compression-challenge
Skill0 starsChanged 5 days agoMITView on GitHub
Ask the learner for a concept three times, in full, as an analogy and as the shortest version that still holds, list the relations that survived and the ones that did not, and record the attempt; use it when the learner wants to test whether an idea survives being shortened, and it writes a file, so run it only when asked.
In this repository35 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- llms.txtllms.txt
- adaptive-scaffoldSkill
- calibration-coachSkill
- counterfactualSkill
- defuddleSkill
- dependency-checkSkill
- error-recoverySkill
- json-canvasSkill
- memory-palaceSkill
- misconception-drillSkill
- mistake-reviewSkill
- new-concept-noteSkill
- new-trackSkill
- no-ai-checkpointSkill
- obsidian-basesSkill
- obsidian-cliSkill
- obsidian-markdownSkill
- pretestSkill
- process-paperSkill
- process-webSkill
- quizSkill
- record-attemptSkill
- recovery-after-inactivitySkill
- retention-auditSkill
- review-queueSkill
- run-checksSkill
- schedule-reviewSkill
- spoken-mockSkill
- teach-the-agentSkill
- transfer-labSkill
- vault-searchSkill
- visual-reconstructionSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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
- Distinctive: line 19
Ask for the three versions, in this order, and do not interrupt:
- Distinctive: line 25
Wait for all three before saying anything about them.
- Distinctive: line 29
Ask the learner to put the missing relation back in their own words.
- Very specific: line 48
Keep the three versions in the compression note of the track.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Tutor mode: the learner attempts first and the agent never interrupts. The agent never rewrites the learner's versions and never writes the learner's notes or answers. Hints come one step at a time and only after an attempt: hint, strong hint, partial example, answer. Confidence from 1 to 5 is asked before the reveal. The agent never grades an attempt the learner did not confirm.
The rule of this exercise: a short version that keeps the causal relations and drops the examples is a good compression. A short version that drops a relation is not a summary, it is a different claim.
- The learner picks a concept they have studied. Read the concept note in
20_Conceptsand the sources it cites in40_References. Note the relations of the concept: what causes what, what depends on what, and what breaks when a part is missing. Ask for the three versions, in this order, and do not interrupt:
- the full version, as the learner would explain it from the notes;
- the analogy, the same idea in terms of something else;
- the shortest version that still holds.
Wait for all three before saying anything about them.
- List which relations of the concept survived in the shortest version and which did not. Use the concept note and its sources to decide. Name each relation in plain words. Only list what is missing. Do not write a shorter version for the learner and do not edit the words they used.
- Ask the learner to put the missing relation back in their own words. One relation at a time.
- Record the attempt only after the learner confirms the result. The mode is
explain, because the exercise asks for the mechanism and not only the definition.
python3 -I 90_System/Scripts/record_attempt.py --objective <id> --item <id> --mode explain \
[--correct|--incorrect] --confidence <1-5> [--probability <0-1>] [--hints <n>] \
[--assistance <level>] [--grading <method>]- A compression that loses a relation is not a wrong answer, it is a different claim. So it goes
to the mistakes log and not to a card about the definition. Ask the learner for the words of
the relation they put back, then add the entry to
50_Review/Mistakes.mdin the format that file describes, with the kind of errorreasoningand the AI usenone:
YYYY-MM-DD, [[Concept_Name|Concept name]]: I answered X. The answer is Y. I went wrong because Z. Kind: reasoning. AI: none. Open.- Keep the three versions in the compression note of the track. If there is no note yet, copy
90_System/Templates/Compression_Challenge.mdinto10_Tracks/<NN>_<Track>/practice/and fill the Date, Concept, Relations kept and Relations lost columns. The three version columns stay in the learner's words. - Show the learner the line that was written. Never edit or delete a line of the ledger.
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.

