ss-learn

1# Learn from project design decisions
2 
3`ss-learn` turns a **specific human-approved correction** into a generalized candidate rule. It
4does not train a model, scrape a repository, or upload telemetry. The CLI is local-only. An
5optional plugin MCP bridge can expose one exact package to its connected client/model only after a
6separate one-time human grant.
7 
8Read `references/privacy-contract.md` before using this skill.
9 
10## When not to use
11 
12- The user did not explicitly ask to capture or share a lesson.
13- The change was accepted only by the agent, not a person.
14- The lesson cannot be expressed without client/product identity, source code, a screenshot,
15  proprietary tokens, or user content.
16- A score or visual pass was not actually measured. Record it as `null` or `not-run`; never infer.
17- The observation belongs only to one project's taste. Keep it in `STYLESEED.md` instead.
18 
19## 1. Initialize local learning
20 
21After explicit user approval:
22 
23```bash
24node <installed-ss-learn>/scripts/learning.mjs init --project-root .
25```
26 
27This creates `.styleseed/learning/config.json` with sharing disabled and all raw-material
28collection disabled. It performs no network request.
29 
30## 2. Draft a candidate
31 
32Use `references/candidate.schema.json`. Generalize the lesson:
33 
34- problem: what design failure was observed;
35- intervention: what bounded change the person accepted;
36- rationale: why it improved the product job;
37- appliesWhen: conditions where the judgment should transfer;
38…