AgentSkillsCN

conversation_miner

在删除聊天记录之前，建立一套协议，从聊天日志中挖掘高价值情境（“黄金”内容），并按技术（操作系统）与元数据（个人资料）两大类别进行分类整理。

SKILL.md
--- frontmatter
name: conversation_miner
description: A protocol for mining high-value context ("Gold") from chat logs before deletion, sorting it into Tech (OS) and Meta (Profile) streams.

Conversation Refinery Skill (v2)

When to use this skill

  • •When the user asks to "mine this chat" or "extract gold".
  • •At the end of a session before the user deletes the conversation history.
  • •When the user asks to "update documentation" based on recent work.

Instructions

1. The Goal: Dual-Stream Refining

Your objective is to separate the conversation into two distinct streams. Do NOT dilute them by mixing.

  • •Stream A (Hard Ore): Technical updates (Code, Config, Fixes).
  • •Stream B (Red Gold): Meta updates (Identity, Decisions, Philosophy, Bio).

2. Target Files (Routing Logic)

Stream A: The Machine (Technical)

  • •src/content/docs/ARCHITECTURE.md (Why code changed)
  • •src/content/docs/handbook/OPERATIONS.md (How to fix it)
  • •src/content/docs/ROADMAP.md (What we finished)
  • •src/content/docs/meta/AGENCY_MEMORY.md (Active focus)

Stream B: The Operator (Meta/Identity)

  • •src/content/docs/meta/AGENT_PROFILE.md (Behavioral updates, new "Laws" of personality)
  • •src/content/docs/project/GROK_LOG.md (Architectural Decrees)
  • •src/content/docs/meta/BIO.md (Public narrative updates)
  • •src/content/docs/colophon/*.md (Philosophical adds)
  • •src/content/docs/prompts/BRANDING_PROMPT.md (Voice/Tone updates)

3. Protocol: "Extract the Gold"

Look for:

  • •Decisions: Why did we choose X over Y?
  • •Config: New environment variables or constants.
  • •Fixes: Reproducible steps for bug fixes.
  • •Traps: Edge cases found.
  • •Meta-Features: Did we build something cool? (Add to Colophon in Roadmap).
  • •Self-Correction: Did the user correct our tone/behavior? (Update AGENT_PROFILE.md).

4. Protocol: "Memory Synapse"

  • •

    Update src/content/docs/meta/AGENCY_MEMORY.md:

    • •Clear completed focus items (Purge closed loops!).
    • •Add new "Active Intelligence" (URLs, Notebooks) ONLY if it's broad meta-focus.
    • •Update "Current Focus State" (Where did we leave off?).
    • •[CRITICAL LAW]: DO NOT log project hydration status or project milestones here.
  • •

    Update src/content/docs/project/MINING_LOG.md:

    • •This is the Sovereign Ledger for Project Status.
    • •If a project was enriched, hydrated, or modified, log it HERE, not in Memory.

5. Output Format

Generate a single markdown document categorized by file. Use tags to indicate status:

  • •[IMPLEMENTED]: Changes already made.
  • •[PROPOSED]: Changes the user needs to make.

Example Output:

markdown
### `docs/handbook/OPERATIONS.md` (Technical Stream)

**Add to "Troubleshooting" section:**

- **Symptom:** Page renders white.
- **Fix:** Ensure `<Layout>` wraps the content.

### `docs/meta/AGENT_PROFILE.md` (Meta Stream)

**Add to "Communication Protocol":**

- **Trigger:** User grew frustrated when I apologized excessively.
- **Calibration:** "Do NOT apologize. Report yields only."