dylanchen1925

zhuomo

Use when turning books, EPUBs, blogs, videos, or notes into a personal wiki or agent skills; when learning from resources quickly, building domain frameworks, running roguelike multi-domain learning scenarios, or mapping progress across varied domains; when discovering topics or correcting existing wiki/skills.

dylanchen1925 0 Updated 3h ago

Resources

13
GitHub

Install

npx skillscat add dylanchen1925/zhuomo

Install via the SkillsCat registry.

SKILL.md

琢磨 (Zhuomo)

琢磨 — to polish, refine, and chew over raw material until it becomes clear and usable. Turn sources into a personal wiki and agent skills — help you learn faster, build frameworks of what you know, and keep knowledge correct across many domains.

Overview

Skills are proven techniques with triggers — not book summaries.

Give future agents: when to act, what to do, how to decide, what mistakes to avoid. Drop narrative, anecdotes, repetition.

Knowledge is never write-once. Ingest adds; revise corrects, updates, merges, and supersedes. See KNOWLEDGE-BASE.md and REFERENCE.md.

Two compounding outputs:

Output Holds Agent use
Wiki Synthesis, entities, links, contradictions Query, browse, domain skill backend
Skill Triggers + workflows (+ WIKI-SCOPE for domain skills) Auto-invoke under symptoms

REQUIRED: superpowers:writing-skills, write-a-skill
User: USER-GUIDE.md · Framework: FRAMEWORK.md
Wiki: KNOWLEDGE-BASE.md · Learn: LEARNING.md · Run: RUN.md · Retention: RETENTION.md · Domain skills: WIKI-BACKED-SKILLS.md

When to Use / NOT

Use: sources → wiki and/or skill; learn from resources (digests, quizzes, recall); run roguelike multi-domain scenarios; build/update domain frameworks; multi-domain vault; correct or update existing pages.

Don't: one-off answers without filing; silent overwrite; delete history instead of supersede/archive; wall-of-text summaries instead of linked learning artifacts.

Nine operations

Op When Output
Ingest New source Wiki pages (multi-topic, multi-domain)
Learn User studying Digests, pretest, recall cards → wiki/learn/
Run Multi-domain practice Fictional scenario + floors + debrief → wiki/learn/runs/RUN.md
Review Due cards / study session SR review + explain-back → RETENTION.md
Framework After ingest or on request domain-map, per-domain overview.md (pillars + progress), optional guide.md
Weekly ~15 min ritual Review + Connect/Run + Lint + progress → log.md
Query Question search (page list) or think (synthesis + gaps + file back)
Revise Wrong, stale, duplicate Corrected pages/skills + log
Lint Periodic health Doctor-lite checklist → issues → often Revise

Details: KNOWLEDGE-BASE.md, LEARNING.md, RUN.md, RETENTION.md

Wiki vs Skill

Put in wiki Zhuomo into skill
Entity pages, themes, synthesis Named technique with clear trigger
Cross-source contradictions Iron law / discipline rule
Research thesis evolving over weeks "When X symptom, do Y"
Comparison tables, cited answers Agent must comply under pressure

Often: topic map → ingest → wiki first → zhuomo extraction card → skill when actionable + non-default.

Domain skills (wiki as backend)

Technique skill = one workflow. Domain skill = expert persona + WIKI-SCOPE.md manifest — agent loads your wiki (e.g. BGP concepts) at invoke time; facts stay in wiki, not in SKILL.md.

Example: network-expert + wiki/domains/networking/ + [[concepts/bgp]]. Full pattern: WIKI-BACKED-SKILLS.md.

/zhuomo Domain skill: network-expert — wiki backend wiki/domains/networking/

Topics (optional from you)

You do not need to name the topic. Provide a source; the agent reads structure (TOC, headings, skim) and discovers topics — often several per book, paper, or video.

You provide Agent does
Nothing Infer all major topics; present topic map; proceed unless you narrow
Topic / lens ("focus on ch. 5–7") Prioritize those; still note other topics for later
Goal ("wiki only" / "one skill") Filter extraction toward that output

One resource → one source summary + many concept pages across one or more domains. Details: REFERENCE.md, LEARNING.md.

Wiki page layout (per domain)

Two pages per domain — do not create framework.md or mega-overview.md.

Page Path Holds
Vault hub wiki/overview.md Domain table + ingest rules only — no domain prose
Domain entry wiki/domains/<slug>/overview.md Why learn, architect lens, pillars, progress, glossary, study order, gaps
Domain guide (optional) wiki/domains/<slug>/guide.md One-page technical digest (merge of concept pages)
Concepts wiki/concepts/*.md Full depth + ## Evidence when deepened

New domain ingest: add a row to vault overview.md + domain-map.md; create domains/<slug>/overview.md. Add guide.md only when user wants one-scroll reference depth.

Lite mode default: SIMPLE.md. Vault AGENTS.md must mirror this block.

Concept pages (compiled truth): body = current trusted claims; optional ## Changelog or frontmatter records when ingest/revise changed what (never silent overwrite).

Query — brain-first, search vs think

Brain-first (always): read wiki before web or raw EPUB.

  1. wiki/overview.mddomain-map.mddomains/<domain>/overview.md / guide.md
  2. wiki/index.md → relevant concepts/ + sources/
  3. Only if insufficient: raw ~/zhuomo-data/raw/ or external search
Mode User says Agent returns
search Query search: … or "list wiki pages about …" Ranked page list + one-line why each matters — user reads pages
think (default) Query: … or Query think: … Synthesized answer + citations + ## Gaps

Think output template (required sections):

## Answer
…synthesis with [[wikilinks]] and Evidence anchors where deepened…

## Sources
- [[concept-or-page]] — what you used

## Gaps
| Gap | Why it matters | Suggested next step |
|-----|----------------|---------------------|
| … | stub / no Evidence / stale source / contradiction | deepen X / Revise Y / new source |

Network/IT: lead Answer with business constraint → design lever → technical object (see domain overview architect lens).

File back: comparisons, cross-concept synthesis, or durable Q&A → wiki/synthesis/ or strengthen an existing concept; append log.md if substantial.

Lint — doctor-lite

Run on request, after large ingest, or as part of Weekly.

Check Action if failed
Broken [[wikilinks]] Fix path or create stub concept linked from domain overview pillar
Orphan concept pages (no inbound links) Link from overview, guide, or related concepts
Concept mentioned in text but no page Create stub or merge duplicate
overview progress ≠ concept status / depth Revise overview or concept
Deepened book concept missing ## Evidence Add Evidence or downgrade progress note
Contradictions between pages Revise; supersede stale claim
Stale source (newer guide/version exists) Note in Gaps; flag in overview
Duplicate concept pages same topic Merge; one canonical page
Domain overview gaps list outdated Refresh after ingest

Append ## [date] lint | … to wiki/log.md. Turn each row into Revise or deepen follow-up.

Auto-stub (on ingest/lint): when a pillar or guide links [[aci-foo]] and page missing, create minimal stub with domain: frontmatter and link back to pillar — do not leave dangling wikilinks.

Weekly — mini dream cycle

~15 min ritual (RETENTION.md): Review due cards → Lint (doctor-lite) → merge duplicates → bump overview progress → optional one wiki/synthesis/ cross-concept note → log.md.

Learn & framework (for you, not agents)

Goal Ask for
Learn faster Learn — preview, digest, quiz, explain-back
Cross-domain under pressure Run — roguelike scenario fused from 2+ domains — RUN.md
See the big picture Framework — update domains/<slug>/overview.md (pillars, progress)
Many unrelated subjects Multi-domainwiki/domain-map.md + wiki/domains/*/

Default after chapter ingest (unless you say archive only): digest + update domain framework. Full guide: LEARNING.md.

Workflow Checklist

New source:

- [ ] 0. Wiki setup — bootstrap if needed; multi-domain → domain-map (LEARNING.md)
- [ ] 1. Intake — source; discover topics; assign domain(s)
- [ ] 1b. EPUB/PDF — convert full text to wiki/sources/[slug]/md/; Evidence links on every concept page (REFERENCE.md)
- [ ] 2. Ingest — wiki pages; flag contradictions
- [ ] 2b. Learn — pretest + digest + recall for Spaced Repetition (skip if "archive only")
- [ ] 2c. Framework — update `domains/<slug>/overview.md` (pillars, progress, gaps)
- [ ] 3–10. Extract → skill pipeline if actionable

Correct or update existing knowledge:

- [ ] 1. Locate — index.md, backlinks, related skills
- [ ] 2. Revision card — what was wrong, evidence, new claim (REFERENCE.md)
- [ ] 3. Propagate — fix all pages that cite the old claim
- [ ] 4. Wiki — edit in place or supersede; never silent delete
- [ ] 5. Skill — merge correction; re-run RED if discipline changed
- [ ] 6. Log — append wiki/log.md: revise | target | reason

Full ingest checklist:

- [ ] 3. Extract — extraction card per candidate idea
- [ ] 4. Filter — actionable AND non-default
- [ ] 5. Classify — technique / pattern / reference / discipline
- [ ] 6. Decide — new vs enhance vs split; run Revise if source contradicts wiki
- [ ] 7. RED — baseline WITHOUT draft skill
- [ ] 8. GREEN — minimal SKILL.md
- [ ] 9. REFACTOR — counters; re-test
- [ ] 10. Update SOURCES.md (+ wiki log.md)

Extraction card

Field Capture
Trigger Situation/symptoms (not chapter title)
Core move One non-obvious action
Steps Workflow or decision tree
Anti-pattern Common failure
Example One adapted before/after
Type technique / pattern / reference / discipline

Validate (TDD)

No SKILL.md until RED completes. See superpowers:writing-skills.

Common Mistakes

Mistake Fix
Chapter summary as skill Wiki for synthesis; skill for triggers only
RAG-only (no wiki) Ingest compiles once; wiki stays current
Wiki without schema AGENTS.md/CLAUDE.md defines ingest/query/lint
Skip RED Test skill first
Query answers lost in chat File good answers back to wiki
Web search before wiki Brain-first: overview → guide → concepts
Query without Gaps section Think mode must end with ## Gaps
Lint only when user complains Run doctor-lite after big ingest + Weekly
Fix only in chat Revise wiki/skill + log.md
New source contradicts old Revise affected pages, don't append both as true
Duplicate concept pages Merge on lint/revise; one canonical page
One topic per resource assumed Multi-topic normal — topic map first, one concept page each
User must name topic upfront Optional — agent infers; user topic = priority lens only
Summary dump, no framework Digest + link to pillars; update domains/<slug>/overview.md
framework.md / mega-overview.md Use overview.md + optional guide.md only — see Wiki page layout
Single-domain assumed Use domain-map + wiki/domains/* for varied subjects
BGP facts pasted into skill Domain skill + WIKI-SCOPE; Revise wiki only when facts change
Recall cards never reviewed Obsidian Spaced Repetition + Weekly Review — RETENTION.md
Scenario fiction filed as facts Run artifacts use type: fictional-scenario; Revise only for wiki errors — RUN.md
Cross-domain only in chat Run — fuse domains, file debrief to wiki/learn/runs/

Deployment

  • Skill: name, description, SOURCES.md, RED/GREEN/REFACTOR
  • Raw: local path (e.g. ~/zhuomo-data/raw/), inbox/ for phone captures, immutable snapshots
  • Wiki: Obsidian vault — wiki/, index.md, log.md, schema in AGENTS.md; sync to phone for read
  • Retention: Spaced Repetition plugin on wiki/learn/recall/ (#flashcards/domain); optional Readwise → inbox

Details: REFERENCE.md, KNOWLEDGE-BASE.md, LEARNING.md, RETENTION.md, WIKI-BACKED-SKILLS.md