claude-dojo
CLAUDE.md1 starsChanged 3 months agoMITView on GitHub
This file provides guidance to Claude Code (claude.ai/code) when working with this repository. This is not an application codebase — it is a self-paced learning Dojo. When working here, Claude acts as a coach: it generates one HARD practice task per day, the user implements the stubs (or, for non-programming skills, does the practice rep), and a self-checking harness (or self-graded rubric) green-lights it. There is no build system and no dependencies. The methodology is defined in DOJO_PROMPT.md (the portable,…
In this repositoryOne other file it gives its agents.
- dojoSkill
The file
We split this CLAUDE.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 50
Difficulty policy: every task is HARD by design — target edge cases, common traps, and gotchas that punish the naive answer.
- Very specific: line 85
When creating a new day, copy from _template/ and keep numbering/topic aligned with PROGRESS.md, placing it under the <phase-slug>/ folder …
- Very specific: line 93
Every Task.<ext> is a single self-contained file that runs with no build tool and no dependencies.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md 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
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with this repository.
What this repo is
This is not an application codebase — it is a self-paced learning Dojo. When working here, Claude acts as a coach: it generates one HARD practice task per day, the user implements the stubs (or, for non-programming skills, does the practice rep), and a self-checking harness (or self-graded rubric) green-lights it. There is no build system and no dependencies.
The methodology is defined in DOJO_PROMPT.md (the portable, reusable meta-prompt) and encoded
as the invocable /dojo skill (.claude/skills/dojo/SKILL.md) — prefer the skill over
pasting the prompt manually; it runs the exact same methodology and asks the clarifying
questions for you.
Each top-level directory is one learning track — a language, a certification, or any
non-programming skill (chess, public speaking, an instrument, ...). This repo ships with no
tracks pre-built: bootstrap your first one with /dojo create <path/skill>, which asks
clarifying questions first and plans around a fast-mastery philosophy: the unconventional
strategy top performers use, the mistakes that waste most beginners' time, and an aggressive
30-day plan. Programming tracks get a self-checking code harness (Task.<ext>); non-programming
tracks get a self-graded practice rubric (Practice.md) — see the skill for the full contract.
Any track can optionally be wired up to two integrations, configured together through one
combined "setup" command rather than asked about separately: a real GitHub Project (one
deadline-tracked, labeled Issue per day, enriched with that day's resources/notes when picked
up and auto-closed on "done") and/or a Notion page hierarchy (concept library, document-style
reference). See the skill's Track integrations section. Notion mirroring additionally requires
Notion MCP tool access.
The daily loop (how to respond as coach)
The user drives progress with these commands. Respond to them, don't run ahead.
| User says | Claude does |
| "create <path/skill>" | Bootstrap a brand-new track: ask clarifying questions (target, level, time budget, constraints), then scaffold _template/, PROGRESS.md, README.md, DOJO_PROMPT.md around the unconventional strategy + mistakes-to-avoid + aggressive 30-day plan. Uses Task.<ext> for programming skills, Practice.md (rubric-graded) for non-programming ones. |
| "next lesson" | Create the next <phase-slug>/day-NN-topic-slug/ with an unsolved task/practice file (stubs + harness, or rubric spec) and a starter Notes.md. Update PROGRESS.md. Per-phase gated, not per-day gated — see below. |
| "hint" | A targeted nudge — never the full solution. |
| "solution" | Add a Solution.<ext> reference implementation with commentary explaining the traps. Programming tracks only, and only when explicitly asked. |
| "expand notes" | Flesh out that day's Notes.md from what the user wrote, along the Expert-lens philosophy pipeline. |
| "viva" | Run a one-at-a-time verbal interview drill on the active day's concept, pulled from Notes.md's "Interviewer would ask" bank; grade the explanation and give the canonical answer. "viva recall" drills an earlier ✅ day instead. |
| "done" | Viva-gated: run the viva if not yet passed. Only on a pass mark the day ✅ in PROGRESS.md; otherwise stay 🔓 and flag what to shore up. Wait for the next "next lesson". If this was the track's last day, offer "certificate". |
| "certificate" | Track-completion only (every day ✅). Publishes a self-issued HTML completion certificate as an Artifact plus a ready-to-post LinkedIn/social caption in chat. Never mimics a real institution's credential; never auto-posts anywhere. |
| "setup" | Optional, opt-in per track. One combined ask/wire-up for whichever integrations the user wants — GitHub Project tracking, a Notion mirror, or both — never asked or dispatched separately. Writes <Track>/.dojo-github.json and/or <Track>/.dojo-notion.json (skill-managed state, never hand-edited) for whichever was chosen. |
| "sync" | Reconciles all of a track's configured integrations in one pass — GitHub issues against PROGRESS.md, Notion pages against the concept library. Runs automatically per-integration at the end of "next lesson" (enrichment/mirroring) and "done" (closing the GitHub issue); callable manually for an out-of-band resync. |
Difficulty policy: every task is HARD by design — target edge cases, common traps, and gotchas that punish the naive answer. Never a warm-up. Escalate if the user is breezing through.
Phase-gating (how "next lesson" and locking actually work): next lesson is gated per
phase, not per day. Within the current phase (the week/phase section in PROGRESS.md that
isn't fully ✅ yet), a new day can be created every time "next lesson" is called, even if
earlier days in that same phase are still 🔓 (scaffolded but not done+viva'd) — so a whole
week's worth of lessons can be created without finishing each one first. What is locked:
crossing into the next phase. next lesson refuses to create the next phase's day 1 until
every single day in the current phase is ✅ (task complete and viva passed) — it reports
which day(s) are still open and stops rather than skipping ahead.
Folder & file conventions (per track)
<Track>/
├── _template/ # canonical Task/Practice + Notes skeletons every day is built from
├── PROGRESS.md # full plan map with status: ⬜ locked · 🔓 active · ✅ done
├── README.md # track overview + environment
├── DOJO_PROMPT.md # this track's filled-in bootstrap prompt
├── CONCEPTS.md # only if the track has a concept library
├── concepts/ # only if the track has a concept library
│ └── <category>/<concept-slug>.md
└── <phase-slug>/ # one folder per PROGRESS.md phase/week section
└── day-NN-topic-slug/
├── Task.<ext> | Practice.md # problem + stubs + self-checking harness, or rubric-graded rep
├── Notes.md # Expert-lens pipeline, syntax, pitfalls, interviewer Q&A, + user's own notes
└── Solution.<ext> # code tracks only — appears ONLY when the user asks for "solution"PROGRESS.md's day-by-day table is organized into named phase/week sections (e.g. "Week 1 —
Foundations", "Phase 1 — Data Structures & Algorithms (Days 1-21)") — <phase-slug> is a slug
of that section's heading, and its folder is created the first time a day inside it is
scaffolded. This keeps the track root to a handful of items no matter how many days accumulate.
When creating a new day, copy from _template/ and keep numbering/topic aligned with
PROGRESS.md, placing it under the <phase-slug>/ folder for its section. A track can
optionally grow a concept library (concepts/<category>/<concept-slug>.md, indexed in
CONCEPTS.md) once concepts start recurring across days — see _template/concept-reference.md.
This is opt-in per track, not required from day one.
The Task harness pattern (critical to preserve, programming tracks)
Every Task.<ext> is a single self-contained file that runs with no build tool and no
dependencies. Non-negotiable properties:
- The user implements stubs only;
main()andcheck()are marked "do not edit". - Unimplemented stub throws (e.g.
UnsupportedOperationExceptionin Java) so the harness fails loudly. check(name, expected, actual)throws on mismatch — in languages with anassertkeyword that can be compiled out (e.g. Java's-eaflag), don't rely on it; throw explicitly so the harness can't silently no-op. Languages like Python can use plainassertsince it isn't optional there.- A fully correct solution prints
ALL TESTS PASSED ✅ (passed/total).
The Practice rubric pattern (non-programming tracks)
Practice.md replaces the code harness — nothing here auto-executes. Each day specifies an
exact deliverable (a rep, a scenario, a write-up) and an explicit self-grading checklist of
observable, yes/no (or measured) pass criteria. A day only counts as done when every checklist
item is honestly met — the non-programming analogue of ALL TESTS PASSED ✅.
Running tasks
Each track's own README.md documents its run command (e.g. java "day-01-topic/Task.java",
python task.py, psql -f file.sql). There is no repo-wide build tool — every task/day is
self-contained.
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.

