agentleFS
janezslapnikjanezslapnik/llm-wiki-kb/.claude/skills/kb-intake/SKILL.md

kb-intake

Skill1 starsChanged 39 days agoMITView on GitHub

Finds and triages candidate sources for the knowledge base, writing a curation file in `raw/curation/` that kb-ingest consumes. Use for: find, discover or 'what's out there on' a topic; search from a prospect, or 'what should we search next'; a refresh sweep of the trusted publishers ('refresh the KB', 'freshness sweep', 'what's new from our publishers'); process my inbox or saved links; triage or approve the candidates in a curation file. Not for answering from the KB (kb-query), or for ingesting or re-fetching a known source (kb-ingest).

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read the prospect first: ## Why it matters shapes the gap, ## What to search for the queries.

    line 94
  2. Distinctive:

    Run the Instance inbox command.

    line 105
  3. Very specific:

    After triage, set every presented item status: done.

    line 109

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

kb-intake

Turns a topic, a prospect, a refresh sweep or the inbox into a curation file under raw/curation/, then runs its triage. It never clips or writes wiki pages; kb-ingest does. Light gist fetches for authority and the citation graph are fine. Routes, tools and the identity key come from CLAUDE.md ## Instance; the authority rubric, tiers and reject list from wiki/source-registry.md.

Modes

The mode decides which steps run and when anything is committed.

ModeCueRunsCommits
discovera topicsteps 1–8, triagenone
prospectnames a raw/prospects/ filesteps 1–8 seeded by the prospect, from_prospect: <slug>, triageafter triage, one commit: curation file + git mv of the prospect to past-searched/ + log prospect | resolved <slug>. Abandoned triage → none
refreshrefresh or freshness sweep, what's new from our publishersInstance refresh route over registry T1/T2 rows, steps 5–8 with the refresh rules, from_refresh: true, since: <last-refreshed>, triageone commit before triage: the file, plus the last-refreshed: bump if every T1/T2 row was searched (a row that returned nothing counts; an unchecked row or a subset sweep doesn't; note thin rows in the file). A bump without its file would silently skip a window; log refresh | <since> → <today>
inboxthe inbox, saved linksinbox rules, steps 5–8, from: inbox, triagenone
triagenames or implies an existing curation filetriage onlynone

"What should we search next" → list the open prospects and the unactioned ### Candidate follow-up items of recent Ingest reports, and ask which to run.

Steps

  1. Gap. One sentence on what the wiki lacks; check wiki/index.md and wiki/sources/ with rg. No gap, or one a single-source ingest covers → say so and stop.
  2. Search per the Instance discover route; operator-given URLs are starting points.
  3. Authority. Apply the registry's ## Tier rubric and drop its reject list. A domain you hard-reject as a whole site, not for one weak page, goes under ## Hard rejects (domain — reason); kb-ingest adds it to the reject list.
  4. Citation graph, a distinct pass rather than another search: follow the top few candidates' references by the Instance route; put what you find through step 3.
  5. Dedup on the Instance identity key against one rg -N --no-filename '^url:' wiki/sources/ pass. Matches go under ## Already in KB with their slug.
  6. Tier from the publisher's registry row. Unknown publisher → tier it yourself and add "new publisher: add a provisional registry row at ingest" to the Why.
  7. Sort into Candidates; Exemplars (artefacts better pointed at than ingested; Kind: example-file | example-folder | example-repo); and Borderline: items that pass the hard rejects but carry a named concern (vendor-aligned, single author, polemic, paywalled), listed with the concern rather than dropped.
  8. Write raw/curation/YYYY-MM-DD-<topic-slug>.md from this template:
---
type: curation
slug: curation-YYYY-MM-DD-<topic-slug>
title: Curation wave — <topic> (YYYY-MM-DD)
topic: <one line>
gap: <one sentence>
status: pending
updated: YYYY-MM-DD
---

## Gap
<one sentence>

## Already in KB (auto-skipped)
- `<slug>` — <title> (matched URL: <url>)

## Hard rejects
- <domain> — <reason>

## Candidates

### 1. <Title>
- **URL:** <url>
- **Publisher:** <publisher>
- **Published:** YYYY-MM-DD | living-doc | unknown
- **Tier:** 1 | 2 | 3
- **Status:** pending
- **Why:** <one paragraph>

## Exemplars

### E1. <Title>
- **URL:** <url>
- **Publisher:** <publisher>
- **Tier:** 1 | 2 | 3
- **Kind:** example-repo
- **Status:** pending
- **Why pointer-only:** <one sentence>

## Borderline

### B1. <Title>
- **URL:** <url>
- **Publisher:** <publisher>
- **Tier:** 1 | 2 | 3
- **Status:** pending
- **Concern:** <one sentence>
- **Why include anyway:** <one sentence>

## Triage
(written at triage when pre-authorised)

## Ingest report
(written by kb-ingest)

Prospects

  • Read the prospect first: ## Why it matters shapes the gap, ## What to search for the queries. ## Resolved when is framing, not a test: a searched prospect closes even with zero approved.
  • To open one, write raw/prospects/<slug>.md (type: prospect, slug, title, updated; ## Why it matters, ## What to search for, ## Resolved when, ## Notes; no status:, the folder is the state) and log prospect | open <slug>.

Refresh rules

  • Reject only on topic or for non-articles (tag pages, pagination, product pages); the tier is the row's.
  • Collapse locale and tracking variants via the identity key. Don't let one prolific publisher dominate the wave: keep its best few and note how many were held back.
  • Living-doc rows: their held stubs past the Instance staleness go under Candidates with a Why that begins refresh <slug>:; kb-ingest re-fetches them.

Inbox rules

  • Run the Instance inbox command. On error, show it verbatim and stop; never run an interactive step (such as a login) on the operator's behalf.
  • Work only raw/inbox/ items with status: new; none → say the inbox is empty and stop, no file. Each gets its saved note or one light gist fetch, no citation graph; put the capture date in its Why.
  • A reject-list domain goes to Borderline: the operator saved it, so they decide.
  • The file is raw/curation/YYYY-MM-DD-inbox.md; a same-day file is appended to, numbering continued.
  • After triage, set every presented item status: done.

Triage

  • Summarise once: counts, then every item by its file number (N, EN, BN) with publisher and tier. Ask one question, end the turn and change nothing until the reply.
  • A bare "yes" approves Candidates and Exemplars, not Borderline ("accept B2" promotes one). "approve 1 3" rejects the unlisted numbers; "all except 5" approves the rest; "reject all" rejects everything. An ambiguous reply gets one question, never a guess.
  • Set each - **Status:** to approved or rejected, then the wave's status: approved, or ingested when nothing was approved.
  • Triage by your own judgment only if the operator pre-authorised it in this session; record it in the file's ## Triage section, quoting the authorisation.
  • Triage ends the skill: report counts and offer kb-ingest; don't start it unasked.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.