wiki-write
Skill1 starsChanged 3 days agoMITView on GitHub
Use when a real decision was made, multiple inputs were synthesized, something broke, or a pattern emerged across events, to append a dated wiki entry under $WIKI_DIR (default ./wiki). Append-only, frontmatter-indexed, queryable by a later reader.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- LucasCooperBey/claude-skillsword for word, first seen Nov 16, 2026
- hmjBill/skillstheir own version, first seen Oct 29, 2026
- wicii2120/skillstheir own version, first seen Aug 29, 2026
- Oshayr/LLM-Wikitheir own version, first seen Apr 10, 2026
Other versions4 projects use a version of this.
Word for word · 1
- LucasCooperBey/claude-skills★ 1100% same
Edited · 0
None found.
Wrote their own version · 3
- Oshayr/LLM-Wiki★ 50same name
- wicii2120/skills★ 1same name
- hmjBill/skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository53 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/deep-recon/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- adversarial-researchSkill
- algorithmic-artSkill
- api-integrationSkill
- breakdownSkill
- c4-architectSkill
- canvas-designSkill
- coding-agentsSkill
- compound-engineeringSkill
- context-budgetSkill
- counterfactualSkill
- critiqueSkill
- d3js-visualizationSkill
- data-analyst-agentSkill
- database-migrationsSkill
- deep-reconSkill
- design-tasteSkill
- e2b-sandboxSkill
- ecosystem-auditSkill
- executing-plansSkill
- frontend-designSkill
- fullstack-agentSkill
- grillSkill
- kickoffSkill
- mcp-builderSkill
- model-migrationSkill
- naming-the-machineSkill
- full-output-enforcementSkill
- package-researchSkill
- pdf-generationSkill
- pp2pSkill
- pp3pSkill
- prompt-engineerSkill
- prompt-groundingSkill
- redesignSkill
- requesting-code-reviewSkill
- reverse-engineer-to-learnSkill
- review-changesSkill
- scratchpadSkill
- shipSkill
- slack-gif-creatorSkill
- sprint-architectSkill
- techdebtSkill
- test-and-fixSkill
- threat-modelSkill
- verification-before-completionSkill
- visual-verdictSkill
- webapp-testingSkill
- web-asset-generatorSkill
- whatdidimissSkill
- writing-plansSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Write an append-only entry to the wiki.
- Distinctive: line 26
If you would reference the entry 6 months from now, write it.
- Very specific: line 30
Before writing a decision (or an ADR-style entry), confirm ALL THREE hold.
- Very specific: line 42
Do not copy a code-local ADR into the wiki.
- Very specific: line 93
Run ~/.claude/skills/wiki-write/scripts/build-index.sh from the same folder to refresh $WIKI_DIR/INDEX.md.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
wiki-write
Write an append-only entry to the wiki.
The scripts live in this skill's scripts/ folder (~/.claude/skills/wiki-write/scripts/ when installed for your user). Run them by that full path from the folder that should hold the wiki, usually your project root. The wiki root is $WIKI_DIR. When that variable is unset, the scripts use ./wiki under the folder you run them from, so set WIKI_DIR to an absolute path to keep one wiki across projects.
When to invoke this skill
- A choice was made that a future reader would benefit from re-reading:
decision - Multiple inputs (transcripts, research, meetings) distilled into findings:
synthesis - Something broke or surprised: production bug, agent failure, mistake:
incident - A pattern emerged across multiple events worth naming:
reflection
When NOT to invoke
- In-progress plans, working drafts: project folders
- Raw tool-call traces: project logs, not the wiki
- Conversation context that will not be referenced again: do not write
- Trivial actions (renamed a variable, fixed a typo): do not write
If you would reference the entry 6 months from now, write it. Otherwise skip.
Decision-worthiness gate (3 gates)
Before writing a decision (or an ADR-style entry), confirm ALL THREE hold. If any is missing, skip it. This stops the wiki filling with non-decisions.
- Hard to reverse: there is real cost to changing your mind later
- Surprising without context: a future reader will wonder why it was done this way
- The result of a real trade-off: genuine alternatives existed and one was chosen for specific reasons
A reversible, obvious, no-alternative call is not a decision. Do not record it.
Where the decision lives (routing fork)
- A decision that lives with the code belongs in that repo's
docs/adr/. - This wiki is for cross-project and strategic decisions and syntheses.
- Do not copy a code-local ADR into the wiki. Reference it in
linksorsources.
How to write
- Pick the type: decision | synthesis | incident | reflection
- Pick a slug: kebab-case, descriptive, no date prefix (the helper adds today's date). Example:
pick-a-weekly-review-day,rename-the-cache-key - Compose frontmatter + body, omitting
id,type,date: the helper injects those. Required fields:actors:,[me, claude]tags:, 2 to 6 kebab-case tags (for example[weekly-review, cache, naming])summary:, one to three sentences as a YAML block scalar. A reader uses this, plus tags and actors, to decide whether to open the file.
- Optional fields:
domain,status,supersedes,sources,links- Linking: populate
links(related entries) andsupersedes(the entry this one replaces) so a later reader can traverse the notes. An unlinked entry is an island: when this entry relates to, replaces, or builds on another, wire the relation.
- Linking: populate
- Body sections depend on type: see SCHEMA.md. Typical patterns:
- decision:
## What changed,## Why,## Tradeoffs - synthesis: numbered findings with one-line headers
- incident:
## What happened,## Root cause,## Fix,## Prevention
- decision:
Invocation
~/.claude/skills/wiki-write/scripts/write-entry.sh <type> <slug> <<'EOF'
actors: [me, claude]
tags: [tag-one, tag-two, tag-three]
summary: >
One to three sentences describing the entry. A reader uses this
plus tags and actors to decide whether to open the file.
domain: meta
status: active
---
# Entry title
## What changed
Body text.
EOFThe script reads $WIKI_DIR (default ./wiki).
The script:
- Validates required fields (
actors,tags,summary), the type, and a kebab-case slug - Refuses a duplicate id
- Injects
id,type, anddate - Files
decisionandincidentunder$WIKI_DIR/decisions/,synthesisunder$WIKI_DIR/synthesis/, andreflectionunder$WIKI_DIR/counterfactuals/ - Prints
WROTE: <path>on success - On any refusal, exits non-zero and prints one error line
After writing
Run ~/.claude/skills/wiki-write/scripts/build-index.sh from the same folder to refresh $WIKI_DIR/INDEX.md.
Read first
SCHEMA.md, full frontmatter contractREADME.md, what this folder is and how to run itexamples/decisions/2026-11-12-example-decision.md, example decision shapeexamples/synthesis/2026-11-12-example-synthesis.md, example synthesis shape
Source
The append-only wiki follows the LLM wiki pattern Andrej Karpathy described. The scripts, the schema and the tests are original to this repo.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

