@reddb-io/red-skills-memory 2.75.2

package/skills/core/wiki/REFERENCES.md

@@ -0,0 +1,48 @@
+# References
+
+Background material for the LLM Wiki pattern and adjacent tooling.
+
+## Karpathy — LLM Wiki
+
+- **Source:** https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
+- **Author:** Andrej Karpathy
+- **Year:** 2026
+- **TL;DR:** Instead of RAG (re-deriving knowledge on every query), the LLM **maintains** an incremental markdown wiki. Every new source is absorbed into the existing pages; cross-references, contradictions, and syntheses get compiled. The human curates; the LLM does bookkeeping.
+- **Why it matters:** the central pattern of this bucket. The design of `wiki-init` and `wiki` derives from it.
+
+## Memex
+
+- **Source:** Bush, Vannevar — _As We May Think_, The Atlantic, July 1945. https://www.theatlantic.com/magazine/archive/1945/07/as-we-may-think/303881/
+- **TL;DR:** Vision of a personal device that stores books, records, and communications, with **associative trails** between documents. The trail is as valuable as the document itself.
+- **Why it matters:** proto-LLM-wiki. The piece Bush couldn't solve — who maintains the trail — is exactly what an LLM solves.
+
+## Tolkien Gateway
+
+- **Source:** https://tolkiengateway.net/wiki/Main_Page
+- **TL;DR:** Fan wiki of Tolkien's legendarium. Thousands of cross-linked pages (characters, places, languages, events) built by volunteers over many years.
+- **Why it matters:** concrete example of an incremental human wiki. Demonstrates what an LLM can build on its own with light supervision.
+
+## qmd
+
+- **Source:** https://github.com/tobi/qmd
+- **Author:** Tobias Lütke
+- **TL;DR:** Local search engine for markdown. BM25 + vector search + LLM re-ranking, fully on-device. CLI + MCP server.
+- **When to install:** once the wiki grows past ~300 pages and `grep`/`ripgrep` becomes a query bottleneck. Update the search section in the consumer repo's `.red/agents/wiki.md` to point at `qmd`.
+
+## Obsidian Dataview
+
+- **Source:** https://github.com/blacksmithgu/obsidian-dataview
+- **TL;DR:** Obsidian plugin that runs SQL-like queries over page YAML frontmatter. Enables dynamic tables, lists filtered by tag/type, and charts.
+- **Why it matters:** enables rich views over the wiki without changing files — useful for a dynamic `index.md` during interactive use. We keep `index.md` static in the canonical format (so GitHub renders it), but the user can add local Dataview blocks without breaking portability.
+
+## Obsidian Web Clipper
+
+- **Source:** https://obsidian.md/clipper
+- **TL;DR:** Browser extension that converts web articles into markdown and saves them into the vault.
+- **Why it matters:** a quick shortcut to feed `.red/wiki/raw/`. An alternative to the `/wiki ingest <url>` flow when the content sits behind a paywall or requires login.
+
+## Adjacent patterns
+
+- **Zettelkasten** (Niklas Luhmann) — slip-box method with atomic notes and links. Difference: Zettelkasten has the human writing atomic notes; LLM Wiki has the agent writing integrated pages.
+- **Personal Knowledge Graphs** — same idea expressed as a graph (Roam, Logseq). LLM Wiki is the markdown-plus-frontmatter variant.
+- **NotebookLM, ChatGPT file uploads** — antagonists: RAG without persistence. Every query re-derives.

package/skills/core/wiki/SKILL.md

@@ -0,0 +1,154 @@
+---
+name: wiki
+description: Operate the LLM Wiki — ingest a source (URL or file), query the wiki and optionally file the answer back as a page, or lint for contradictions/orphans/stale claims. Requires `/wiki-init` to have run. Use when ingesting a doc, asking a question against the knowledge base, or checking the wiki for contradictions and orphans.
+---
+
+# wiki
+
+**Operate the LLM Wiki — ingest a source, query for answers, or lint for contradictions — reading `.red/agents/wiki.md` first (source of truth for this repo's conventions).**
+
+<what-to-do>
+
+## Preconditions
+
+- `.red/wiki/` exists.
+- `.red/agents/wiki.md` exists.
+- `.red/wiki/` is in `.gitignore` (verify before any operation that writes).
+
+If any of those fails, stop and tell the user to run `/wiki-init`.
+
+## Routing
+
+Map the opening verb to an operation; if ambiguous, ask which one:
+
+- "ingest …" / "add source …" → **Ingest**
+- question words / free-form question → **Query**
+- "lint" / "health check" → **Lint**
+
+## Ingest
+
+### 1. Resolve the input
+
+- **URL** → `WebFetch` the content, slugify the title (kebab-case), save to `.red/wiki/raw/<slug>.md` with a minimal YAML header (`url:`, `fetched:`, `title:`). Linked images: download to `.red/wiki/raw/assets/<slug>-<n>.<ext>` and rewrite the refs.
+- **Local path (PDF)** → `pdftotext <pdf> .red/wiki/raw/<slug>.txt` (or similar), then normalise to md at `raw/<slug>.md`. Keep the original PDF alongside.
+- **Local path (md/txt)** → copy/move to `raw/<slug>.md` if it isn't already there.
+
+### 2. Read and discuss
+
+`Read` the source. Before writing pages, discuss the 3–5 main takeaways with the user. Ask about emphasis: "highlight X or Y more?".
+
+### 3. Write pages
+
+Use the templates in [../wiki-init/](../wiki-init/):
+
+1. **Source page** — `pages/<slug>.md`, `type: source`. Bullets, key claims, quotes, "touches" filled in at the end.
+2. **Entity pages** — one for each new person/place/object. Slug kebab-case. If a page already exists, **update** it: add to `sources:`, revise `Overview`, update `updated:`, add new facts.
+3. **Concept pages** — same approach.
+4. **Contradictions** — if a claim conflicts with an existing page, **do not silence it**. Add a `## Contradictions` section on the affected page citing both sides.
+
+### 4. Update index.md
+
+Add or update entries in the right sections. Alphabetical order within each section. Each entry: `[Title](./pages/slug.md) — one-liner. N sources.`
+
+### 5. Log
+
+Append to `.red/wiki/log.md`:
+
+```markdown
+## [YYYY-MM-DD] ingest | <Source title>
+- source: raw/<slug>.md
+- touched: pages/a.md, pages/b.md, pages/c.md
+- notes: <one line on the most important change>
+```
+
+### 6. C4 awareness (optional, complexity-gated)
+
+If `.red/wiki/C4.md` exists, check whether the new source introduces a service, integration, or dependency not yet on the diagram. If so, update C4.md and bump its `updated:` field. If the project doesn't yet have a C4 but the new source describes enough architectural surface to warrant one (≥3 services or non-trivial integration), propose creating it. When C4 work triggers, read the [C4 reference](./C4-reference.md) for the model structure and conventions.
+
+### 7. Verify gitignore
+
+Before finishing: confirm `.red/wiki/` is still in `.gitignore`. If it was removed, alert and re-add.
+
+## Query
+
+### 1. Map
+
+`Read` `.red/wiki/index.md`. Identify candidate pages by title and one-liner. If nothing matches, fall back to `grep -ri "<term>" .red/wiki/pages/` or `ripgrep`.
+
+### 2. Drill
+
+`Read` the candidate pages. Follow cross-links if a page cites another relevant one.
+
+### 3. Synthesise
+
+Pick the most useful output format:
+
+- **Markdown prose** — for open-ended questions.
+- **Markdown table** — for comparisons or structured lists.
+- **Mermaid diagram** — for relationships, timelines, flows.
+
+Other formats (slides, charts) — only on explicit request. The agent can improvise but it's not the default.
+
+Every answer cites the pages (and via them the sources) used. Format: `(via [page-x](./pages/page-x.md))`.
+
+### 4. File back?
+
+If the answer has durable value (a fresh analysis, a non-obvious comparison, a timeline), ask: "file as a `synthesis` page?" If yes:
+
+1. Create `pages/<slug>.md` with `type: synthesis` and the appropriate template.
+2. Update `index.md`.
+3. Append log: `## [date] query | "<question>" → answer-filed: pages/<slug>.md`.
+
+If it isn't worth a page, just append to the log: `## [date] query | "<question>"`.
+
+## Lint
+
+Runs checks against `.red/wiki/`. **Does not auto-fix** — reports and asks what to do.
+
+### Checks
+
+1. **Contradictions** — scan pages; flag pairs that assert conflicting things. Look for: negations ("X is Y" in one, "X is not Y" in another); different numbers for the same attribute; conflicting dates.
+
+2. **Stale** — pages whose `updated:` is >90 days old and whose referenced `sources:` have `created:` more recent than the page's `updated:` (or whose sources have been superseded by newer sources on the same topic via tags).
+
+3. **Orphans** — pages with no inbound link (`grep -l "./<slug>.md" .red/wiki/pages/` returns empty).
+
+4. **Stubs** — pages with `< 10` lines of content (after the frontmatter).
+
+5. **Missing concepts** — terms repeated `>= 5x` across pages but lacking a page of their own (heuristic: lowercase + frequent bigrams/trigrams).
+
+6. **Gaps** — questions in "Open questions" unresolved for `>60` days.
+
+7. **C4 staleness** — if `.red/wiki/C4.md` exists, flag when its `updated:` is older than the most recent `created:` of any source page whose `touches:` list names a container or component appearing in the diagram. Suggested action: regenerate the affected level — when C4 work triggers, read the [C4 reference](./C4-reference.md).
+
+### Output
+
+Markdown report grouped by the checks above. For each item:
+
+- Path of the affected page(s).
+- Detail (e.g. "claim X in pages/a.md vs claim Y in pages/b.md").
+- Suggested action (e.g. "consolidate into pages/a.md, cite both sources").
+
+Append to log: `## [date] lint — orphans: N, stale: N, contradictions: N, stubs: N, gaps: N`.
+
+</what-to-do>
+
+<supporting-info>
+
+## Anti-patterns
+
+- ❌ Never operate without reading `.red/agents/wiki.md` first.
+- ❌ Never edit `raw/` — it is immutable.
+- ❌ Never silence contradictions — surface them.
+- ❌ Never use wikilinks `[[...]]` — use standard markdown `[...](./...)`.
+- ❌ Never commit `.red/wiki/` — verify `.gitignore` before any `git add`.
+
+## References
+
+- [LLM Wiki — Karpathy gist](./REFERENCES.md#karpathy-llm-wiki) — origin of the pattern.
+- [Memex — Vannevar Bush, 1945](./REFERENCES.md#memex) — proto-LLM-wiki.
+- [Tolkien Gateway](./REFERENCES.md#tolkien-gateway) — example of incremental human wiki.
+- [qmd](./REFERENCES.md#qmd) — local search engine for md, optional.
+- [Obsidian Dataview](./REFERENCES.md#obsidian-dataview) — dynamic queries via frontmatter.
+
+</supporting-info>

package/skills/core/wiki-init/SKILL.md

@@ -0,0 +1,127 @@
+---
+name: wiki-init
+description: Bootstrap the LLM Wiki pattern in a repo — create `.red/wiki/` layout, write `.red/agents/wiki.md` schema, gitignore wiki artifacts, register a `### Wiki` entry under `## Agent skills` in CLAUDE.md/AGENTS.md. Run once per repo before using `/wiki`. Re-run only to reset.
+disable-model-invocation: true
+---
+
+# wiki-init
+
+**Bootstrap the LLM Wiki with a short interview, then write the layout and agent schema.**
+
+Sets up the [LLM Wiki](../wiki/REFERENCES.md) pattern. After this runs, the agent has a persistent, incrementally-maintained knowledge base inside `.red/wiki/` and a schema in `.red/agents/wiki.md` that teaches future sessions how to ingest sources, query, and lint.
+
+<what-to-do>
+
+## Process
+
+Light grilling — three questions, then preview, then write. Don't dump all three at once; ask one, get the answer, move on.
+
+### 1. Explore
+
+Before asking anything, check:
+
+- `.red/wiki/` — does it already exist? If yes, refuse to overwrite without an explicit `--reset` from the user.
+- `.red/agents/wiki.md` — already present?
+- `CLAUDE.md` and `AGENTS.md` — does either have an `## Agent skills` block?
+- `.gitignore` — is `.red/wiki/` already listed?
+- `gh repo view --json visibility` — sanity check (informational only; the wiki is gitignored either way).
+
+### 2. Ask three questions
+
+The three questions are Domain, Source types, and Voice. Walk through them one at a time and recommend a default for each before waiting for an answer.
+
+**Q1 — Domain.** "In one sentence, what will this wiki accumulate?"
+Examples: "research on LLMs applied to finance", "reading notes on The Lord of the Rings", "competitive intel on reddb.io competitors", "personal journals and self-tracking".
+
+**Q2 — Source types you expect to feed in.** Multi-select, with defaults:
+- web articles (URL fetch)
+- PDFs (papers, books, reports)
+- transcripts (podcasts, calls, meetings)
+- personal notes (markdown drop)
+- other (describe in free text)
+
+This defines what `/wiki ingest` must support out of the gate.
+
+**Q3 — Solo or team?**
+- **Solo** → wiki voice is the user's first person; no "we", no "the team".
+- **Team** → neutral/collective voice, optional `team.md` page to map stakeholders.
+
+Default: infer from `git config user.name` + whether the repo has multiple contributors in `git shortlog -sn` (>1 → team).
+
+### 3. Preview
+
+Before writing, show the user:
+
+1. The paths that will be created.
+2. The rendered `.red/agents/wiki.md` (with domain/source-types/voice substitutions applied).
+3. The `.gitignore` line that will be appended.
+4. The `### Wiki` block that will land under `## Agent skills` in CLAUDE.md/AGENTS.md.
+
+Wait for approval.
+
+### 4. Write
+
+**Create the structure:**
+
+```
+.red/
+├── agents/wiki.md
+└── wiki/
+    ├── raw/
+    │   └── assets/
+    ├── pages/
+    │   └── .gitkeep
+    ├── index.md
+    └── log.md
+```
+
+Use the seed templates bundled with this skill:
+- [schema-template.md](./schema-template.md) → `.red/agents/wiki.md` (substituting placeholders `{{domain}}`, `{{source-types}}`, `{{voice}}`)
+- [index-template.md](./index-template.md) → `.red/wiki/index.md`
+- `log.md` starts with the heading `# Log` and nothing else.
+- For the full bundled template inventory, see [TEMPLATE-REFERENCE.md](./TEMPLATE-REFERENCE.md).
+
+**Append to `.gitignore`:**
+
+```
+# RedSkills wiki — local-only, never committed
+.red/wiki/
+```
+
+If `.gitignore` doesn't exist, create it. If the line already exists (same prefix), skip.
+
+**Update CLAUDE.md/AGENTS.md:**
+
+Pick rules identical to `red-setup`:
+- If `CLAUDE.md` exists, edit it.
+- Else if `AGENTS.md` exists, edit it.
+- If neither exists, ask which one to create.
+
+Append under `## Agent skills`:
+
+```markdown
+### Wiki
+
+Incremental LLM Wiki for accumulating knowledge about `{{domain}}`. Schema at `.red/agents/wiki.md`. Use `/wiki` for ingest, query, and lint.
+```
+
+If a `### Wiki` block already exists, update it in place instead of duplicating.
+
+### 5. Done
+
+Tell the user:
+- What was created.
+- That `.red/wiki/` is gitignored (nothing leaks).
+- Next step: run `/wiki ingest <first source>`.
+- That the schema is living — they can edit `.red/agents/wiki.md` directly, or ask the agent to update it as conventions emerge.
+
+</what-to-do>
+
+<supporting-info>
+
+## References
+
+- Bundled templates and examples: [TEMPLATE-REFERENCE.md](./TEMPLATE-REFERENCE.md).
+- External LLM Wiki pattern reference: [REFERENCES.md](../wiki/REFERENCES.md) in the `wiki` skill.
+
+</supporting-info>

package/skills/core/wiki-init/TEMPLATE-REFERENCE.md

@@ -0,0 +1,17 @@
+# Template Reference
+
+Reference-only inventory for `wiki-init/SKILL.md`. Load this when checking the
+bundled template set or following the external LLM Wiki pattern.
+
+## Bundled Templates
+
+- [schema-template.md](./schema-template.md)
+- [index-template.md](./index-template.md)
+- [page-template-entity.md](./page-template-entity.md)
+- [page-template-concept.md](./page-template-concept.md)
+- [page-template-source.md](./page-template-source.md)
+- [page-template-synthesis.md](./page-template-synthesis.md)
+- [examples/research.md](./examples/research.md)
+- [examples/book-reading.md](./examples/book-reading.md)
+
+External references: see [REFERENCES.md](../wiki/REFERENCES.md) in the `wiki` skill.

package/skills/core/wiki-init/examples/book-reading.md

@@ -0,0 +1,42 @@
+# Example — Book reading wiki
+
+A repo dedicated to a single reading: "The Fellowship of the Ring" (Tolkien).
+
+## After `/wiki-init`
+
+`.red/agents/wiki.md`:
+- `{{domain}}` = "companion wiki for reading The Lord of the Rings"
+- `{{source-types}}` = "book chapters (personal notes in md), academic commentary (PDFs), Tolkien interviews (transcripts)"
+- `{{voice}}` = "first person (solo)"
+
+## Emergent convention
+
+After ingesting the first chapter, the agent asks: "Chapters are the common source here — want a naming pattern?" The user agrees on `ch01-<slug>.md`. The agent **adds it to the schema** (`.red/agents/wiki.md`):
+
+```markdown
+## Naming exceptions
+
+- **Book chapters:** `ch<NN>-<slug>.md` (zero-padded, NN from 01 to 22). Example: `ch01-a-long-expected-party.md`.
+```
+
+## After 5 chapters
+
+```
+pages/
+├── ch01-a-long-expected-party.md   # type: source
+├── ch02-the-shadow-of-the-past.md  # type: source
+├── ch03-three-is-company.md        # type: source
+├── ch04-a-short-cut-to-mushrooms.md
+├── ch05-a-conspiracy-unmasked.md
+├── frodo-baggins.md                # type: entity, sources: 5
+├── bilbo-baggins.md                # type: entity, sources: 2
+├── gandalf.md                      # type: entity, sources: 3
+├── samwise-gamgee.md               # type: entity, sources: 3
+├── the-shire.md                    # type: entity (place)
+├── the-one-ring.md                 # type: entity (object)
+├── ring-bearer.md                  # type: concept
+├── eucatastrophe.md                # type: concept
+└── ring-temptation-patterns.md     # type: synthesis (4 sources)
+```
+
+Typical query: "how does the Ring affect each bearer?" → the agent reads the relevant pages, builds a comparison table, and offers to file it back as `pages/ring-bearer-effects.md` (synthesis).

package/skills/core/wiki-init/examples/research.md

@@ -0,0 +1,70 @@
+# Example — Research wiki
+
+A research repo covering "LLM Wiki patterns and adjacent tooling".
+
+## After `/wiki-init`
+
+```
+.red/wiki/
+├── index.md                     # empty, with stub sections
+├── log.md                       # header only
+├── raw/
+│   └── assets/
+└── pages/
+```
+
+`.red/agents/wiki.md` populated with:
+- `{{domain}}` = "research on LLM Wiki patterns and adjacent tooling"
+- `{{source-types}}` = "web articles, PDFs (papers), podcast transcripts, personal notes"
+- `{{voice}}` = "first person (solo)"
+
+## After 3 ingests
+
+```
+.red/wiki/
+├── index.md
+├── log.md
+├── raw/
+│   ├── karpathy-llm-wiki.md       # from the gist
+│   ├── memex-as-we-may-think.md   # extracted PDF
+│   └── qmd-readme.md              # from GitHub
+└── pages/
+    ├── karpathy-llm-wiki.md       # type: source
+    ├── memex-as-we-may-think.md   # type: source
+    ├── qmd-readme.md              # type: source
+    ├── andrej-karpathy.md         # type: entity
+    ├── vannevar-bush.md           # type: entity
+    ├── llm-wiki.md                # type: concept
+    ├── memex.md                   # type: concept
+    ├── associative-trails.md      # type: concept
+    ├── rag.md                     # type: concept (contrast with llm-wiki)
+    └── wiki-vs-rag.md             # type: synthesis
+```
+
+Example `index.md`:
+
+```markdown
+# Index
+
+## Sources
+
+- [Karpathy — LLM Wiki gist](./pages/karpathy-llm-wiki.md) — 2026. The incremental LLM-maintained wiki pattern.
+- [Bush — As We May Think](./pages/memex-as-we-may-think.md) — 1945. Original Memex vision.
+- [qmd README](./pages/qmd-readme.md) — Tobi Lütke. BM25+vector search over local markdown.
+
+## Entities
+
+- [Andrej Karpathy](./pages/andrej-karpathy.md) — author of the gist. 1 source.
+- [Vannevar Bush](./pages/vannevar-bush.md) — proto-author of the idea (Memex). 1 source.
+
+## Concepts
+
+- [Associative trails](./pages/associative-trails.md) — links between documents as knowledge in their own right. 2 sources.
+- [LLM Wiki](./pages/llm-wiki.md) — the central pattern. 1 source.
+- [Memex](./pages/memex.md) — Bush's proto-LLM-wiki. 1 source.
+- [RAG](./pages/rag.md) — antagonist to the LLM Wiki: retrieval without synthesis. 1 source.
+
+## Syntheses
+
+- [Wiki vs RAG](./pages/wiki-vs-rag.md) — accumulating vs re-deriving. 3 sources.
+```

package/skills/core/wiki-init/index-template.md

@@ -0,0 +1,19 @@
+# Index
+
+Wiki catalogue. Updated on every ingest and on every query that becomes a page.
+
+## Sources
+
+_(no sources yet — run `/wiki ingest <url|path>`)_
+
+## Entities
+
+_(no entities yet)_
+
+## Concepts
+
+_(no concepts yet)_
+
+## Syntheses
+
+_(no syntheses yet)_

package/skills/core/wiki-init/page-template-concept.md

@@ -0,0 +1,38 @@
+---
+title: {{Concept Name}}
+type: concept
+tags: []
+created: {{YYYY-MM-DD}}
+updated: {{YYYY-MM-DD}}
+sources: []
+---
+
+# {{Concept Name}}
+
+**Definition:** one precise sentence. If a canonical definition exists in a source, cite it.
+
+## Origin
+
+Who coined the term, where, when, in what context.
+
+## Mechanism / How it works
+
+How the concept operates. Components, dynamics, examples.
+
+## Instances
+
+- Appears in [Entity X](./entity-x.md)
+- Also seen in [Source Y](./source-y.md)
+
+## Related concepts
+
+- Close to [Another Concept](./another-concept.md): similarities and differences.
+- Contrast with [Antagonist](./antagonist.md).
+
+## Open questions
+
+- What is still unclear in the sources.
+
+## Contradictions
+
+_(if sources disagree on scope or definition)_

package/skills/core/wiki-init/page-template-entity.md

@@ -0,0 +1,34 @@
+---
+title: {{Display Name}}
+type: entity
+tags: []
+created: {{YYYY-MM-DD}}
+updated: {{YYYY-MM-DD}}
+sources: []
+---
+
+# {{Display Name}}
+
+**One-liner:** who/what/where, in one sentence.
+
+## Overview
+
+Short paragraph situating the entity. Who/what it is, primary role, why it appears in this wiki.
+
+## Key facts
+
+- Fact 1 (source: [slug](./slug.md))
+- Fact 2
+
+## Relationships
+
+- Links to [Another Entity](./another-entity.md) via X.
+- Contrast with [Another](./another.md): Y.
+
+## Open questions
+
+- Question not yet answered by the sources.
+
+## Contradictions
+
+_(add if sources disagree; cite both)_

package/skills/core/wiki-init/page-template-source.md

@@ -0,0 +1,43 @@
+---
+title: {{Source Title}}
+type: source
+tags: []
+created: {{YYYY-MM-DD}}
+updated: {{YYYY-MM-DD}}
+sources: [{{slug}}]
+---
+
+# {{Source Title}}
+
+- **Author:** _(who)_
+- **Published:** _(when)_
+- **URL / file:** `raw/{{slug}}.md` (origin: `<url>` or `<original file>`)
+- **Format:** article | paper | podcast | book-chapter | transcript | note
+
+## TL;DR
+
+3–5 bullets capturing the central argument.
+
+## Key claims
+
+1. Claim with a reference to the source's section/paragraph.
+2. Claim 2.
+
+## Methods / evidence
+
+How the source supports its claims. Useful for papers and reports.
+
+## Quotes
+
+> Citation worth preserving verbatim.
+
+## Touches
+
+Pages updated by this source:
+
+- [Entity X](./entity-x.md)
+- [Concept Y](./concept-y.md)
+
+## My take
+
+_(optional — human or agent annotation on the source's strengths/weaknesses)_

package/skills/core/wiki-init/page-template-synthesis.md

@@ -0,0 +1,45 @@
+---
+title: {{Synthesis Title}}
+type: synthesis
+tags: []
+created: {{YYYY-MM-DD}}
+updated: {{YYYY-MM-DD}}
+sources: []
+---
+
+# {{Synthesis Title}}
+
+**Question / thesis:** what this synthesis answers or argues, in one sentence.
+
+## Context
+
+Why this synthesis exists. The question that came from the human, or the pattern that emerged from the wiki.
+
+## Argument
+
+Structure as bullets, short paragraphs, or a table:
+
+| Aspect | Source A | Source B | Consequence |
+|--------|----------|----------|-------------|
+| ...    | ...      | ...      | ...         |
+
+Or prose when tabular comparison doesn't fit.
+
+## Evidence
+
+Pages and sources that support the argument:
+
+- [Page X](./page-x.md) — contributes Y
+- [Source Z](./source-z.md) — contributes W
+
+## Tensions / disagreements
+
+Where the sources diverge. Name the conflict; don't hide it.
+
+## Implications
+
+What changes if this synthesis is true.
+
+## Open questions
+
+What still needs investigating.