菜单
Identifies behavioral patterns from recent conversations and reaction feedback, updates learned-patterns.md when 3+ observations confirm a pattern.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
| name | pattern-learning |
| description | Identifies behavioral patterns from recent conversations and reaction feedback, updates learned-patterns.md when 3+ observations confirm a pattern. |
| allowed-tools | Read, Write, Edit, Bash |
Review recent interactions and reaction feedback. Update agent-data/learned-patterns.md when you see something recurring 3+ times.
agent-data/memories/): jokes that landed/flopped, successful/failed approaches, repeated mistakesagent-data/reaction-feedback.jsonl): emoji reactions on your messages. Interpret emoji names semantically (e.g., "omegaLUL" = positive, "sadge" = negative). Compare GIF vs text reaction rates, roast vs earnest, per-person patterns.agent-data/learned-patterns.md): what's already capturedAdd when you see the same thing work/fail 3+ times with clear cause-and-effect. Don't add one-off observations, vague patterns, or things already in the file.
Categories in learned-patterns.md: Humor, Conversation, Per-Person, Games, Self-Corrections, What Lands, Timing, Meta-Patterns. Pick the right section.
Hypothesis creation and review is owned by the hypothesis-review skill — if you notice something that's a testable prediction rather than a confirmed pattern, note it in today's memory for hypothesis-review to pick up.
After significant interactions, when you notice repeated mistakes, or during idle reflection.
Cooldown: 24 hours
Tags transcript message threads with semantic concepts (roast, game-recommendation, etc.) so past conversations can be searched by idea, not just keywords.
Captures notable interactions missed during live conversation — new person info, game preferences, boundaries, jokes that landed/flopped, relationship insights.
The single owner of the hypothesis lifecycle — creates, tests, promotes, and kills hypotheses. Uses extensions as programmatic data collectors and reviews their output.
Consolidates impression files by applying weight decay, removing redundant entries, and enforcing per-user limits. Maintains relationship file summaries.
Consolidates old memories by removing duplicates, promoting recurring patterns to learned-patterns.md or relationship files, and archiving outdated entries.
Evaluates overall effectiveness — are idle behaviors producing value, are extensions working, what's blocking improvement. Higher-level thinking about approach, not individual conversations.