capture-learning
Skill1 starsChanged 5 months agoApache-2.0View on GitHub
Use when a real outcome just landed and the lesson should be banked so the OS stops repeating mistakes - an eval result, a customer reply, a metric move, a failed or won launch, a Share-of-Model move - when the founder says "we just learned", "that worked", "that failed", "log this", "note this for next time", "remember this", or after any real result. Appends a dated, sourced lesson to the right one of four KB files (PRINCIPLES, DOS-AND-DONTS, PLAYBOOK-LESSONS, MOATS-LEDGER); append, never overwrite. Skill before: a result that taught nothing and gets repeated. After: a dated sourced lesson the next founder inherits.
14 other projects use a version of this.
Adopted by
Most recently taken up by
- callumflack/skillstheir own version, first seen Nov 8, 2026
- RoyoTech/royo-learntheir own version, first seen Nov 8, 2026
- netgrade-digital/shared-agentstheir own version, first seen Nov 1, 2026
- DentalEdge-Solutions/claude_codetheir own version, first seen Nov 1, 2026
- dsvid/claude-skillstheir own version, first seen Oct 8, 2026
- hussainweb/skillstheir own version, first seen Sep 18, 2026
Other versions14 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 14
- NVIDIA/multi-storage-client★ 91same name
- Kastalien-Research/thoughtbox★ 64same name
- PIXARTSeu/Synapse★ 9same name
- RoyoTech/royo-learn★ 7same name
- netgrade-digital/shared-agents★ 5same name
- hussainweb/skills★ 2same name
- impactbrussels/FounderOS★ 2same name
- subarnasaikia/kaggle-skills★ 2same name
- callumflack/skills★ 1same name
- DentalEdge-Solutions/claude_code★ 1same name
4 more not listed
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 repository27 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/ai-native-os.mdcCursor rule
- agentic-build-loopSkill
- ai-assisted-salesSkill
- apply-ai-native-modelsSkill
- architect-before-codeSkill
- cognitive-architecture-reviewSkill
- compliance-readinessSkill
- customer-discovery-that-doesnt-lieSkill
- design-a-loopSkill
- design-the-mvaSkill
- eval-and-safety-harnessSkill
- frame-the-hypothesisSkill
- gateway-agent-opsSkill
- geo-contentSkill
- hypothesis-miningSkill
- map-the-terrainSkill
- measure-ai-native-pmfSkill
- moat-strategySkill
- pay-down-agentic-debtSkill
- red-team-the-agentSkill
- secure-the-connectorsSkill
- self-healing-fallbacksSkill
- share-of-model-auditSkill
- start-hereSkill
- write-the-claude-mdSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 14 edited versions we read keep these.
1
Distinctive
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
- Very specific: line 23
Use the PLAYBOOK-LESSONS format: a one-line lesson as the heading, then Context, What happened, Lesson, Source.
- Distinctive: line 36
Add below the existing entries.
“Common” is measured: kept by at least 6 in 10 of the 14 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
Capture Learning
The OS gets smarter only when real outcomes are banked. A skill that ran once and taught you nothing was a waste of the run; the same mistake made twice is the OS failing at its one job. This skill takes a thing that actually happened, an eval score, a customer reply, a number that moved, a launch that died, and writes it down as a dated, sourced lesson the next founder inherits. Append, never overwrite. Memory you can edit away is not memory.
The method
Full append protocol, what is worth capturing, the file-routing table, a worked KB entry,
anti-patterns, and a copyable template: references/learning-loop-method.md.
Step 1 - Capture the real outcome
Write what happened, not what you hoped. The eval scored 0.71, not "the agent works well". The customer asked for a refund, not "mixed signal". A guess dressed as a result poisons the file for everyone downstream, so if you cannot point to the run, the reply, or the metric, you are not done capturing.
Step 2 - Write the dated lesson
Use the PLAYBOOK-LESSONS format: a one-line lesson as the heading, then Context, What happened, Lesson, Source. State the lesson so the next founder avoids the mistake, not so you feel resolved about it.
Step 3 - Append to the right file
Pick the home by shape of the learning:
| The learning is... | Append to |
| A durable rule that should govern every design | PRINCIPLES.md |
| A concrete tactic, one move and its reason | DOS-AND-DONTS.md |
| A dated outcome from one real run or case | PLAYBOOK-LESSONS.md |
| Evidence about what compounds or what trust you earned | MOATS-LEDGER.md |
Step 4 - Append, never overwrite
Add below the existing entries. Past lessons stay even when later ones sharpen them; the record of being wrong is itself the lesson.
Output
- One logged lesson, dated and sourced, appended to the correct knowledge-base file.
Constraints
- Append, never overwrite. The history is the asset.
- A real outcome, not a guess. No run, reply, or metric means nothing to bank yet.
- Stay theme-agnostic; the lesson must read true for a founder in any sector.
Dictionary
the learning loop · a demo is not evidence
Copy-paste version
For non-coders: paste into any chatbot (Claude.ai, ChatGPT, Gemini).
Act as my AI-native learning archivist. Here is a real outcome I just had: [WHAT HAPPENED - the
actual eval score, customer reply, metric move, or failure]. Stage it came from: [DOMAIN / stage].
1. Restate the outcome in plain fact, stripped of hope. If I gave you a hope, not a result, say so.
2. Write it as a dated lesson in this exact shape:
### [today's date] - [one-line lesson]
Context: what I was attempting.
What happened: the real outcome.
Lesson: the durable takeaway, written so the next founder avoids my mistake.
Source: where this came from.
3. Tell me which file to append it to: a durable rule (PRINCIPLES), a tactic (DOS-AND-DONTS), a dated
outcome (PLAYBOOK-LESSONS), or moat evidence (MOATS-LEDGER).
Append it below what's already there. Never overwrite an older lesson.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.

