interface:
  display_name: "OKF Audit"
  short_description: "Audit content for OKF and agent-readable knowledge quality."
  default_prompt: "Audit this URL or draft for OKF and agent-readable knowledge quality."
policy:
  allow_implicit_invocation: true

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "title": "AMTECH OKF Audit Report",
  "type": "object",
  "required": ["summary", "scores", "findings", "priorityFixes", "remediationPrompt"],
  "properties": {
    "summary": { "type": "string" },
    "scores": {
      "type": "object",
      "required": ["overall", "okfReadability", "entityGraph", "discoveryRendering", "trustSourceQuality"],
      "properties": {
        "overall": { "type": "number", "minimum": 0, "maximum": 30 },
        "okfReadability": { "type": "number", "minimum": 0, "maximum": 5 },
        "entityGraph": { "type": "number", "minimum": 0, "maximum": 5 },
        "discoveryRendering": { "type": "number", "minimum": 0, "maximum": 5 },
        "trustSourceQuality": { "type": "number", "minimum": 0, "maximum": 5 }
      }
    },
    "findings": {
      "type": "array",
      "items": {
        "type": "object",
        "required": ["severity", "title", "detail"],
        "properties": {
          "severity": { "type": "string", "enum": ["high", "medium", "low"] },
          "title": { "type": "string" },
          "detail": { "type": "string" }
        }
      }
    },
    "missingConceptsAndEdges": { "type": "array", "items": { "type": "string" } },
    "materializedViewOpportunities": { "type": "array", "items": { "type": "string" } },
    "priorityFixes": { "type": "array", "items": { "type": "string" } },
    "remediationPrompt": { "type": "string" }
  }
}

MIT License

Copyright (c) 2026 AMTECH

Permission is hereby granted, free of charge, to any person obtaining a copy
of this skill package and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.

Agent-Readable Content Checklist

Use this when reviewing whether a page or package can be consumed by AI systems that may only fetch the exact URL the user provides.

First-Fetch HTML

• The exact URL includes the essential instructions in static HTML.
• The title and meta description name the task, not only the brand.
• The first visible section tells an AI what to do next.
• Links to markdown, JSON, and source files are plain anchors.
• The page works without JavaScript for core content.

Markdown Views

• A short bootstrap markdown file exists.
• The bootstrap says how to use the resource in context.
• The canonical workflow is available as markdown.
• Long references are split and routed by need.

Machine-Readable Views

• A manifest lists all files, roles, hashes, URLs, and load policies.
• Structured data identifies the page as a creative work, tool, software application, or dataset when appropriate.
• Checksums are available for archives and raw files.
• Archives expand to a predictable folder.

Discovery Surfaces

• Sitemap includes public human pages.
• llms.txt or equivalent links to the skill catalog.
• Internal links point from related articles or docs to the skill.
• The page can be found from site navigation or a logical hub.

Trust And Execution

• Scripts are listed separately from references and assets.
• Script purpose, language, permissions, and run policy are explicit.
• Install paths are optional.
• Local instructions and user instructions are given precedence.

AMTECH Knowledge Graph Insights

Use this reference to recommend AMTECH-style improvements after the generic OKF audit.

AMTECH Pattern

AMTECH's strongest publishing pattern is one canonical model with many projections:

• Human React page.
• Static prerendered HTML.
• Markdown or OKF concept files.
• JSON or manifest view.
• Sitemap and discovery files.
• Source package or raw files when the artifact is reusable.
• Optional database projection.

What To Look For

Missing Entities

Identify people, businesses, places, industries, tools, products, services, and use cases that are implied but not named.

Missing Edges

Find relationships that should be explicit:

• solves
• requires
• produces
• cites
• depends on
• compares with
• belongs to
• is useful for

Missing Consumer Views

Recommend views based on who consumes the knowledge:

• Customer: concise human page.
• Search engine: crawlable HTML and metadata.
• AI search: first-fetch summary and clear claims.
• Coding agent: raw markdown, manifest, source tree.
• Security reviewer: checksums, script index, source refs.
• Internal team: canonical registry or database row.

Missing Action Surface

Useful knowledge should tell an agent what to do. Recommend:

• A copy-paste implementation prompt.
• A remediation checklist.
• A local-file creation path.
• A validation command or acceptance test.

AMTECH Voice

Recommendations should be direct, practical, and implementation-oriented. Avoid vague SEO advice. Name the missing file, link, entity, view, or validation step whenever possible.

OKF Audit Rubric

Use this rubric to score articles, sites, drafts, markdown bundles, and OKF-like knowledge packages.

Scoring

Score each category from 0 to 5.

• 0: absent or actively misleading.
• 1: present only as vague prose.
• 2: partially present but not usable by an agent without interpretation.
• 3: usable with gaps.
• 4: strong and mostly complete.
• 5: excellent, explicit, and easy for agents and humans to verify.

Categories

1. First-Fetch Clarity

The exact URL a user shares should tell an AI what the page is, what task it supports, and what to do next.

Strong signals:

• Clear title and H1.
• The opening paragraph names the task and audience.
• If it is a tool/skill, visible agent instructions appear near the top.
• Critical links are in body text, not hidden behind scripts.

2. Concept Packaging

The content should identify the core concept, related concepts, and practical scope.

Strong signals:

• One page or markdown file maps to one durable concept.
• Frontmatter or metadata names title, description, type, tags, updated date, and source URL.
• Definitions are stable enough to cite.
• The page makes clear what is in scope and out of scope.

3. Entity And Relationship Coverage

The content should expose the people, places, industries, tools, use cases, and decisions it relates to.

Strong signals:

• Explicit links to related concepts.
• Entity names are consistent.
• Relationships explain why two concepts connect.
• No important orphan concepts.

4. Source And Citation Quality

Claims should be traceable.

Strong signals:

• External sources are linked near the claims they support.
• Internal sources and source-of-truth files are named.
• Dates are concrete.
• Generated or inferred claims are labeled.

5. Materialized Views

Different consumers need different surfaces.

Strong signals:

• Human HTML page.
• Markdown or text view.
• JSON manifest or structured data.
• Sitemap and discovery links.
• Raw files or source package when relevant.
• Download/archive/checksum when the content is a reusable package.

6. Agent Execution Readiness

An AI should be able to act from the content.

Strong signals:

• The workflow is imperative and ordered.
• Inputs and outputs are explicit.
• Optional references are routed by use case.
• Failure modes and permission boundaries are stated.
• The page includes a copy-paste remediation prompt or action prompt.

Overall Interpretation

• 26-30: agent-native and strong.
• 20-25: usable with targeted fixes.
• 14-19: promising but incomplete.
• 8-13: human-readable but weak for agents.
• 0-7: not agent-ready.

--- name: okf-audit description: Use when auditing an article, website, draft, sitemap, llms.txt file, markdown bundle, or OKF bundle for AI-readable knowledge quality, OKF-style concept packaging, entity coverage, citations, internal links, materialized views, and AMTECH knowledge graph improvements. ---

OKF Audit

Use this skill to audit content for agent-readable knowledge quality. The user may provide a URL, pasted text, article draft, sitemap, llms.txt, markdown bundle, or OKF bundle.

Default behavior: use the skill in the current conversation. Do not install or create files unless the user asks or the environment clearly requires it.

Read Order

1. Read this SKILL.md. 2. Read references/okf-audit-rubric.md when scoring or explaining findings. 3. Read references/agent-readable-content-checklist.md when reviewing first-fetch HTML, snippets, markdown views, manifests, or discovery files. 4. Read references/amtech-knowledge-graph-insights.md when recommending AMTECH-style entity graph or materialized-view improvements. 5. Use assets/report-schema.json only when the user asks for JSON or structured output.

Audit Workflow

1. Identify the submitted surface: article, whole website, draft text, sitemap, OKF bundle, skill package, or mixed input. 2. Summarize what the content is trying to teach or enable. 3. Check whether the first fetched surface gives an AI enough context to act without hunting through adjacent files. 4. Extract core concepts, entities, relationships, audience, tasks, citations, and intended actions. 5. Score the content against the rubric. 6. Identify missing materialized views: HTML, markdown, JSON, manifest, sitemap, llms.txt, raw source, archive, API endpoint, or hosted tool. 7. Recommend concrete fixes in priority order. 8. End with a copy-paste remediation prompt the user can give to an implementation agent.

Output Format

Return:

## Summary

## Score
- Overall:
- OKF/agent-readability:
- Entity graph:
- Discovery/rendering:
- Trust/source quality:

## Findings

## Missing Concepts And Edges

## Materialized View Opportunities

## Priority Fixes

## Copy-Paste Remediation Prompt

If the user asks for JSON, follow assets/report-schema.json.

Safety And Local Rules

• User instructions, repo AGENTS.md, local project rules, and sandbox restrictions override this skill.
• If you can only browse the web, use the fetched skill text in context.
• If you can write files and the user wants reuse, ask before creating a local skill folder.
• If scripts are added in a future version, inspect them and ask before running them.
• Do not claim a site is OKF-compliant unless the required machine-readable files and relationships are actually present.

Source and verification

Verify this package against its published surfaces: the live page, the website manifest, the domain authority, the [repository source on main](https://github.com/benamtech/amtech-skills-registry/tree/main/skills/okf-audit), and the repository catalog.