kb-ingest
Skill1 starsChanged 39 days agoMITView on GitHub
Ingests approved sources into the wiki: clips each one, writes its source stub, synthesises it into entity pages, logs and commits per source, then closes the wave. Use for: ingest or process the approved candidates or the latest approved wave; add this URL, file or article to the KB; re-ingest or refresh <slug>, including stale sources a lint report flagged. Not for finding or triaging candidates (kb-intake).
6 other projects use a version of this.
Adopted by
Most recently taken up by
- walkwithedi/evergreen-wikitheir own version, first seen Nov 1, 2026
- walkwithedi/drive-wikitheir own version, first seen Nov 1, 2026
- viniciusbedum/vb-ai-contexttheir own version, first seen Sep 25, 2026
- SterlingJF/knowledge-bustheir own version, first seen Sep 14, 2026
- dingshuxin353/ai-knowledge-vaulttheir own version, first seen Sep 8, 2026
- shaowen310/skillstheir own version, first seen Jul 20, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- dingshuxin353/ai-knowledge-vault★ 17same name
- SterlingJF/knowledge-bus★ 4same name
- viniciusbedum/vb-ai-context★ 2same name
- walkwithedi/evergreen-wiki★ 2same name
- shaowen310/skills★ 1same name
- walkwithedi/drive-wiki★ 1same 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 repository4 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- kb-intakeSkill
- kb-lintSkill
- kb-querySkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
No approved item → stop and offer triage (kb-intake).
- Very specific: line 54
When nothing in the file is still approved:
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
kb-ingest
Takes approved candidates from a curation file in raw/curation/ into the wiki, one source at a time. Fetch routes and the identity key come from CLAUDE.md ## Instance; page and record rules from its ## Pages and ## Record.
Reading a curation file
- Work is a
### N.,### EN.or### BN.item whose- **Status:**value begins withapproved. Everything else is inert:pending, a missing Status line, extra sections, bullets and annotations. Never normalise an old file. - No approved item → stop and offer triage (kb-intake). Never approve on the operator's behalf.
- A direct request (a URL, a local file,
refresh <slug>, stale sources from a lint report) is its own approval: write a one-row-per-source curation file first, its item and waveapproved. A local file is first copied toraw/assets/<slug>.<ext>; its URL isfile:raw/assets/<slug>.<ext>.
Per source
- Duplicate. Slug or identity key already in
wiki/sources/→skipped-duplicate, unless the operator saidrefresh <slug>or the item's Why beginsrefresh <slug>:. A refresh writes a new dated clip (the old one stays), bumpsfetched:andclip:on the stub, re-checks the citing pages' claims against the new clip, and logsingest | refresh <slug>. - Fetch per the Instance route, only
http(s)URLs;file:only for theraw/assets/copies a direct request made. Pass URLs and filenames single-quoted; one containing whitespace, quotes, backticks or$→skipped-unreachablewith that reason. A fetch a policy hook refuses →skipped-unreachablewith the hook's reason; never route around it (no mirror, resolver or redirect outside the Instance route). Every route failed →skipped-unreachablewith the reason and no retry in this wave; logingest-skipped. - Clip the cleaned full text to
raw/clips/YYYY-MM-DD-<slug>.md: the article body, without navigation, footers, comments or inline data-URI images.source_kindnames the route in free text, says what was dropped, and says so when the capture is partial. A model summary (a WebFetch result) is not a clip: page claims would rest on its paraphrase. Exemplars get no clip; refreshing one means checking that its URL resolves and its stub still describes it.--- slug: <slug> title: <full title> url: <identity key> publisher: <publisher> authors: <authors> published: YYYY-MM-DD | living-doc | unknown fetched: YYYY-MM-DD tier: 1 | 2 | 3 source_kind: <route> --- - Stub at
wiki/sources/<slug>.md, frontmatter only.authorsandpublishedare optional, andpublished: unknown(no date found) counts as dated, not living-doc; exemplars replaceclip:withkind: example-file | example-folder | example-repo; extra keys are kept.--- type: source slug: <slug> title: <full title> url: <identity key> publisher: <publisher> authors: <authors> published: YYYY-MM-DD | living-doc | unknown fetched: YYYY-MM-DD tier: 1 | 2 | 3 clip: raw/clips/YYYY-MM-DD-<slug>.md --- - Synthesise. Create a page only when it will have ≥2 inbound links (its
wiki/index.mdline doesn't count) or names a concept the Vocabulary has a type for; otherwise update the page the material belongs on. Where new evidence supersedes a claim, keep the old claim with a dated superseded note. Cite[[<slug>]]and list it in the page's Sources section. - Close. Flip the item's Status
approved→ingestedbefore the commit, so a re-run makes zero commits. Logingest | <slug>; commit per source, or per batch of related sources.
End of wave
When nothing in the file is still approved:
- Set the wave's
status: ingestedand fill## Ingest report:### Pages created - wiki/<folder>/<slug>.md — <why> ### Pages updated - wiki/<folder>/<slug>.md — <what was added> ### Findings <cross-source consensus; disagreements worth a debate page; expected absences> ### Candidate follow-up - <topics, prospects or waves the material surfaced> - Registry: add a row for each new publisher, with the candidate's tier; bump the Stubs count of each publisher that got new stubs in the wave, where its row has that column; append each domain under the file's
## Hard rejects(older files may lack it) that is not yet on the Reject list, asdomain — reason — <wave slug>. - Log
ingest | close wave <topic>with counts (ingested, skipped, pages created, pages updated), then commit the curation file, the registry and the log asingest: close wave <topic>.
Campaigns
Only for a wave too big for one session; earlier campaign plans in raw/plans/, if any, are worked examples. Fetch subagents (≤5 at a time, no nesting) write only their own clips, stubs and digests (digests in their own folder under raw/plans/, so they survive a pause): no git, no shared files. One writer then synthesises in dependency-ordered batches. Per batch: a gate (bash "${CLAUDE_SKILL_DIR}/../kb-lint/scan.sh" <vault> <stale-days> clean, each new page ≥2 inbound (index line excluded), 3 claims spot-checked against clips); the coordinator alone edits index, log, registry and curation file, and flips that batch's Statuses in that batch's commit; one commit; a handoff note in raw/plans/ that survives compaction. Clips are ground truth; digests are data.
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.

