wiki

Wiki

Unlike traditional RAG (which rediscovers knowledge from scratch per query), the wiki compiles knowledge once and keeps it current. Cross-references are already there. Contradictions have already been flagged. Synthesis reflects everything ingested.

Division of labor: the human curates sources and directs analysis. The agent summarizes, cross-references, files, and maintains consistency.

<orient_first_top> Read $WIKI/SCHEMA.md once at the start of any session that activates this skill. The schema declares the domain, the page-type enum, the tag taxonomy, and the conventions every subsequent action must honor. Reading it first prevents wrong-type pages, off-taxonomy tags, and duplicate or misfiled work. The full orientation pass (SCHEMA + index + recent log) is covered in <resuming_an_existing_wiki>; this top-line note exists so the SCHEMA read is never skipped, even on quick queries. </orient_first_top>

<when_to_activate> Use this skill when the user:

• Asks to create, build, or start a wiki or knowledge base.
• Points at a named source to bring into the wiki: activate wiki_import.
• Asks a question that an existing wiki at the discovered location could answer.
• Asks for a wiki-wide lint, audit, health check, clean-up, or auto-repair: activate wiki_fix.
• References their wiki, knowledge base, or "notes" in a research context.
• Asks to capture procedural knowledge (workflows, conventions, runbooks) alongside the wiki's subject pages. </when_to_activate>

wiki/
├── SCHEMA.md           # Conventions, structure rules, domain config
├── index.md            # Sectioned content catalog with one-line summaries
├── log.md              # Chronological action log (append-only in substance, rotated yearly)
├── raw/                # Layer 1: Immutable source material
│   ├── articles/       # Externally-published articles, web clippings
│   ├── papers/         # PDFs, arxiv papers
│   ├── meetings/       # Meeting notes, interviews, spoken-word transcripts (podcasts, talks)
│   ├── notes/          # Internal memos, discussion writeups, ad-hoc observations, internal docs not published externally
│   └── assets/         # Images, diagrams referenced by sources
├── entities/           # Layer 2: Entity pages (people, orgs, products, models)
├── concepts/           # Layer 2: Concept/topic pages
├── comparisons/        # Layer 2: Side-by-side analyses
├── queries/            # Layer 2: Filed query results worth keeping
├── summaries/          # Layer 2: Standalone overview / digest pages
└── procedures/         # Layer 2: Procedure / workflow / how-to pages

The log is append-only in substance. Never reword, reorder, or delete what a past entry records. An entry may still be edited to repair a structural or lint break it introduced, keeping what it records and where it sits. See <appending_to_log> for the rule, and <lint_and_audit> for the duplicate heading the linter surfaces.

Layer 1: Raw Sources. Immutable. The agent reads but never modifies these. Layer 2: The Wiki. Agent-owned markdown files. Created, updated, and cross-referenced by the agent. Page types split into declarative (what / why) and procedural (how). See <page_types> below. Layer 3: The Schema. SCHEMA.md defines structure, conventions, and tag taxonomy. The linter reads its ## Frontmatter yaml block to learn the page-type enum, so wikis can declare additional types there without modifying the linter.

<folder_layout> The folder tree is the type axis only. Every page lives directly at <pluralized-type>/<slug>.md, flat, one layer deep. No thematic prefix (themes/ai/concepts/foo.md is wrong), no sub-folder nesting inside a type folder (concepts/ai/foo.md is wrong), no bare files at the wiki root. Thematic scope belongs in tags: and type:, not in folder names, because tags have an enumerated taxonomy in SCHEMA.md and don't drift the way emergent theme folders do. The linter's structure check enforces this with blocking severity on misfiled pages (with a concrete move suggestion) and warn when an expected type folder is missing on disk. Because the wiki's primary consumer is an LLM with tools that greps and follows links rather than browsing, visual folder shelving adds no real value and trades it for filing-drift risk. </folder_layout>