技能檔案

Continuous Learning V2

Name: Continuous Learning V2
Author: RogerioSobrinho

Instinct-based learning system with project scoping. Creates atomic instincts with confidence scoring and evolves them into skills, commands, or agents. v2 adds project-scoped instincts to prevent cross-project contamination. Use when tracking a heuristic specific to one project, promoting a project instinct to global scope, or reviewing project-specific patterns.

RogerioSobrinho0 星標2026年3月28日

職業
分類: 機器學習

技能內容

Continuous Learning v2 — Project-Scoped Instincts

Extends the base continuous-learning skill with project isolation: React patterns stay in your React project, Java conventions stay in your Java project. Universal patterns (always validate input, grep before edit) are shared globally.

What's Different from v1

Feature	v1	v2
Scope	All instincts are global	Project-scoped by default, promotable to global
Storage	One flat file per pattern	Separate dirs per project
Cross-project	Contamination risk	Isolated by default
Promotion	Manual	Explicit `/promote` step

Instinct Model (same as v1 + scope field)

相關技能

Continuous Learning V2 | Skills Pool

## Instinct: {Name}

**Heuristic:** [One sentence — what does the instinct say?]

**Confidence:** {1-5} / 5

**Scope:** project | global

**Project:** {project name, or "all projects" if global}

**Evidence log:**
| # | Context | Outcome | Score delta |
|---|---|---|---|
| 1 | {Where this was applied} | {Worked/Failed/Partial} | +1 |

**Promoted to pattern:** [Link to pattern, or "pending"]
**Contradictions:** [What evidence challenged this?]

~/.copilot/
├── instincts/
│   ├── global/                    # Apply in all projects
│   │   ├── grep-before-edit.md
│   │   └── always-validate-input.md
│   └── projects/
│       ├── my-spring-api/         # Project-specific
│       │   ├── avoid-eager-fetch.md
│       │   └── use-record-dtos.md
│       └── my-react-app/
│           └── prefer-signals.md
└── evolved/                       # Promoted to skills/agents/commands
    ├── skills/
    ├── agents/
    └── commands/

Pattern Type	Scope
Language/framework conventions	project
File structure preferences	project
Code style choices	project
Error handling strategy	project
Security practices	global
General best practices	global
Tool workflow preferences (grep before edit)	global
Git practices (conventional commits)	global

Occurrence	Action
1st	Note as project-scoped instinct (confidence: 2)
2nd	Update instinct (confidence: 3–4); note whether it's project-specific
3rd in same project	Promote to named pattern in that project's SKILL.md section
3rd across different projects	Promote to global instinct (confidence: 4–5); candidate for a new SKILL.md

{repo-root}/
└── docs/
    └── adr/
        ├── ADR-001-choose-flyway-over-liquibase.md
        └── ADR-002-use-record-dtos.md

Continuous Learning V2

Continuous Learning v2 — Project-Scoped Instincts

What's Different from v1

Instinct Model (same as v1 + scope field)

Continuous Learning V2

Continuous Learning v2 — Project-Scoped Instincts

What's Different from v1

Instinct Model (same as v1 + scope field)

Storage Layout

Scope Decision Guide

Three-Time Rule (same as v1, now with scope awareness)

Promotion: Project → Global

Review Protocol (end of session)

ADR Storage

Migration from v1

Continuous Learning V2

Continuous Learning V2

Continuous Learning V2

Continuous Learning

Continuous Learning

Pytorch Patterns