Agent Skills
› NeverSight/learn-skills.dev
› repository-knowledge-compaction
repository-knowledge-compaction
GitHub将仓库探索结果压缩为紧凑、可复用的知识摘要，生成 brief.md 和 facts.yaml，确保信息稳定且易于后续智能体高效检索与复用。
Trigger Scenarios
完成仓库探索后
需要刷新现有仓库简介时
Install
npx skills add NeverSight/learn-skills.dev --skill repository-knowledge-compaction -g -y
SKILL.md
Frontmatter
{
"name": "repository-knowledge-compaction",
"description": "Compress repository findings into durable human-readable and machine-readable memory so later agents can reuse repo knowledge efficiently."
}
Repository Knowledge Compaction
Use this skill after repository exploration or when refreshing an existing repository brief.
Goals
- Preserve repository knowledge in a compact, reusable form.
- Make later retrieval cheap for both humans and agents.
- Avoid stale or bloated memory artifacts.
Required Inputs
- Raw repository findings
- Existing repository brief or facts file, if any
.ai-team/framework/memory/wiki/repositories/for existing entries
Required Outputs
- Updated
brief.mdinwiki/repositories/<repo-slug>/ - Updated
facts.yamlinwiki/repositories/<repo-slug>/ - Updated
wiki/repositories/_index.yamlwhen needed
Compaction Rules
- Keep only high-value facts that downstream roles are likely to reuse.
- Prefer stable facts over transient implementation trivia.
- Record source path, revision, and last-updated timestamp in the machine-readable file.
- Use arrays and short strings in
facts.jsonso other tooling can parse it easily. - Archive uncertainty as
open_questionsinstead of hiding it inside prose.
Version History
- e0220ca Current 2026-07-05 23:31
