Build and maintain a persistent, compounding knowledge base as interlinked markdown files. Based on Andrej Karpathy's LLM Wiki pattern.
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.
Use this skill when the user: - Asks to create, build, or start a wiki or knowledge base - Asks to ingest, add, or process a source into their wiki - Asks a question and an existing wiki is present at the configured path - Asks to lint, audit, or health-check their wiki - References their wiki, knowledge base, or "notes" in a research context
Location: Set via WIKI_PATH environment variable (e.g. in ~/.hermes/.env).
If unset, defaults to ~/wiki.
WIKI="${WIKI_PATH:-$HOME/wiki}"
The wiki is just a directory of markdown files — open it in Obsidian, VS Code, or any editor. No database, no special tooling required.
wiki/
├── SCHEMA.md # Conventions, structure rules, domain config
├── index.md # Sectioned content catalog with one-line summaries
├── log.md # Chronological action log (append-only, rotated yearly)
├── raw/ # Layer 1: Immutable source material
│ ├── articles/ # Web articles, clippings
│ ├── papers/ # PDFs, arxiv papers
│ ├── transcripts/ # Meeting notes, interviews
│ └── 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
Layer 1 — Raw Sources: Immutable. The agent reads but never modifies these. For large assets, use a hybrid pattern: keep small raw markdown/JSON/manifests local, but store large PDFs, rendered page images, audio, and video in S3-compatible object storage with local .s3stub pointer files and a manifest.json.
Layer 2 — The Wiki: Agent-owned markdown files. Created, updated, and
cross-referenced by the agent.
Layer 3 — The Schema: SCHEMA.md defines structure, conventions, and tag taxonomy.
A strong adaptation of Brad Bonanno's self-updating wiki pattern is to treat the wiki as: - a normal markdown vault - tracked in git - optionally opened in Obsidian - updated by scheduled Hermes jobs that act like "context farmers"
Recommended additions for a production wiki repo:
wiki/
├── context/
│ ├── watchlists.md # sources, channels, feeds, competitors, query seeds
│ └── farmers/
│ └── .state/ # last-run markers per farmer/job
├── raw/
├── concepts/
├── entities/
└── ...
Use this pattern when the user wants the wiki to keep growing without manual ingestion every day.
For Hermes, scheduled updates map naturally to cronjob, not Claude's cloud scheduler.
A "context farmer" is just a recurring Hermes workflow that:
1. reads a watchlist or source config
2. fetches only new material since the last run
3. writes raw source files into raw/
4. updates entity/concept pages
5. records the run in log.md
6. updates a last-run state file
cronjobcronjob + optional delegate_taskweb_extract, web_search, terminal, provider CLIs, APIs)Store last-run timestamps under a repo-local state path such as:
context/farmers/.state/youtube-last-run.txt
context/farmers/.state/research-last-run.txt
These can be gitignored if they are purely operational.
context/watchlists.md should capture what matters, not implementation details. Example sections:
- tracked YouTube channels
- tracked blogs or RSS feeds
- tracked companies / competitors
- key topics / keywords
- exclusions / noise filters
For recurring wiki updates, create one cron job per source class or one orchestrator job that fans out. Examples: - every morning: ingest tracked YouTube channels - every 2 hours: ingest monitored RSS/blog feeds - every evening: synthesize new raw files into entity/concept pages
Use separate jobs when source classes have different failure modes or cadences.
When the user has an existing wiki, always orient yourself before doing anything:
① Read SCHEMA.md — understand the domain, conventions, and tag taxonomy.
② Read index.md — learn what pages exist and their summaries.
③ Scan recent log.md — read the last 20-30 entries to understand recent activity.
WIKI="${WIKI_PATH:-$HOME/wiki}"
# Orientation reads at session start
read_file "$WIKI/SCHEMA.md"
read_file "$WIKI/index.md"
read_file "$WIKI/log.md" offset=<last 30 lines>
Only after orientation should you ingest, query, or lint. This prevents: - Creating duplicate pages for entities that already exist - Missing cross-references to existing content - Contradicting the schema's conventions - Repeating work already logged
For large wikis (100+ pages), also run a quick search_files for the topic
at hand before creating anything new.
When the user asks to create or start a wiki:
$WIKI_PATH env var, or ask the user; default ~/wiki)SCHEMA.md customized to the domain (see template below)index.md with sectioned headerlog.md with creation entryAdapt to the user's domain. The schema constrains agent behavior and ensures consistency:
# Wiki Schema
## Domain
[What this wiki covers — e.g., "AI/ML research", "personal health", "startup intelligence"]
## Conventions
- File names: lowercase, hyphens, no spaces (e.g., `transformer-architecture.md`)
- Every wiki page starts with YAML frontmatter (see below)
- Use `[[wikilinks]]` to link between pages (minimum 2 outbound links per page)
- When updating a page, always bump the `updated` date
- Every new page must be added to `index.md` under the correct section
- Every action must be appended to `log.md`
- **Provenance markers:** On pages that synthesize 3+ sources, append `^[raw/articles/source-file.md]`
at the end of paragraphs whose claims come from a specific source. This lets a reader trace each
claim back without re-reading the whole raw file. Optional on single-source pages where the
`sources:` frontmatter is enough.
## Frontmatter
```yaml
---
title: Page Title
created: YYYY-MM-DD
updated: YYYY-MM-DD
type: entity | concept | comparison | query | summary
tags: [from taxonomy below]
sources: [raw/articles/source-name.md]
# Optional quality signals:
confidence: high | medium | low # how well-supported the claims are
contested: true # set when the page has unresolved contradictions
contradictions: [other-page-slug] # pages this one conflicts with
---
```
`confidence` and `contested` are optional but recommended for opinion-heavy or fast-moving
topics. Lint surfaces `contested: true` and `confidence: low` pages for review so weak claims
don't silently harden into accepted wiki fact.
### raw/ Frontmatter
Raw sources ALSO get a small frontmatter block so re-ingests can detect drift:
```yaml
---
source_url: https://example.com/article # original URL, if applicable
ingested: YYYY-MM-DD
sha256: <hex digest of the raw content below the frontmatter>
---
The sha256: lets a future re-ingest of the same URL skip processing when content is unchanged,
and flag drift when it has changed. Compute over the body only (everything after the closing
---), not the frontmatter itself.
[Define 10-20 top-level tags for the domain. Add new tags here BEFORE using them.]
Example for AI/ML: - Models: model, architecture, benchmark, training - People/Orgs: person, company, lab, open-source - Techniques: optimization, fine-tuning, inference, alignment, data - Meta: comparison, timeline, controversy, prediction
Rule: every tag on a page must appear in this taxonomy. If a new tag is needed, add it here first, then use it. This prevents tag sprawl.
_archive/, remove from indexOne page per notable entity. Include: - Overview / what it is - Key facts and dates - Relationships to other entities ([[wikilinks]]) - Source references
One page per concept or topic. Include: - Definition / explanation - Current state of knowledge - Open questions or debates - Related concepts ([[wikilinks]])
Side-by-side analyses. Include: - What is being compared and why - Dimensions of comparison (table format preferred) - Verdict or synthesis - Sources
When new information conflicts with existing content:
1. Check the dates — newer sources generally supersede older ones
2. If genuinely contradictory, note both positions with dates and sources
3. Mark the contradiction in frontmatter: contradictions: [page-name]
4. Flag for user review in the lint report
### index.md Template
The index is sectioned by type. Each entry is one line: wikilink + summary.
```markdown
# Wiki Index
> Content catalog. Every wiki page listed under its type with a one-line summary.
> Read this first to find relevant pages for any query.
> Last updated: YYYY-MM-DD | Total pages: N
## Entities
<!-- Alphabetical within section -->
## Concepts
## Comparisons
## Queries
Scaling rule: When any section exceeds 50 entries, split it into sub-sections
by first letter or sub-domain. When the index exceeds 200 entries total, create
a _meta/topic-map.md that groups pages by theme for faster navigation.
# Wiki Log
> Chronological record of all wiki actions. Append-only.
> Format: `## [YYYY-MM-DD] action | subject`
> Actions: ingest, update, query, lint, create, archive, delete
> When this file exceeds 500 entries, rotate: rename to log-YYYY.md, start fresh.
## [YYYY-MM-DD] create | Wiki initialized
- Domain: [domain]
- Structure created with SCHEMA.md, index.md, log.md
When the user provides a source (URL, file, paste), integrate it into the wiki:
① Capture the raw source:
- URL → use web_extract to get markdown, save to raw/articles/
- PDF → use web_extract (handles PDFs), save to raw/papers/
- Pasted text → save to appropriate raw/ subdirectory
- Name the file descriptively: raw/articles/karpathy-llm-wiki-2026.md
- Add raw frontmatter (source_url, ingested, sha256 of the body).
On re-ingest of the same URL: recompute the sha256, compare to the stored value —
skip if identical, flag drift and update if different. This is cheap enough to
do on every re-ingest and catches silent source changes.
② Discuss takeaways with the user — what's interesting, what matters for the domain. (Skip this in automated/cron contexts — proceed directly.)
③ Check what already exists — search index.md and use search_files to find
existing pages for mentioned entities/concepts. This is the difference between
a growing wiki and a pile of duplicates.
④ Write or update wiki pages:
- New entities/concepts: Create pages only if they meet the Page Thresholds
in SCHEMA.md (2+ source mentions, or central to one source)
- Existing pages: Add new information, update facts, bump updated date.
When new info contradicts existing content, follow the Update Policy.
- Cross-reference: Every new or updated page must link to at least 2 other
pages via [[wikilinks]]. Check that existing pages link back.
- Tags: Only use tags from the taxonomy in SCHEMA.md
- Provenance: On pages synthesizing 3+ sources, append ^[raw/articles/source.md]
markers to paragraphs whose claims trace to a specific source.
- Confidence: For opinion-heavy, fast-moving, or single-source claims, set
confidence: medium or low in frontmatter. Don't mark high unless the
claim is well-supported across multiple sources.
⑤ Update navigation:
- Add new pages to index.md under the correct section, alphabetically
- Update the "Total pages" count and "Last updated" date in index header
- Append to log.md: ## [YYYY-MM-DD] ingest | Source Title
- List every file created or updated in the log entry
⑥ Report what changed — list every file created or updated to the user.
A single source can trigger updates across 5-15 wiki pages. This is normal and desired — it's the compounding effect.
When the user asks a question about the wiki's domain:
① Read index.md to identify relevant pages.
② For wikis with 100+ pages, also search_files across all .md files
for key terms — the index alone may miss relevant content.
③ Read the relevant pages using read_file.
④ Synthesize an answer from the compiled knowledge. Cite the wiki pages
you drew from: "Based on [[page-a]] and [[page-b]]..."
⑤ File valuable answers back — if the answer is a substantial comparison,
deep dive, or novel synthesis, create a page in queries/ or comparisons/.
Don't file trivial lookups — only answers that would be painful to re-derive.
⑥ Update log.md with the query and whether it was filed.
When the user asks to lint, health-check, audit, or prepare for a large/source-sensitive ingestion, run or create a repo-local QA gate before bulk processing. For high-value corpora (lectures, books, PDFs, domain archives), do not treat lint as optional cleanup after the fact: add scripts/wiki_qa.py or equivalent first, document it in docs/QA.md, run it after every ingestion phase, and commit only when it passes. See references/source-grounded-qa-gates.md for the Human Design/large-media pattern.
① Orphan pages: Find pages with no inbound [[wikilinks]] from other pages.
# Use execute_code for this — programmatic scan across all wiki pages
import os, re
from collections import defaultdict
wiki = "<WIKI_PATH>"
# Scan all .md files in entities/, concepts/, comparisons/, queries/
# Extract all [[wikilinks]] — build inbound link map
# Pages with zero inbound links are orphans
② Broken wikilinks: Find [[links]] that point to pages that don't exist.
③ Index completeness: Every wiki page should appear in index.md. Compare
the filesystem against index entries.
④ Frontmatter validation: Every wiki page must have all required fields (title, created, updated, type, tags, sources). Tags must be in the taxonomy.
⑤ Stale content: Pages whose updated date is >90 days older than the most
recent source that mentions the same entities.
⑥ Contradictions: Pages on the same topic with conflicting claims. Look for
pages that share tags/entities but state different facts. Surface all pages
with contested: true or contradictions: frontmatter for user review.
⑦ Quality signals: List pages with confidence: low and any page that cites
only a single source but has no confidence field set — these are candidates
for either finding corroboration or demoting to confidence: medium.
⑧ Source drift: For each file in raw/ with a sha256: frontmatter, recompute
the hash and flag mismatches. Mismatches indicate the raw file was edited
(shouldn't happen — raw/ is immutable) or ingested from a URL that has since
changed. Not a hard error, but worth reporting.
⑨ Page size: Flag pages over 200 lines — candidates for splitting.
⑩ Tag audit: List all tags in use, flag any not in the SCHEMA.md taxonomy.
⑪ Log rotation: If log.md exceeds 500 entries, rotate it.
⑫ Report findings with specific file paths and suggested actions, grouped by severity (broken links > orphans > source drift > contested pages > stale content > style issues).
⑬ Append to log.md: ## [YYYY-MM-DD] lint | N issues found
# Find pages by content
search_files "transformer" path="$WIKI" file_glob="*.md"
# Find pages by filename
search_files "*.md" target="files" path="$WIKI"
# Find pages by tag
search_files "tags:.*alignment" path="$WIKI" file_glob="*.md"
# Recent activity
read_file "$WIKI/log.md" offset=<last 20 lines>
When ingesting multiple sources at once, batch the updates: 1. Read all sources first 2. Identify all entities and concepts across all sources 3. Check existing pages for all of them (one search pass, not N) 4. Create/update pages in one pass (avoids redundant updates) 5. Update index.md once at the end 6. Write a single log entry covering the batch
When the user corrects the operating frame of a wiki-backed project — e.g. “this is execution-first, not research-first,” “we already have relationships/buyers,” or “organize around equipment/deployment, not validation” — treat it as a first-class wiki maintenance task, not a chat-only clarification. Create or update a concept page for the corrected thesis, update relevant object/inventory pages, patch the main plan/roadmap and swarm brief, and update index/log/action registers. If a deployed console/cockpit reads or hardcodes summaries from the wiki, patch and redeploy the app/API too so the UI does not continue surfacing stale priorities.
See references/strategic-correction-reframe.md for the tested pattern and verification checklist.
For wikis that ingest PDFs/books/media, preserve speed by keeping compiled markdown plus small raw markdown/JSON/manifests local, while storing large immutable source PDFs, rendered page images, scans, audio, and video in S3-compatible object storage. Use local .s3stub pointer files and raw/books/<slug>/manifest.json; fetch assets into .cache/s3/ only when exact raw reinspection is needed, and verify with SHA-256 before use.
See references/hybrid-s3-raw-assets.md for the tested Alex VPS pattern, manifest shape, gitignore rules, and command examples.
For MEGA-hosted corpora, especially when raw source files must not persist on the VPS, see references/mega-cloud-drive-ingest.md. It captures the megajs workaround for public MEGA folder listing/download, golden-sample selection, transient /tmp download → S3 upload → local deletion workflow, and verification checks.
For large corpora that must keep ingesting autonomously beyond the current chat session, see references/background-corpus-swarm-ingest.md. It captures the run-until-complete Hermes cron swarm pattern: dedicated project space, state file, reports, package coverage ledger, orchestrator/digest jobs, and explicit stop conditions.
For productized multi-tenant ingestion UIs (e.g. Hermes Spawn), see references/spawn-kb-ingestion-sessions.md. Treat every KB ingest as a durable Hermes ingestion session with a visible transcript/status, scope note, final report, resumable artifacts, and interrupt support. Do not run hidden one-shot ingestion that can fail silently behind a raw queued source.
When the user wants the swarm actively processing now rather than waiting for a scheduled heartbeat, use references/foreground-continuation-workers.md. It captures the pattern of spawning a bounded hermes chat worker from the swarm workdir, verifying the live process/PID, leaving cron enabled, and committing/pushing clean wiki changes between bounded units.
For background swarms that touch S3, provider SDKs, or other tools under cron, see references/autonomous-swarm-runtime-bootstrap.md. It captures the self-contained .env + .venv + scripts/bootstrap_runtime.sh pattern, real S3 put/head/delete verification before ledger unblocking, JSON .s3stub cleanup, and prompt language that tells the orchestrator to install missing dependencies before declaring blockers.
For Human Design wiki swarm synthesis, do not impose fixed page-count targets. See references/human-design-wiki-swarm.md for Alex's corrected dynamic-synthesis rule: load the relevant extracted source context into GPT-5.5 and let source coverage determine whether the run creates one deep page, many atomic pages, query/comparison pages, source-page expansion, or candidate work items, while preserving provenance and QA cleanliness. That reference also captures the compounding synthesis rule: update existing Markdown pages when topics already exist, create new pages only with source support, and add/repair [[wikilinks]].
For a concrete Human Design golden-sample ingest, see references/human-design-lyd-golden-sample.md. It captures the RA LYD 7h lecture + 94-page slide PDF lessons: timestamp-free ASR is provisional, PowerPoint PDFs need slide-aware vision routing, .cache/ raw media can still fail QA, and --s3 failures may come from loading credentials for the wrong bucket.
For the completed high-fidelity LYD pass, see references/human-design-lyd-timestamped-slide-ingest.md. It captures the timestamped Venice Whisper workflow, 94-slide vision extraction, resumable missing-page reruns after provider 429s, candidate slide/transcript alignment, disk cleanup, and the QA pitfall where combined transcripts reset timestamps at each source-audio heading.
For turning a large MEGA/wiki effort into a long-running autonomous project, see references/human-design-wiki-swarm.md. It captures the dedicated /home/avalon/hd-wiki-swarm project-space pattern, Hermes cron orchestrator + daily digest jobs, state/report layout, role definitions, bounded-run policy, and the chart-analysis-oriented synthesis target for gates, lines, colors, tones, bases, variables, centers, channels, transits, and chart comparison.
For source-sensitive bulk ingests (long lectures, PDFs, slide decks, domain archives), add committed QA gates before bulk processing. See references/source-grounded-qa-gates.md for checks covering no-local-media policy, manifests/stubs, S3 HEAD verification, transcript timestamps, PDF rendered-page assets, in-body provenance markers, and avoiding ungrounded placeholder pages.
For single long strategy recordings or interviews that need raw transcript preservation plus structured wiki pages, use references/long-audio-to-wiki-ingest.md. It covers cloud-link acquisition when Telegram cannot cache oversized voice files, chunked faster-whisper transcription, watcher/idempotency pattern, parallel timestamp-slice analysis, wiki page synthesis, and disk-space pitfalls on Alex's VPS.
For long Telegram/voice recordings that should become a new wiki, use the durable incoming-folder + watcher + chunked faster-whisper pattern in references/long-audio-telegram-wiki-ingest.md. Check the gateway log for Failed to cache voice: File is too big before assuming the audio exists locally, and use alternate upload routes when Telegram bot download rejects the file.
When the corpus is too large for a single session and the user wants autonomous progress until exhaustion, run the wiki through a dedicated background swarm/project space rather than ad-hoc chat turns. The pattern is: a swarm directory beside the wiki, a persistent state file, recursive inventory, package coverage ledger, reports per run, a local/noiseless Hermes cron orchestrator, and a separate daily digest job. The orchestrator should process bounded resumable units, but its global stop condition must be explicit: all ingestible packages/files are complete or intentionally skip_*/blocked with reasons. Do not stop at golden samples.
See references/background-corpus-swarm-ingest.md for the tested run-until-complete pattern and prompt requirements.
When a scheduled farmer writes into the wiki, add enough provenance that later sessions can tell human and automated updates apart. Good patterns:
- raw filenames that encode source + date, e.g. raw/articles/youtube-nate-herck-2026-05-01.md
- frontmatter fields such as source: farmer/youtube, farmed: 2026-05-01T06:00:00Z
- a matching log entry describing which farmer/job ran and which files changed
This keeps the wiki auditable as it compounds over time.
When adapting an LLM wiki into a chat product, do not let the agent silently write every interesting chat turn into durable memory. Put a memory-harvest pass after the assistant answer that considers both sides of the exchange: what the user shared and what the assistant synthesized. The pass should draft reviewable save proposals rather than directly mutating the wiki.
Good proposal fields: kind, title, summary, proposedMarkdown, targetPath, crossLinks, confidence, provenance (userMessageId, assistantMessageId, timestamp), and status (pending, approved, rejected, superseded). Candidate kinds include durable person facts, life events, dream notes, reading conclusions, chart/HD corrections, transit windows, substantial query syntheses, and user corrections.
Before approving writes, orient to SCHEMA.md, index.md, recent log.md, and relevant pages; update existing pages before creating new ones; cross-link updates; and record contradictions instead of overwriting. In multi-tenant products, approval must go through constrained KB tools/path locking, not general filesystem access. Public UI should show simple human labels like “Possible memory update noticed,” not raw internal file paths, source chunks, skill ids, or hidden routing metadata.
When content is fully superseded or the domain scope changes:
1. Create _archive/ directory if it doesn't exist
2. Move the page to _archive/ with its original path (e.g., _archive/entities/old-page.md)
3. Remove from index.md
4. Update any pages that linked to it — replace wikilink with plain text + "(archived)"
5. Log the archive action
The wiki directory works as an Obsidian vault out of the box:
- [[wikilinks]] render as clickable links
- Graph View visualizes the knowledge network
- YAML frontmatter powers Dataview queries
- The raw/assets/ folder holds images referenced via ![[image.png]]
For a hosted/web-native viewer, publish the Markdown vault with Quartz instead of running the Obsidian desktop app through VNC/noVNC. Quartz provides a fast static site with backlinks, search, and graph-style navigation while keeping the vault files as source of truth. Treat Quartz as the viewer/publishing layer, not the whole Karpathy LLM wiki pattern: the full pattern still requires ingestion/context-farmer jobs that create raw source records, synthesize entity/concept/comparison pages, and update index.md/log.md.
See references/quartz-viewer-and-context-farmers.md for the VPS multi-vault Quartz deployment pattern and the viewer-vs-farmer distinction learned from Alex's Obsidian migration.
For best results:
- Set Obsidian's attachment folder to raw/assets/
- Enable "Wikilinks" in Obsidian settings (usually on by default)
- Install Dataview plugin for queries like TABLE tags FROM "entities" WHERE contains(tags, "company")
If using the Obsidian skill alongside this one, set OBSIDIAN_VAULT_PATH to the
same directory as the wiki path.
On machines without a display, use obsidian-headless instead of the desktop app.
It syncs vaults via Obsidian Sync without a GUI — perfect for agents running on
servers that write to the wiki while Obsidian desktop reads it on another device.
Setup:
# Requires Node.js 22+
npm install -g obsidian-headless
# Login (requires Obsidian account with Sync subscription)
ob login --email <email> --password '<password>'
# Create a remote vault for the wiki
ob sync-create-remote --name "LLM Wiki"
# Connect the wiki directory to the vault
cd ~/wiki
ob sync-setup --vault "<vault-id>"
# Initial sync
ob sync
# Continuous sync (foreground — use systemd for background)
ob sync --continuous
Continuous background sync via systemd:
# ~/.config/systemd/user/obsidian-wiki-sync.service
[Unit]
Description=Obsidian LLM Wiki Sync
After=network-online.target
Wants=network-online.target
[Service]
ExecStart=/path/to/ob sync --continuous
WorkingDirectory=/home/user/wiki
Restart=on-failure
RestartSec=10
[Install]
WantedBy=default.target
systemctl --user daemon-reload
systemctl --user enable --now obsidian-wiki-sync
# Enable linger so sync survives logout:
sudo loginctl enable-linger $USER
This lets the agent write to ~/wiki on a server while you browse the same
vault in Obsidian on your laptop/phone — changes appear within seconds.
raw/ — sources are immutable. Corrections go in wiki pages.log-YYYY.md and start fresh.
The agent should check log size during lint.authority, type, or center has not yet been source-grounded, leave it as plain text or a working note until an ingest actually supports it. Broken wikilinks should fail QA rather than encouraging ungrounded pages.For high-value media/PDF ingests, prefer fidelity over speed/cost — use timestamped transcription, rendered PDF page images, vision extraction for artifacted PDFs, and phase-level QA gates. Structural lint is necessary but not sufficient; include manual/LLM-assisted spot checks against the original audio/page images.
When the user corrects the project frame, propagate it through the wiki and any cockpit. Do not leave a strategic correction buried in chat or in one plan page. Create a durable concept for the corrected thesis, update plans/swarm briefs/action registers/index/log, and if a live console has hardcoded recommendations or gates, patch/redeploy it. See references/strategic-correction-reframe.md.
llm-wiki-compiler is a Node.js CLI that compiles sources into a concept wiki with the same Karpathy inspiration. It's Obsidian-compatible, so users who want a scheduled/CLI-driven compile pipeline can point it at the same vault this skill maintains. Trade-offs: it owns page generation (replaces the agent's judgment on page creation) and is tuned for small corpora. Use this skill when you want agent-in-the-loop curation; use llmwiki when you want batch compile of a source directory.