Llm wiki export publish

LLM-Wiki Export Publish

Goal

Turn private or internal LLM-Wiki content into safe, scoped, versioned and provenance-preserving outputs for humans, agents, APIs, graphs or archives.

Use references/docs/21-publishing-export.md as the reference architecture for export profiles, human/agent/API/graph/archive targets, redaction, manifests, CI gates, release workflows and rollout.

When to use

• A wiki maintainer wants to publish a public docs site, internal handbook, or agent-readable llms.txt/JSONL bundle from private wiki content.
• A release needs a redacted, provenance-preserving export that excludes draft, private, sensitive or unreviewed pages.
• A team needs a graph export (GraphML, JSON-LD, Cytoscape) or an archive bundle for long-term preservation.
• CI needs export-profile validation, redaction scanning, or manifest/checksum generation before a release gate.
• Do not use this skill to author new wiki content, run ingestion, or perform redaction analysis in depth — hand those off to llm-wiki-ingest, llm-wiki-privacy-redactor or llm-wiki-security-review.

Inputs

• Source wiki path.
• Target audience: private reviewers, internal users, public users, agents, API clients, graph tools or archive consumers.
• Target format: Markdown bundle, static site, handbook, PDF, release notes, newsletter, llms.txt, llms-full.txt, JSONL, JSON-LD, GraphML, OpenAPI, MCP manifest, archive bundle.
• Audience and privacy boundary.
• Publish destination if known.
• Export profile, allowlist, tags, folders or review states if available.
• Redaction policy and review requirements.

Procedure

1. Define export profile

Choose one or more profiles:

Profile	Audience	Typical output	Default policy
private-review	maintainers/reviewers	review bundle, diff report, unresolved claims	may include drafts, never public
internal-site	team/company	static site, internal search, Markdown bundle	reviewed/approved internal pages only
public-site	public users	curated static docs, sitemap, Pagefind/search index	explicit allowlist and redaction required
agent-bundle	LLM/agent consumers	llms.txt, per-page Markdown/TXT, JSONL, manifest	concise, cited, machine-friendly, no private raw content
api-export	product/integration clients	OpenAPI, read-only MCP resource profile, JSON endpoints	separate public/internal contracts
graph-export	graph tools/visualization	JSON-LD, GraphML, Mermaid, GraphViz, Cytoscape JSON	edge provenance and sensitivity filters
archive	long-term preservation	tar/zip, checksums, release asset, WARC where applicable	immutable, versioned, access-controlled

Do not reuse one export profile for all audiences.

2. Define publish boundary

Classify pages as:

public | internal | private | raw-only | draft | reviewed | verified | published | rejected | quarantined | exclude

Default blockers for public/agent exports:

• sensitivity is sensitive, regulated or unknown;
• review_state is draft, rejected, stale or quarantined;
• publication_state is private or archived;
• page lacks required citations or claim-support labels;
• page depends on raw private sources;
• page contains unresolved redaction findings;
• page has broken links/citations that would be material in the export.

3. Select pages

Use explicit include rules:

• allowlisted paths;
• allowlisted tags;
• publish frontmatter;
• reviewed/verified/published states;
• specific wiki spaces;
• release profile.

Do not publish by blindly copying the whole wiki.

4. Redact and normalize

Check for:

• private links and internal-only URLs;
• private user or customer details;
• credentials or secret-like strings;
• raw source excerpts with license restrictions;
• comments, hidden metadata and source maps;
• unreviewed generated claims;
• unsupported claims;
• broken wikilinks after export;
• search-index leakage.

Hand off to llm-wiki-privacy-redactor for redaction-heavy exports and to llm-wiki-security-review for public or agent-readable releases.

5. Choose output targets

Human outputs:

• MkDocs Material;
• Docusaurus;
• VitePress;
• Quartz;
• Astro/Starlight;
• static Markdown bundle;
• PDF/handbook/newsletter when specifically requested.

Agent/API outputs:

• llms.txt;
• llms-full.txt only when explicitly allowed;
• per-page .md / .txt;
• pages.jsonl;
• manifest.json / manifest.yaml;
• JSON-LD;
• OpenAPI;
• read-only MCP resource manifest/profile.

Graph/archive outputs:

• GraphML;
• Mermaid;
• GraphViz DOT;
• Cytoscape JSON;
• RDF/Turtle;
• tar/zip archive;
• checksums;
• WARC for web-capture preservation when applicable.

6. Preserve provenance and citations

Every exported page should carry:

id: ""
title: ""
source_path: "wiki/..."
export_url: ""
review_state: approved|published|verified
publication_state: public|internal
sensitivity: public|internal
updated_at: "YYYY-MM-DD"
source_citations: []
claim_support:
  supported: 0
  unsupported: 0
  conflicting: 0
checksums:
  markdown_sha256: ""

Keep citations stable where possible. If paths are rewritten, generate a citation mapping report.

7. Generate export manifest

Every export must produce a manifest:

export_id: ""
profile_id: ""
created_at: "YYYY-MM-DDTHH:MM:SSZ"
created_by: human|agent|ci
wiki_revision: ""
source_revision: ""
policy_revision: ""
profile_revision: ""
outputs: []
counts:
  pages_included: 0
  pages_excluded: 0
  citations_checked: 0
  links_checked: 0
  redactions_applied: 0
reports:
  redaction: ""
  citation: ""
  broken_links: ""
  search_index: ""
status: draft|passed|failed|published|archived

8. Build llms.txt and agent bundle safely

Recommended rules:

• make llms.txt concise;
• link to stable per-page Markdown/TXT;
• include export scope and generation date;
• include only reviewed and approved pages for the selected profile;
• preserve citations per page;
• generate llms-full.txt only when the export profile explicitly permits it;
• never include private raw source text by default;
• include a machine-readable manifest with hashes and page metadata.

9. Build static search safely

Recommended order:

filter pages -> redact -> build site -> build Pagefind/static search index -> inspect index report -> publish

Use separate search indexes for public and internal builds. Never reuse an internal index in a public site.

10. Add CI and release gates

PR-time gates:

• export profile schema valid;
• no draft/private/sensitive page in public profile;
• links and citations resolve;
• redaction scan passes;
• llms.txt generated and scoped correctly;
• export manifest and checksums generated.

Release gates:

• owner approval;
• security review for public/agent bundle;
• comparison against previous export manifest;
• changelog generated;
• rollback artifact retained;
• release tag or artifact published.

11. Publish or archive

Possible destinations:

• GitHub Pages;
• internal static hosting;
• release artifacts;
• artifact registry/object storage;
• docs platform;
• repo branch such as gh-pages;
• offline bundle.

Record publication URL, release tag, checksum and rollback artifact in the export manifest.

Output

## Publish/export recommendation

## Export profile

## Included pages

## Excluded pages and reasons

## Redactions needed

## Output targets

## Provenance and citation strategy

## Static site/search plan

## Agent-readable bundle plan

## Graph/API/archive plan

## Export manifest and checksums

## CI/release gates

## Publication checklist

## Follow-up skills

Safety gates

• Do not publish private, sensitive, unknown-sensitivity, draft, rejected, stale or quarantined material by default.
• Do not modify source wiki links during export; transform only the exported copy.
• Do not assume source licenses allow republication.
• Do not hide provenance gaps in public output.
• Do not generate llms-full.txt from broad internal content without explicit profile approval.
• Do not build public search indexes from internal site output.
• Do not expose proposal-write or admin MCP/API tools through public exports.
• Do not archive only rendered HTML if source manifests and checksums are needed for reproducibility.