agentsclimarketplace

Wiki ingest

Skill Tedydev-web/llm-wiki-skills/skills/wiki-ingest

Process raw source documents into LLM Wiki pages. Use when the user adds files to raw/ and wants them ingested, says "process this source", "ingest this article", "I added something to raw/", or wants to incorporate new material into their knowledge base / second brain.From its SKILL.md

Install
npx -y skills add Tedydev-web/llm-wiki-skills --skill wiki-ingest

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • reads credentialsReads from 1 credential source: `$WIKI_MEMORY_VAULT`.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 2 commands, including `cat ~/.config/wiki/sidecar.json | jq -r .vault_path` and 1 more.

SKILL.md

4.0 KB, 951 tokens by cl100k_base, as published. Nobody here has run it

LLM Wiki — Ingest

Process raw source documents into structured, interlinked wiki pages.

Vault path: read from $WIKI_MEMORY_VAULT env, else cat ~/.config/wiki/sidecar.json | jq -r .vault_path, else cat ~/.config/wiki-memory/vault-path (legacy), else ask user.

Commands

  • /wiki-ingest — process files in raw/ (and raw/sessions/ if wiki-memory enabled)
  • /wiki-ingest --promote-qa <slug> — graduate wiki/qa/<slug>.md into a permanent concept page

Identify Sources to Process

  1. If the user specifies a file or files, use those
  2. If the user says "process new sources" or similar, detect unprocessed files:
    • List all files in raw/ (excluding raw/assets/)
    • Read wiki/log.md and extract all previously ingested source filenames from ingest entries
    • Any file in raw/ not listed in the log is unprocessed
  3. If no unprocessed files are found, tell the user

Workflow (High Level)

For each source file:

  1. Step 0 — State check: read wiki/.state.json, compute SHA256, skip unchanged files. For full spec: Read skills/wiki-ingest/references/state-json-spec.md
  2. Steps 1–8 — Ingest pipeline: read source → discuss takeaways → create source page → update entity/concept pages → add wikilinks → update index → update log → report. For full walkthrough: Read skills/wiki-ingest/references/ingest-walkthrough.md
  3. Step N — State write: update .state.json atomically with computed hashes + output paths. For full spec: Read skills/wiki-ingest/references/state-json-spec.md

For --promote-qa mode (skips pipeline entirely): Read skills/wiki-ingest/references/promote-qa-workflow.md

Quick Example

User: "I added raw/article-on-aws.md, please ingest it."

  1. Read .state.json — check SHA256 of raw/article-on-aws.md; if unchanged since last run, skip
  2. Read the article; share 3-5 key takeaways; wait for user confirmation
  3. Create wiki/sources/article-on-aws.md
  4. Create or update wiki/entities/aws.md, wiki/concepts/cloud-cost-optimization.md, etc.
  5. Update wiki/index.md + wiki/cache.md + append to wiki/log.md
  6. Write updated .state.json

Conventions

  • Source summary pages are factual only. Save interpretation for concept and synthesis pages.
  • A single source typically touches 10–15 wiki pages. This is normal.
  • When new information contradicts existing content, update the page and note the contradiction with both sources cited.
  • Prefer updating existing pages over creating new ones.
  • Use [[wikilinks]] for all internal references — never raw file paths.
  • A1 (aliases): Every entity/concept page MUST have aliases: [<H1 Title Case>].
  • K1 (inline citation): On entity/concept/synthesis pages, every factual claim gets (source: filename.md) inline. Source pages exempt.
  • K2 (needs verification): Unsourceable claims get [needs verification] — surfaced by next lint pass.
  • F3 (sources array): When updating an existing page, append the current source filename to sources: YAML array.

References

  • skills/wiki-ingest/references/state-json-spec.md — Step 0 + Step N full spec (SHA256, dedup, atomic write, mkdir-lock)
  • skills/wiki-ingest/references/ingest-walkthrough.md — Steps 1–8 pipeline (source page, entity/concept update, wikilinks, index, log)
  • skills/wiki-ingest/references/promote-qa-workflow.md--promote-qa mode (preconditions → mutation → postconditions)

What's Next

After ingesting sources, the user can:

  • Ask questions with /wiki-query
  • Ingest more sources — clip another article and run /wiki-ingest again
  • Health-check with /wiki-lint after every 10 ingests

What ships with it: 3 files

12.2 KB alongside SKILL.md

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.