continual-learning（continual learning 命令行）

Continual Learning

Keep AGENTS.md current by mining this session's conversation history.

Inputs

• Current session messages (already in context)
• Existing memory file: AGENTS.md

Workflow

1. Read existing AGENTS.md.
2. Review messages from the current session conversation.
3. Extract only high-signal, reusable information:

- Recurring user corrections/preferences - Durable workspace facts (patterns, conventions, tech choices)

1. Merge with existing bullets in AGENTS.md:

- Update matching bullets in place - Add only net-new bullets - Deduplicate semantically similar bullets

1. Write back to AGENTS.md with only these two sections added or updated:

- ## Learned User Preferences - ## Learned Workspace Facts

AGENTS.md Output Contract

• Manage only these sections:

- ## Learned User Preferences - ## Learned Workspace Facts

• Use plain bullet points only.
• Do not write evidence/confidence tags.
• Do not write process instructions, rationale, or metadata blocks.
• Maximum 12 bullets per section.

Inclusion Bar

Keep an item only if all are true:

• Actionable in future sessions
• Stable across sessions
• Repeated in the conversation, or explicitly stated as a broad rule
• Non-sensitive

Exclusions

Never store:

• Secrets, tokens, credentials, private personal data
• One-off task instructions
• Transient details (branch names, commit hashes, temporary errors)