Knowledge sources

Sources swept for this bootstrap discovery pass

Source	Yields	Re-check by
skills/*/SKILL.md	the six skill Playbooks	re-read each router; a changed step list, guardrail, or frontmatter description is a drift signal
skills/*/references/*.md, skills/ktl-prose/scripts/prose-check.py, skills/ktl-prose/assets/voa-word-book.txt, skills/ktl-docent/scripts/*.sh	detail behind each skill Playbook; the prose skill’s check script, which playbooks/ktl-prose-skill.md describes; and, since 2026-10-05, ktl-docent’s own copies of knowledge-report.sh and knowledge-feedback.sh, which playbooks/ktl-docent-skill.md describes; the references include the librarian’s agent-skills.md and retellings.md and the prose skill’s young-readers.md; voa-word-book.txt is the word list prose-check.py reads under a band of readers	diff against the claims in the corresponding concept body; check 11a holds the docent’s two scripts byte-identical to their templates
skills/ktl-sidecar/templates/	what the sidecar skill actually writes: the toolkit dependency and its [build] extra, the justfile with its schema variable for a domain schema, .gitattributes, gitignore, llms.txt, queries.http, the two READMEs, the knowledge/ skeleton, the scripts (preflight, conventions and its Python half, librarian wrapper, apply and its Python half, report, provenance gate, feedback recorder; since 0.44.0 the federation build knowledge-federate.py; since 0.47.0 the sync knowledge-sync.sh; since 0.49.0 the ingest knowledge-ingest.sh and its Python half), the workflows (registrar, librarian and, since 2026-09-24, release; since 0.44.0 knowledge-federation.yaml with a domain host’s domain.yaml; since 0.47.0 knowledge-sync.yaml; since 0.49.0 knowledge-ingest.yaml; the last two each with a pull request template), since 0.47.0 knowledge-sync.manifest, the list of what a sync keeps current, since 0.49.0 ingest.yaml, the ingest’s manifest, and, also since 2026-09-24, m365/: the Copilot skills’ builder and the docent’s instructions for Copilot	diff pyproject.toml (the lokf floor), the justfile recipes against explanation/domain-schemas.md, and the template lists in the skill’s Step 1 and Step 5 tables
skills/ktl-website/templates/, skills/ktl-website/scripts/website-preflight.sh	what the website skill lays down in a host: the site under templates/website/, which this repository’s own website/ copies, and the deploy workflow knowledge-website.yaml; and the preflight it runs in place. playbooks/ktl-website-skill.md since 2026-10-14	diff the step list, the four placeholders and the three files a host owns against the concept; the whole of the site’s src/ and its scripts are consciously excluded as concepts, as under website/; since 0.53.0 the scripts include raw-html.mjs, which holds raw HTML to what GitHub renders, and playbooks/ktl-website-skill.md names it; the template’s astro.config.mjs, package.json, package-lock.json, tsconfig.json, gitignore and NOTICE are copied to website/ byte for byte, and its README.md and site.config.mjs hold the placeholders Step 1 substitutes, which playbooks/ktl-website-skill.md describes
skills/ktl-sidecar/templates/knowledge-sync.manifest, skills/ktl-sidecar/templates/scripts/knowledge-sync.sh, skills/ktl-sidecar/templates/github/knowledge-sync.yaml, knowledge-sync-pull-request.md	since 0.47.0, how a host keeps its sidecar current by itself: the manifest that lists each template, its copy and the rule for it, which the sync script, the preflight and contract check 11 read; the sync script, which scripts/sync-sidecar.sh runs in each sibling; and the weekly workflow that keeps one pull request refreshed to the newest release. playbooks/releasing.md and playbooks/repository-validation.md describe them	diff the manifest’s rules against check 11 in playbooks/repository-validation.md, and the workflow’s schedule and the script’s steps against playbooks/releasing.md; this repository has nothing to sync itself to, so it carries no copy
docs/ingest.md, skills/ktl-librarian/references/ingest.md, skills/ktl-sidecar/templates/ingest.yaml, skills/ktl-sidecar/templates/scripts/knowledge-ingest.sh, knowledge-ingest.py, skills/ktl-sidecar/templates/github/knowledge-ingest.yaml, knowledge-ingest-pull-request.md	since 0.49.0, the ingest that mirrors knowledge from outside the repository: the manifest .lokf/ingest.yaml, the script that writes each source as Markdown under .lokf/mirror/ with .lokf/ingest.lock.yaml recording what was fetched, the workflow that opens one pull request with what changed, and the librarian’s page on reading and citing the mirror; since 0.50.0 the ingest restores a lost mirror file and prunes on a schedule. playbooks/knowledge-outside-the-repository.md carries it	diff the four kinds of source, the manifest’s refusals, the workflow’s jobs and secrets and the librarian’s rules for the mirror against that concept; this repository has no .lokf/ingest.yaml, so it carries no copy
README.md, docs/architecture.md, docs/install.md, docs/repository-layout.md	project identity: since 2026-10-13 the README is a doorway that keeps the poem, the opening paragraph, LOKF and LinkML as inspirations and the agents note, lists the Exhibitions (the sites that show a repository’s collection) since 0.42.0, with the two Obsidian plugins’ sites among them since 2026-10-16 and the site of KTL Federation, the domain that federates member bundles into one workspace, and points to the website at https://w3id.org/ktl/ and, since 0.43.1, to the site’s pages for the quick start, the retelling and docs/architecture.md, which holds the four-role narrative, the trust ladder, the reading list and the credits; the install commands, prerequisites table and pinning rule, on their own page since 2026-09-19 when the README became a front door, and the quick start since 2026-10-13; the repository tree, likewise on its own page	diff the roles table and the Read on table on docs/architecture.md; diff docs/install.md against references/gh-skill-cli.md and references/open-skills-cli.md; diff the tree against the working copy
docs/architecture.en-simple.md	a retelling of docs/architecture.md for readers aged 13 to 18, which that page links, and whose page on the site the README links since 0.43.1; until 2026-10-13 it was README.en-simple.md and retold README.md	consciously excluded as a concept: a retelling says nothing the page it retells does not say; re-check that it still links docs/architecture.md and that docs/architecture.md still links it
docs/m365.md, skills/ktl-sidecar/references/m365.md	playbooks/docent-in-m365-copilot.md (2026-09-24): getting, building and adding the docent’s Copilot skill, and why only read-only roles go there; the reserved Microsoft 365 identity shape lives in the curator’s domain-schemas.md, carried by explanation/domain-schemas.md	diff the routes, the builder’s limits and the role table against the concept; diff .lokf/m365/ktl-docent-m365.md’s trust-label table against glossary/trust-label.md (check 17 already holds it to ktl-docent’s); re-read Microsoft’s custom-skills page for changed limits while the feature is in preview
docs/for-the-curious.md, docs/obsidian.md	the mechanics docs/architecture.md delegates: who writes the bundle and what checks it, the four levels of checking and the domain-schema escape hatch. Since 2026-09-13, obsidian.md is the human guide to opening the bundle as a vault of its own with the two plugins. Since 0.53.0 for-the-curious.md also says how readers and verdicts steer the next run, with what KTL took from the WikiSkill and SkillWiki papers	diff the four-levels table against glossary/trust-label.md and ktl-curator/references/domain-schemas.md; diff obsidian.md’s two-vault steps against playbooks/open-bundle-in-obsidian.md and ktl-docent/references/obsidian.md, which must agree with it
.claude-plugin/plugin.json, marketplace.json	the Claude Code plugin route to the skills (2026-09-19), six of them since ktl-website (0.39.0), as docs/install.md counts, carried by explanation/why-a-distribution-repository.md	diff the install commands in docs/install.md against that concept; check 3c holds the two keyword lists identical
CONTRIBUTING.md	the contributing playbook	diff the layout table and the pre-PR checklist; since 2026-09-14 the release-process and signing detail live in docs/releasing.md/docs/signing-commits.md instead, and a word-budget check (validate-repository.sh check 10) holds this file to 1000 words
docs/releasing.md, docs/signing-commits.md	the release-process detail (CONTRIBUTING.md used to carry it) that playbooks/releasing.md and policies/versioning.md’s bump rule now derive from; docs/signing-commits.md backs the one-sentence summary in playbooks/contributing.md, no concept of its own	diff playbooks/releasing.md and policies/versioning.md’s Conventional-Commits table against docs/releasing.md’s
docs/three-lines.md	the roles placed in the three lines of defence, what an auditor can check, the critics, quoted, and what a bundle answers and leaves open, and what remains to do and who does it	diff explanation/three-lines-of-defence.md; confirm every external link on the page still resolves and still says what is quoted, and that the verdict table’s links still meet their headings
the WikiSkill paper (Tang et al., 2026) and the SkillWiki paper (Huang et al., 2026)	explanation/wikiskill.md and explanation/skillwiki.md, the record of what KTL took from each paper, citing the paper, CHANGELOG.md and the skill files where each change lives. Since 0.53.0 docs/for-the-curious.md holds what KTL took from the two papers, and docs/wikiskill.md and docs/skillwiki.md are gone	re-check each change a concept lists against CHANGELOG.md and the skill files it cites
docs/skills-repositories.md	whether KTL helps a repository whose content is Agent Skills, with two runs and what remains to do	diff explanation/skills-repositories.md; re-check the run on anthropics/skills
docs/context-layer.md, docs/context-layer-technology.md, docs/federation.md	what a context layer is, one bundle as one repository’s layer, and why KTL federates bundles through domains. Since 0.53.0 docs/federation.md holds the whole federation design, from domains as manifests to the decisions an adopter makes, and the technology page names the products a workspace can use. explanation/context-layer.md summarises the three pages, the glossary term defines the layer, and playbooks/domain-bundle.md cites the federation page	diff the concept; re-check a vendor’s page only where the concept states what it says
website/, .github/workflows/knowledge-website.yaml	the Astro Starlight site built from the README, docs/, the bundle and CHANGELOG.md, and the workflow that publishes it to GitHub Pages, named website.yml until 2026-10-14; playbooks/website.md since 2026-10-12. Since 0.39.0 both are copies of ktl-website’s templates, which check 11w holds byte-identical	diff website/README.md and the workflow’s triggers and jobs against the concept; the whole of src/ and the scripts under website/ are consciously excluded as concepts, and so is the build’s configuration, astro.config.mjs, package.json and its lock, tsconfig.json and .gitignore; playbooks/website.md describes site.config.mjs, the site’s values and options, and the Astro override in package.json
SECURITY.md	the slim security policy (reporting, supported versions, a surface table)	diff the surface table and the reporting/supported-versions text
docs/threat-model.md	the shared threat model (2026-09-14 on, replacing SECURITY.md’s own design section) - repository hardening, the human: attribution gate, prompt-injection guards	diff policies/threat-model.md; its section headings are deep-linked by the sibling repositories, the two plugins’ SECURITY.md and ai-linkmo among them, and check 9 records each anchor, so a heading rename here is a breaking change there
AI_COVENANT.md, CODE_OF_CONDUCT.md	governance policies	diff each; both are adapted from upstream documents that may themselves change
.github/workflows/validate.yml, publish.yml	the validation and releasing playbooks	diff job names, triggers, and the pinned action SHAs
.github/workflows/semantic-release.yml, .github/scripts/changelog-release.mjs, .releaserc.json	the version-and-changelog automation playbooks/releasing.md and policies/versioning.md describe	diff the release job’s steps, the script’s verifyRelease/generateNotes behaviour, and .releaserc.json’s releaseRules (which commit types map to which bump) and the preset it names, since the default Angular preset reads no ! after the type, against the concept’s Overview; the release job runs behind the release Environment, as publish.yml does, while the plan job runs on every pull request with no environment
.github/workflows/knowledge-registrar.yaml, knowledge-librarian.yaml, knowledge-release.yaml, .lokf/scripts/knowledge-librarian.sh, knowledge-conventions.sh, knowledge-conventions.py, knowledge-preflight.sh, knowledge-provenance.sh, knowledge-feedback.sh, knowledge-apply.sh, knowledge-apply.py, knowledge-report.sh, .lokf/m365/knowledge-m365.sh, ktl-docent-m365.md, .lokf/.gitattributes, .lokf/.gitignore	this repository’s dogfooded copies of the workflow templates, scripts and attributes the sidecar skill ships	diff each against its counterpart under skills/ktl-sidecar/templates/; all sixteen are kept byte-identical (check 11 and the preflight report any drift) apart from the two values of knowledge-librarian.yaml’s skills pin, TRUST_LADDER_SKILLS_REF and TRUST_LADDER_SKILLS_SHA: since 2026-09-24 the release commit moves that pin in the template only, because GITHUB_TOKEN may not push under .github/workflows/, and the install step that reads it is skipped here by an if on the repository name, since this repository publishes the skills it uses and the wrapper finds them under bare skills/. Any other difference is a template bump not yet copied across
.lokf/curators/*.asc, *.pub	the curator public keys knowledge-provenance.sh verifies human: confirmations against (the first, noelmcloughlin.asc, registered 2026-09-24)	consciously excluded as concepts: key material, not knowledge; re-check that playbooks/ktl-curator-skill.md and policies/threat-model.md still describe the directory as the gate reads it
.github/ISSUE_TEMPLATE/bug_report.md and feature_request.md, .github/pull_request_template.md, .github/dependabot.yml	contributor intake forms and pin maintenance	consciously excluded as concepts: they are intake forms and maintenance configuration, not knowledge about the repository; re-check only that each template still names all six skills and that its AI_COVENANT.md link is absolute
scripts/*.sh	what the validation playbook claims CI enforces; since 2026-09-24 sync-sidecar.sh, the maintainer’s tool the releasing playbook describes, which is not a CI check; and since 0.39.0 smoke-test-website.sh, which validate.yml’s smoke-test-website job runs	re-read the assertions; a new check is a gap in the playbook; diff sync-sidecar.sh’s refusals and steps against playbooks/releasing.md
tests/*.py, tests/requirements.txt, pytest.ini	the behavioural tests of the sidecar scripts and the prose checker, which ran inside scripts/validate-repository.sh until 2026-10-07; validate.yml runs them in its test-scripts job and publish.yml runs them before a release	consciously excluded as concepts; re-check that playbooks/repository-validation.md still names the job that runs them and the contract checks that moved there
evals/, tests/docent-holdout.yaml, tests/docent-holdout-schema.yaml, tests/test_docent_holdout.py, justfile, .github/workflows/docent-eval.yml	since 0.58.0,1 the docent holdout: the eight captured docent questions and the concept each answer should rest on, as data, with its LinkML shape; the same questions as claude plugin eval cases; the test that checks that each expected concept still exists and that evals/ asks the same questions; the root justfile’s docent-eval and docent-holdout recipes; and the workflow that runs the suite weekly and by hand under a cost cap, never on a pull request2	consciously excluded as concepts: the questions and the concept each expects are kept outside .lokf/, so the docent never reads a question beside its answer,3 and no concept names a question or the concept it expects; policies/threat-model.md and policies/security.md carry the workflow’s guards; re-check that evals/scaffold-bundle.sh still copies the bundle alone
CHANGELOG.md	what changed between releases	read the [Unreleased] section for behaviour changes not yet reflected in concepts
external URLs cited across the repo	the seven Reference concepts; the commit-analyzer’s README and default release rules and the Conventional Commits preset’s parser, which policies/versioning.md and playbooks/releasing.md cite; and the toolkit’s cli.py, which references/lokf-toolkit.md cites	confirm each still resolves and still says what the concept claims
.assets/*.svg, .assets/companions/, scripts/companion-diagrams.mjs	the social-preview card, which docs/architecture.md shows since the README left it on 2026-10-13, the project’s logo mark, the diagrams docs/ pages embed, and the lifecycle diagram, which the README showed until ktl-architecture.svg replaced it. Since 2026-10-19 .assets/companions/ holds the Archify flow diagrams that some figures link beside them, each a light and a dark SVG drawn from its .archify.json, and scripts/companion-diagrams.mjs draws them again when a maintainer runs it by hand, as docs/repository-layout.md says	consciously excluded as concepts: the card and the mark carry no knowledge, and each diagram, a companion too, draws what the page that embeds or links it says; re-check only that each diagram still matches that page
.markdownlint-cli2.jsonc, lychee.toml	the Markdown lint and link-check configuration that validate.yml’s validate-markdown job reads	consciously excluded as concepts; re-check that playbooks/repository-validation.md still names the job that runs them
LICENSE, llms.txt, .gitignore, docs/examples/docent.md, curator.md and prose.md	licensing boilerplate, the agent-facing pointer file, the repository’s ignore rules, and captured skill transcripts (the docent’s, and ktl-prose’s runs for younger readers; the curator’s page is a placeholder)	consciously excluded as concepts: LICENSE is the licence’s standard text, llms.txt restates the README’s note for agents, .gitignore is configuration, and the examples are captured output, not facts about the repository; re-check that llms.txt still matches what the “Agents:” callout in README.md tells agents, and that docs/examples/docent.md’s trust-label claims still match glossary/trust-label.md

Notes for the next run

• This map keeps no account of a run. Each row names a source, what it yields and how to re-check it, as ktl-librarian’s references/bootstrap.md defines the map. What a run did goes in its log lines and its hand-off, and git keeps the accounts this map held until 2026-10-19.
• No Service, Dataset, Table, Metric or AttestedComputation concept exists, and that is correct. This repository holds documentation, agent skills and CI, and publishes a static website: it runs no API, stores no data and defines no measured quantity. playbooks/website.md covers the website.
• No Person or Organization concept. The maintainer is recorded once, as the bundle’s publisher in index.md, and no concept links to a person.
• base_iri is https://w3id.org/ktl/knowledge/, a persistent namespace the project controls. Every concept’s @id is made under it, and the w3id rule redirects each one to the concept’s page on the site, its JSON-LD or its Markdown file. KTL Federation names this bundle by it, so the namespace stays constant across releases, as docs/federation.md asks of a member.

Open questions

• 2026-10-27, human:noelmcloughlin: exclude all of src/, in website/ and in ktl-website’s template: its pages, assets/, content.config.ts, routeData.ts, socialCards.ts and trust.ts are site code like the components and styles, so both website rows should say the whole of src/ is consciously excluded as concepts. playbooks/website.md cites the files it names through website/README.md.

Footnotes

1. CHANGELOG.md, 0.58.0, Added ↩

2. docs/repository-layout.md, the lines for .github/workflows/docent-eval.yml, tests/, evals/ and justfile ↩

3. docs/threat-model.md, The docent eval ↩