AgentSkillsCN

knowledge-harvester

指导如何从本地Slack对话记录文件中提取持久化的组织知识。

SKILL.md
--- frontmatter
name: knowledge-harvester
description: Guidance for extracting durable organizational knowledge from local Slack transcript files.

Knowledge Harvester

Use this skill when analyzing local transcript snapshots to identify preservable knowledge.

Scope

  • Input is transcript text from slack-forge/transcripts/*.md.
  • Do not fetch Slack data directly.

Knowledge Signals

  • Decisions and rationale
  • Process or policy changes
  • Ownership/responsibility changes
  • Project milestone/scope updates
  • Terminology/acronym definitions
  • Durable architecture/technical context

Filter Out

  • Social/casual chat
  • Ephemeral status chatter
  • Acknowledgments without content
  • Repetitive bot noise with no decision context

Durability Test

Keep items likely useful in 2+ weeks.

Confidence

  • high: explicit decision/announcement
  • medium: strong contextual insight
  • low: tentative/speculative statements

Provenance Requirements

For each extracted knowledge item, preserve:

  • source channel name and ID
  • source author
  • source timestamp
  • supporting transcript quote/context

Memory Hints

Use tags to signal likely destination type at promote time:

  • person
  • project
  • glossary
  • general