Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add johnnywuj81/tokenknows --skill distillgit clone --depth 1 https://github.com/johnnywuj81/tokenknowsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/johnnywuj81/tokenknows/distill)<a href="https://agentmods.dev/skills/johnnywuj81/tokenknows/distill"><img src="https://agentmods.dev/badge/skills/johnnywuj81/tokenknows/distill/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/johnnywuj81/tokenknows/distill"><img src="https://agentmods.dev/badge/skills/johnnywuj81/tokenknows/distill.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00085 | $0.01463 |
| Opus 5.5 | $0.00034 | $0.00585 |
| Sonnet 5.5 | $0.00017 | $0.00293 |
| Haiku 4.5 | $0.00009 | $0.00146 |
Grade A, and why
distill scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 118 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Distill Session
把当前 Claude session 蒸馏成 TokenKnows 后端的结构化文档之一。
用户意图判断
用户说下列任一表达,即应调用此 skill:
| 用户说 | 推断文档类型 |
|---|---|
| "蒸馏 / 总结 / 整理" 这次对话 | 让用户选, 默认 weekly_report |
| "出周报" / "生成周报" / "weekly" | weekly_report |
| "技术方案" / "设计文档" / "tech design" | tech_design |
| "ADR" / "架构决策" / "决策记录" | adr |
| "复盘" / "故障复盘" / "postmortem" | incident |
| "技术书籍" / "教程" / "long-form" | book |
| "蒸馏 skill" / "提炼专家技能" | agent_skill |
| "知识图谱" / "KG" / "实体关系" | knowledge_graph |
如果用户没明示,优先问一句 "你想蒸馏成哪种类型?周报 / 技术方案 / ADR / 复盘 / 书籍 / Skill / 知识图谱?",然后再走流程。
标准流程 (5 步)
Step 1 · 整理本次 session 的关键事件
把对话拆成 3-10 条 event (每条聚焦一个语义单元):
- 用户提的需求 / 问题 / 决策点
- Claude 给出的方案 / 关键代码 / 取舍说明
- 重要工具调用 (e.g. Edit/Bash 执行的 PR 合并、测试结果)
每条 event 形如:
{
"external_id": "<session_uuid>-<msg_index>",
"source_ref": "<session_uuid>",
"event_type": "ai_conversation_turn", // 或 tool_call/code_change
"content": "<对话片段或代码变更摘要>",
"title": "<一句话标题>",
"author_name": "user" 或 "Claude",
"tags": ["关键词1", "关键词2"]
}
关键: external_id 务必唯一 (同 session_uuid + msg_index)避免后端 dedup 时丢失。
Step 2 · 批量提交 events
调 MCP tool: submit_session_events
events: [...上面整理的 3-10 条...]
返回 {ingested, skipped, project_id}。如果 skipped > 0,说明部分 event 被 backend 去重 (content_hash 已存在),正常不用担心。
Step 3 · 触发蒸馏 pipeline
调 MCP tool: distill_document
document_type: "<step 1 用户选的类型>"
time_window: "this_week" // 默认; 用户明说 "上周/最近 7 天" 时改
返回 {asset_id, status: "generating", view_url, estimated_seconds: 60}。view_url 是完整可点击的绝对 URL (前缀由 TOKENKNOWS_WEB_BASE 决定,默认 http://127.0.0.1:5173)。
告诉用户:"已触发蒸馏,大约 60 秒。我会轮询完成状态。"
Step 4 · 轮询完成
每 5-10 秒调一次 get_asset(asset_id),直到 status == "draft" (或 failed)。
- 如果 60 秒还在 generating,告诉用户后端 LLM 还在跑,继续等
- 如果超 3 分钟 → 后端可能挂,提示用户查 backend log
- 如果 status=
draft→ 进入 Step 5
Step 5 · 展示蒸馏结果
调 get_asset_chapters(asset_id),拿到完整 markdown。
展示方式按类型分:
- weekly_report / tech_design / adr / incident: 把每个 chapter 的
title+content用 markdown headings 直接输出给用户 - book: 章节多,先列大纲 (chapter titles) 给用户,再问要看哪几章
- agent_skill: 输出 SKILL.md 风格 markdown, 让用户决定是否落地到
~/.claude/skills/ - knowledge_graph: layout 含 nodes/edges; 用 ASCII art 简述 (e.g.
Alice --authored_by--> PR#127),指引用户开浏览器看可视化 (URL 在 view_url 里,绝对 URL 可直接点)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 118 lines · 85 tokens per session scan A 95509a3d3bfb
distill is a skill published in the GitHub repository johnnywuj81/tokenknows (4 stars, last pushed 1mo ago), licensed MIT. It adds 85 tokens to every session and 1,463 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
oma-hwp
Convert Hangul HWP, HWPX, or HWPML documents to Markdown with kordoc. Use for text, table, and image extraction from these formats.
oma-pdf
Extract PDF text, headings, tables, and images into Markdown using opendataloader-pdf. Use for PDF parsing, not PDF layout creation.
pkf-writing
Author, update, convert, validate, persist, and render PKF (.pkf) content through message-inline, Library Asset, or Memory Page carriers. Use for structured reports, knowledge pages, delivery cards, or any .pkf body. NOT for recall, memory-graph placement, or curation decisions — those belong to the memory skill.
dex-obsidian-setup
Turn on Obsidian mode and migrate the vault to wiki links. Use when the user says 'I use Obsidian', 'enable wiki links', 'make this work in Obsidian'. Not for connecting an external tool/API; use create-mcp/integrate-mcp.
document-actions
Extract cited facts, obligations, deadlines and proposed actions from documents, preserving OCR uncertainty and reconciling Prismer tasks.
honey-px
Read huge read-only text as PNG pages; big input cut.