update-claude-learnings（update Claude learnings 搜索）

Non-negotiable rules:

1. Only record learnings about Claude Code behavior or workflow in this project.
2. Reject application-code, subagent, or skill-authoring learnings and route them to the proper maintenance workflow instead.
3. Check for duplicates or near-duplicates before writing.
4. Get explicit user confirmation before modifying CLAUDE.md.
5. Preserve existing structure and only add the smallest correct instruction.

Update Claude Learnings

Inputs

• $request: Optional learning candidate, category hint, or reminder about what the session revealed

Goal

Add one validated Claude-behavior learning to project memory by:

• extracting the right rule from the session
• verifying it belongs in CLAUDE.md
• placing it in the correct section
• preserving the surrounding memory structure
• reporting exactly what changed

Step 0: Confirm the learning belongs here

This skill is only for persistent Claude Code behavior in this project.

Valid examples:

• workflow rules for when Claude should use a specific skill
• scope-control rules for how Claude should ask before expanding work
• session-management rules for checkpoints, progress updates, or stopping conditions
• project-specific Claude behavior that improves future sessions

Invalid examples:

• application implementation rules better suited for agent files
• skill-authoring patterns better suited for skill learnings
• transient one-off notes that do not deserve durable memory

Load references/learning-scope.md for routing, category placement, and failure modes.

If the learning does not belong in CLAUDE.md, stop and say where it should go instead.

Success criteria: The learning clearly belongs in persistent Claude project memory.

Step 1: Extract one concrete learning from the session

Review the session and identify the smallest useful rule.

Rules:

• prefer one precise learning over several vague ones
• write it in imperative mood
• tie it to a concrete behavior, not a general aspiration
• avoid duplicating rules that are already implied by stronger existing guidance

Good shape:

• "Use /commit instead of manual git commit flows when the user explicitly asks to commit."
• "Ask before expanding a bugfix into adjacent refactors."

Bad shape:

• "Be more careful."
• "Use better workflows."

Success criteria: You have a single actionable learning candidate with a clear rationale.

Step 2: Check the current CLAUDE memory and choose placement

Read the project CLAUDE.md. If it does not exist, use references/claude-md-template.md as the structural fallback.

Choose the correct section:

• Workflow Rules
• Session Management
• Scope Control
• Behavioral Patterns

Before writing:

• search for duplicates or near-duplicates
• merge with existing wording if a similar rule already exists
• keep the new instruction small and local

Load references/learning-scope.md for placement and duplicate-handling guidance.

Success criteria: The target section is known and duplication risk has been checked.

Step 3: Confirm with the user

Before editing CLAUDE.md, present:

• category
• final wording
• reason this learning was extracted
• intended section placement

Use AskUserQuestion if confirmation or wording refinement is needed.

Do not write until the user explicitly approves the learning.

Success criteria: The user has approved the learning and its placement.

Step 4: Update CLAUDE.md

Apply the minimal correct edit:

• preserve file structure
• insert the learning in the chosen section
• avoid deleting unrelated content
• update the "Last updated" marker only if the file already uses one

Rules:

• if CLAUDE.md is missing, create it using the provided template and then add the learning
• if the section is missing, create the smallest compatible section rather than restructuring the whole file
• keep formatting consistent with the existing document

Success criteria: CLAUDE.md contains the approved learning in the right place without collateral churn.

Step 5: Verify and report

Verify:

• the learning was added exactly once
• the file structure still makes sense
• the instruction remains actionable and specific
• the update did not drift into agent or skill memory territory

Report:

• category
• section path
• final wording
• whether the file was created or updated

Success criteria: The user can see exactly what durable Claude memory changed.

Guardrails

• Do not let the model invoke this skill proactively; it mutates durable memory.
• Do not add context: fork; this workflow edits the active repository.
• Do not add paths:; this is a generic maintenance skill.
• Do not keep routing matrices, quality scorecards, or long failure catalogs inline in SKILL.md.
• Do not add a learning without explicit user approval.
• Do not rewrite large parts of CLAUDE.md when a small targeted insertion is enough.

When To Load References

• references/learning-scope.md Use for deciding whether the learning belongs in Claude memory, choosing section placement, handling duplicates, and checking common failure modes.
• references/claude-md-template.md Use only when CLAUDE.md is missing or its structure needs a minimal compatible fallback.

Output Contract

Report:

1. whether the learning was accepted or redirected elsewhere
2. the chosen category and section
3. the final approved wording
4. whether CLAUDE.md was created or updated
5. any duplicate merge or placement decisions