learning-review
Skill2 starsChanged 3 months agoMITView on GitHub
Automatically use after a meaningful project milestone, completed experiment, substantial refactor, important bug/failure investigation, new architectural boundary, evaluation result, or security decision when there are transferable legal-engineering concepts worth consolidating. Also use whenever the owner asks what they should learn or understand from recent work. Do not invoke for trivial edits, more than once for the same milestone, or repeatedly within a single session.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- nicolas-diez/personal-ostheir own version, first seen Nov 8, 2026
- swenbo1-web/agents-learningtheir own version, first seen Oct 25, 2026
- rikilamadrid/lorekeepertheir own version, first seen Oct 13, 2026
- rikilamadrid/pathfindertheir own version, first seen Sep 30, 2026
- brianbolze/truffletheir own version, first seen Jun 25, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- nicolas-diez/personal-os★ 4same name
- brianbolze/truffle★ 1same name
- rikilamadrid/lorekeeper★ 1same name
- rikilamadrid/pathfinder★ 1same name
- swenbo1-web/agents-learning★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository21 other files it gives its agents.
- .agents/AGENTS.mdAGENTS.md
- AGENTS.mdAGENTS.md
- .claude/rules/agents.mdAGENTS.md
- evals/AGENTS.mdAGENTS.md
- tests/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- architecture-reviewSkill
- code-reviewSkill
- evaluation-reviewSkill
- explainSkill
- interview-reviewSkill
- security-reviewSkill
- start-projectSkill
- architecture-reviewSkill
- code-reviewSkill
- evaluation-reviewSkill
- explainSkill
- interview-reviewSkill
- learning-reviewSkill
- security-reviewSkill
- start-projectSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. 0 are common, 6 are distinctive, and 7 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Do not give a generic software-engineering lecture.
- Very specific: line 17
Before teaching anything, read docs/learning-log/.
- Distinctive: line 32
When it is empty, pitch from the conversation and from what the repository shows has been built, and do not treat the absence as evidence …
- Very specific: line 76
Write a record to docs/learning-log/ using templates/LEARNING_RECORD.md.
- Distinctive: line 80
Write one only when one of these is true:
- Very specific: line 96
Do not duplicate a definition that belongs in docs/GLOSSARY.md, or an explanation that belongs in docs/concepts/.
- Distinctive: line 100
When a later record contradicts an earlier one — the owner's understanding deepened, or what they believed turned out to be wrong — mark …
- Distinctive: line 111
After trivial edits, routine fixes, or ordinary iteration.
- Distinctive: line 113
When no genuinely transferable concept appeared.
- Very specific: line 122
If the work involved an AI behaviour whose quality is still an impression rather than a measurement, move into evaluation-review before …
- Very specific: line 124
If the review reveals that the next step requires a structural design decision, move into architecture-review when that work begins.
- Very specific: line 125
If it reveals a new confidentiality, permission, data, provider, or side-effect concern, move into security-review.
- Very specific: line 126
If the completed work now forms a credible external story, follow with interview-review — and point it at the learning log, which is where …
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<!-- Generated from ../../../.claude/skills/learning-review/SKILL.md - do not edit this copy.
Edit the source and run `uv run python scripts/sync_agent_files.py`. -->Learning Review
Turn the work that has actually happened in this repository into a concise learning session.
Do not give a generic software-engineering lecture. Teach only concepts that arose naturally from the owner's project.
1. Read what is already understood
Before teaching anything, read docs/learning-log/. It records what the owner has
already demonstrated they understand, what they told you they knew before starting, and
which misconceptions have been corrected.
This is the difference between a review that builds and one that repeats. Without it, every session pitches at the same level: provenance gets explained a fourth time while the thing the project is actually about to need goes unmentioned. Aim just past what the log shows — far enough to be worth the session, close enough to connect to something already held.
Also read docs/GLOSSARY.md, and use its terms rather than synonyms. Consistent
vocabulary is most of what makes a concept transferable to a conversation with an engineer.
An empty log is the normal early state, not a missing file. The bar for a record is deliberately high, so the log stays empty until something has genuinely been demonstrated. When it is empty, pitch from the conversation and from what the repository shows has been built, and do not treat the absence as evidence that nothing is known.
The cheapest way to stop it being empty is criterion 2 below: what the owner already
knows. Someone who has written SQL, run a document review platform, or managed a system
implementation is starting from somewhere quite specific, and a session that discovers that
in week six has wasted five. start-project §0 asks the question in a first session and
writes the record; if you reach this skill and no such record exists — because the project
began before that, or the question went unanswered — ask it now. One question, and a record
of the answer with the depth claimed.
2. Teach the concepts the work has earned
For each important concept:
- explain the practical problem that caused the project to need it;
- explain the concept in ordinary English;
- point to where it appears in the repository;
- explain what problem it solves;
- explain when it would be unnecessary or over-engineering;
- relate it to legal technology or legal AI;
- give one short question or scenario that checks whether the owner genuinely understands the idea.
Useful concepts may include, only when the project has earned them:
- deterministic versus probabilistic behaviour;
- modules and separation of responsibilities;
- schemas and structured outputs;
- provenance and grounding;
- tests versus AI evaluations;
- least privilege and narrow AI tools;
- authentication versus authorisation;
- human approval and side-effect control;
- audit trails;
- idempotency;
- prompt injection;
- data boundaries;
- prototype versus production maturity.
Where a concept deserves a durable write-up rather than an explanation in a transcript,
add a note to docs/concepts/ following the guidance there.
3. Record what was learned, when it was actually learned
Write a record to docs/learning-log/ using templates/LEARNING_RECORD.md. Number it
sequentially from the highest existing file.
The bar is deliberately high, because a log that records everything records nothing. Write one only when one of these is true:
- The owner demonstrated genuine understanding of something non-trivial. Not that they were shown it — that they used it correctly, caught a problem with it, or explained it back in their own words. This sets a new floor for what to teach next.
- The owner disclosed prior knowledge. "I already know how X works." Record it, and record the depth claimed, so no future session spends time re-teaching it.
- A misconception was corrected. These are the highest-value records in the log, because a wrong model of one concept usually predicts trouble with the neighbouring ones.
- The direction of the project shifted because of something learned. Record what changed and why.
Coverage is not learning. Material that was merely explained does not qualify, however well it was explained. Wait for the evidence.
Do not duplicate a definition that belongs in docs/GLOSSARY.md, or an explanation that
belongs in docs/concepts/. The log records that something is now understood and what
that changes; the other two hold the substance.
When a later record contradicts an earlier one — the owner's understanding deepened, or what they believed turned out to be wrong — mark the earlier record superseded rather than deleting it. How an understanding developed is worth more than a tidy final state, and it is exactly the kind of thing worth being able to describe in an interview.
4. Point at what is worth learning next
Identify one or two things worth learning next because the project is beginning to need them. Explain why now is the right time. Do not prescribe a broad curriculum detached from the work.
When not to use this skill
- After trivial edits, routine fixes, or ordinary iteration.
- More than once for the same milestone, or repeatedly within a session.
- When no genuinely transferable concept appeared. A review that manufactures a lesson from work that did not contain one devalues the reviews that do.
- To write a learning record for a concept that was merely explained. If the session produced teaching but no evidence of understanding, teach, and leave the log alone.
Handoff to other skills
Hand off to at most one of these, and note the rest rather than running them.
- If the work involved an AI behaviour whose quality is still an impression rather than a
measurement, move into
evaluation-reviewbefore drawing lessons from it. - If the review reveals that the next step requires a structural design decision, move into
architecture-reviewwhen that work begins. - If it reveals a new confidentiality, permission, data, provider, or side-effect concern, move into
security-review. - If the completed work now forms a credible external story, follow with
interview-review— and point it at the learning log, which is where the "what changed in my understanding" part of an interview answer comes from. - Otherwise return to normal implementation or the next experiment.
The goal is retention and judgement, not proving that the owner can recite terminology.
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.

