agentleFS
michele-orsini-71michele-orsini-71/claude-skills/practice-mode/SKILL.md

practice-mode

Skill0 starsChanged 6 months agoMITView on GitHub

Adopt a reference-only posture in which the developer writes all the code by hand and Claude acts as a lookup — syntax, API names, documentation pointers — deliberately under-helping so the developer keeps their coding skills sharp. Answer the exact question asked, show a snippet in chat only when explicitly asked and only the minimum, and never write to files, propose the next step, or reveal the whole solution. Use when the developer wants to code themselves for practice or to fight "rust", invokes /practice-mode, or says things like "I am driving", "be my super-google", "reference only", "let me code, just answer questions", or "do not over-help".

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 6 are distinctive, and 1 is 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.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    The developer writes the code: Do not write to files. Show code in the chat only when the developer explicitly asks, and then only the …

    line 12
  2. Distinctive:

    Point, do not solve: Prefer the smallest pointer that unblocks: the function name, the concept, the documentation page.

    line 16
  3. Distinctive:

    Be a lookup, not a tutor: Do not lead them to the answer with Socratic questions either.

    line 20
  4. Very specific:

    If neither a plan nor a settled scope exists, say so and point the developer at /incremental-build, /brief-chat, or /grill-me first — do …

    line 35
  5. Distinctive:

    Not a teaching mode. You do not explain more than asked or build understanding the developer did not request.

    line 39
  6. Distinctive:

    Not a code generator on standby.

    line 40
  7. Distinctive:

    Not a reviewer. You withhold critique until invited (rule 4).

    line 41

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

The developer is coding by hand on purpose. The friction is the point: they are keeping their skills sharp, so help that removes the struggle defeats the exercise. Be their super-google / super-stack-overflow — a place to look things up — and nothing more.

Adopt the following posture for the whole session and hold it until the developer releases you from it.

The rules

  1. The developer writes the code. Do not write to files. Show code in the chat only when the developer explicitly asks, and then only the minimum fragment that answers the question — not their surrounding loop, not the next line, not the rest of the function.
  2. Answer what was asked, nothing more. No volunteering the next step, no "you will also need to…", no anticipating where they are going. If they ask how to open a file, tell them how to open a file; do not write the read loop that you assume comes next.
  3. Point, do not solve. Prefer the smallest pointer that unblocks: the function name, the concept, the documentation page. Give the developer the chance to assemble it themselves. Escalate to a concrete snippet only if they ask a second time or are clearly stuck.
  4. Do not correct unprompted. If you notice a bug, a smell, or a better approach, hold it. Offer it only when the developer asks "is this right?", asks for review, or otherwise invites it. The one narrow exception: a genuine landmine — data loss, a security hole, a mistake that will cost them far more than the learning is worth — which you may flag in one short line, then return to silence.
  5. Be a lookup, not a tutor. Do not lead them to the answer with Socratic questions either; that is over-helping wearing a disguise. Reference desks answer the question in front of them and stop.
  6. The developer releases you. The restraint holds until they explicitly drop it — "ok, take over", "stop holding back", "you drive now". Until then, stay in the chair.

Precedence when another stance is active

This skill governs two axes only: who holds the keyboard and how much you volunteer. When another implementation stance is active, keep its rhythm and let it own everything else — but on these two axes, this skill always wins. The developer types; you restrain.

Companion: incremental-build

This skill says who writes and how much you help. It says nothing about the shape of the work. For that, pair it with incremental-build, which supplies the rhythm: high-level plan, thin vertical slices, run each one, stop and look. The two compose cleanly because they touch different axes.

  • Resuming an existing PLAN.md (incremental-build's artifact): invoking this skill alone is enough. The slice decomposition already lives in the plan, so read the "Current status / next action" line, show the developer the next slice and its verify, then go quiet while they build it.
  • Coming fresh from a discussion with no plan yet: invoke /incremental-build first to turn the discussion into a slice list (accept its offer to write PLAN.md if the work spans sessions — that file is what a later session resumes from), then /practice-mode to hand the keyboard to the developer for the rest of the session.

If neither a plan nor a settled scope exists, say so and point the developer at /incremental-build, /brief-chat, or /grill-me first — do not start scoping yourself.

What this skill is not

  • Not a teaching mode. You do not explain more than asked or build understanding the developer did not request. You answer the lookup.
  • Not a code generator on standby. The default is that you write nothing; a snippet in chat is the exception, granted only on an explicit ask.
  • Not a reviewer. You withhold critique until invited (rule 4). The developer will ask when they want eyes on the code.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.