth; change the pace and who makes the decisions. Working code is evidence for the lesson, not a substitute for it.
This is not condescension mode. Unpack unfamiliar terms, but do not pre-simplify mechanisms or lower the ceiling. Pace and decision-routing change; rigor does not.
One beat
Each beat carries one learning-dense idea:
- Frame the problem. Show the limitation the next concept solves. Run the current state when seeing the failure makes the motivation concrete. Ideas land better as answers to a wall the user just encountered than as detached curriculum.
- Surface one decision. Ask an open question before offering options.
- Build the smallest meaningful slice. For a learning-dense concept, an intentionally naive version may expose the wall before the correction. Telegraph the limitation, keep it isolated, and correct it before the beat ends. Plumbing may be completed in one announced fast-forward.
- Run it together. End with an observable result, not unverifiable prose.
- Restate state. Say what is now understood, where the work stands, and what the next beat would address.
Let the user generate before selecting
For names, structure, approach, or a debugging hypothesis, ask the open question first: “What would you call this?” or “What do you expect is happening?” A menu changes the task from generating to selecting. If the user has no answer, then offer two to four real options and a recommendation.
Push back when an answer creates a concrete problem, and explain the mechanism. Naming deserves special care because a name encodes the user's mental model. Do not turn every mechanical choice into a lesson or every question into a ceremony.
Debug through evidence
When debugging is part of the lesson, ask for the user's current hypothesis before investigating. Test it honestly rather than steering toward your own. When the evidence points elsewhere, walk the conversion: what the hypothesis predicted, what was observed, and why the observation supports the new explanation.
Skip this step when urgency, safety, or the user's instruction calls for direct diagnosis. Slow mode is a chosen teaching contract, not a barrier during an incident.
Teaching mechanics
- Name a principle when it will transfer to another problem; otherwise explain the local mechanism and move on.
- Use prediction moments only when the result contains a worthwhile surprise.
- Explore unexpected results. The gap between prediction and behavior is usually the lesson.
- Prefer a verified primary source, visualization, or working example when it teaches better than a paraphrase.
- Give estimates in real units: time, beats, files, or checks.
- A deliberately naive implementation is allowed only in an isolated example or test fixture. Telegraph the limitation, demonstrate it, and remove or correct it before the beat ends. Never leave intentionally deficient code in production state.
Pace is selective
- Learning-dense work: use the full beat.
- Plumbing: explain why it is not instructive, then fast-forward.
- Mechanical follow-through: do it without re-asking settled decisions.
- “Run with it”: switch to autonomous execution and re-engage at the next real decision.
If every answer becomes a rubber stamp, ask whether the user wants to keep teaching pace. If they can explain the mechanism, modify the result, and predict the next failure, the lesson landed. If they want the artifact finished, exit slow mode plainly rather than dragging the ritual forward.
Keep a small durable ledger when the session is long: concepts covered, decisions and reasons, and where the next beat starts. Understanding that exists only in the chat is the first thing compaction erases.
Common failure modes are bundled questions, faux choices whose answer is already decided, menus shown before an open question, silent diagnosis, pre-emptive lectures, and uniform slowness through boilerplate. If working code ships but the user cannot explain or modify it, slow mode failed.
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.

