AI Features

AI assistants, agent adapters, skills, and automated pipelines available in this docs repo.

AI Features

This repo has AI capabilities at three levels: reader-facing assistants, contributor-facing agent tooling, and automated pipelines that keep content fresh without human intervention.

For governance and registry details, see docs-guide/tooling/ai-tools.mdx.

AI Assistants for Readers

Mintlify AI Assistant

Built into docs.livepeer.org. Uses agentic RAG with tool-calling to answer questions from the published docs. No configuration required.

Conversation analytics are available in the Mintlify dashboard and surface unanswered questions as content gap signals.

llms.txt

This repo governs llms.txt as a committed AI-first public root artifact. It is generated from the repo docs navigation by operations/scripts/generators/ai/llm/generate-llms-files.js. llms-full.txt is retired and not part of the active root contract.

MCP Server

Mintlify provides an auto-syncing MCP server that lets Claude, Cursor, and Windsurf query the published docs in real time:

npx mcp add livepeer

Agent Adapters

AGENTS.md is the baseline instruction set. Every AI agent reads it first, then loads its native adapter for tool-specific rules.

Agent	Adapter	Scope
Claude Code	.claude/CLAUDE.md	Full project context, session management, co-work skills
Cursor	.cursor/rules/repo-governance.mdc + no-deletions.mdc	Governance rules, deletion guards
Windsurf	.windsurf/rules/repo-governance.md	Governance rules
GitHub Copilot	.github/copilot-instructions.md	Inline code suggestions
Augment Code	.augment/rules/ (3 files)	Governance, git safety, deletion guards
Codex	.github/AGENTS.md	Task isolation, checkpoints, lock contracts

Codex Branch Contract

AI agent implementation branches use codex/<issue-id>-<slug> naming. These branches are hook-enforced with:

• Task contract file: .codex/task-contract.yaml
• Pre-push validation: contract + scope + non-fast-forward block
• PR body generation: node operations/scripts/dispatch/ai/codex/create-codex-pr.js --create

Full policy: docs-guide/policies/agent-governance-framework.mdx

AI Skills

28 skills provide structured workflows for both humans and agents. Invoke with /skill-name in Claude Code or follow the SKILL.md instructions directly.

Co-work and Process (9 skills)

Skill	Purpose
/thread	Session anchor, drift prevention, finalisation reports
/pm	Project management, priorities, bottlenecks, risks
/research	Agent-delegated investigation, synthesise and recommend
/design	First-principles co-creation, architecture with test criteria
/build	Agent-delegated execution against approved design
/iterate	Review against criteria and best practice
/dispatch	Parallel agent coordinator for batch work
/agent-brief	Standard instruction template for spawned agents
/diagnose	Systematic debugging, root-cause analysis

Content Pipeline (4 skills)

Skill	Purpose
content-pipeline-pass-a	Content layer writing (audience, voice, veracity flags)
content-pipeline-pass-b	Structure and layout pass
content-pipeline-tab-map	Tab-level IA mapping
page-authoring	Single-page authoring workflow

Audit and Governance (8 skills)

Skill	Purpose
repo-audit-orchestrator	Full repo audit coordination
docs-coverage-and-route-integrity-audit	Navigation and routing coverage
docs-quality-and-freshness-audit	Content quality and staleness
script-footprint-and-usage-audit	Script usage analysis
component-layout-governance	Component placement rules
generated-mdx-banners-governance	Generated file banner integrity
cleanup-quarantine-manager	Quarantine and cleanup workflows
cross-agent-packager	Cross-agent context packaging

Review and Specialised (7 skills)

Skill	Purpose
docs-review-packet-generation	Review packet assembly
docs-review-fix-execution	Fix execution from review findings
rubric-static-review	Rubric-based static content review
docs-copy	Copy editing pass
style-and-language-homogenizer-en-gb	UK English style enforcement
product-thinking	Product-oriented analysis
skill-docs	Skill documentation generation

Skill Chaining

Skills chain in a standard workflow: /thread (anchor) then /pm (plan) then /research (investigate) then /design (architect) then /build (execute) then /iterate (review).

Portable Templates

10 portable skill templates live in ai-tools/ai-skills/templates/. These can be synced to Codex or other environments:

node operations/scripts/dispatch/ai/agents/sync-codex-skills.js
# or: bash tools/lpd setup --yes (includes skill sync)

AI-Powered Pipelines

Glossary Companion Generation

Extracts glossary data from SearchTable components in MDX pages and writes companion JSON files for AI and crawler discoverability.

• Generator: operations/scripts/generators/content/reference/generate-glossary-companions.js
• CI generate: .github/workflows/generate-ai-companions.yml (push to deploy branch)
• CI check: .github/workflows/check-ai-companions.yml (PR gate)

Content Health Automation

Weekly scheduled workflow (content-health.yml) scans for stale pages, broken links, and content drift. Outputs reports to workspace/reports/.

Language Translation

Translation pipeline for multi-locale content. Frontmatter taxonomy validator at tools/lib/docs/frontmatter-taxonomy.js enforces valid locale values.

Contributing AI Tools

Write a new skill

1. Copy a template from ai-tools/ai-skills/templates/
2. Create ai-tools/ai-skills/your-skill/SKILL.md
3. Follow the SKILL.md format: metadata block, when to use, steps, principles
4. Register in ai-tools/registry/ai-tools-registry.json
5. Run the validator: node operations/scripts/validators/governance/compliance/validate-ai-tools-registry.js --check

Add rules

Rules live in ai-tools/ai-rules/:

• best-practices.md - general AI agent best practices
• HUMAN-OVERRIDE-POLICY.md - when humans can override AI safety rules
• ROLLBACK-GUIDE.md - how to roll back AI-generated changes
• rules/ - granular rule files

Agent packs

Pre-packaged agent configurations in ai-tools/agent-packs/ for Claude, Codex, Cursor, Windsurf, and portable skills. Generated from templates and registry data.