Claude Code SEO Workflow: Terminal Agents With Ranked Orders

Why Claude Code needs connected orders

Claude Code excels at repo edits, test runs, and multi-file refactors. SEO work still defaults to side chats with pasted CSV exports. The agent guesses priorities from training data. Monday repeats. The Claude Code SEO workflow replaces guesswork with structured JSON from your asset: Mission Brief orders, opportunity backlog, credit balance.

Cursor SEO workflow covers the same architecture for Cursor users. This article targets Claude Code terminal patterns and project instruction files.

Architecture: REST API, CLI, human publish

Setup: token, CLI, first brief pull

1. •Create API token in app settings. Export LEARN_DOMAINS_TOKEN in shell profile.
2. •Install @learndomains/cli. Verify with learn-domains me --json.
3. •Add CLAUDE.md or project instructions: always pull Mission Brief before SEO tasks.
4. •Run learn-domains mission-brief latest --website yourdomain.com --json.
5. •Optional: learn-domains opportunities list --json for full backlog context.
6. •Check learn-domains credits balance --json before batch content jobs.

Activation path still applies: website added, Search Console connected, GA4 connected, Knowledge Base populated. Agents cannot raise Confidence without syncs.

Weekly agent loop

Monday: ingest Mission Brief JSON into Claude Code context. Assign order one with URL, query cluster, ICEE rationale. Tuesday to Thursday: technical fixes in repo, Content Operations draft for refresh orders, AI Analyst in app when Confidence is medium.

Content create via CLI or REST returns draft IDs. Agent prepares CMS-ready markdown. Human publishes manually. Mark published in app after live. Auto-publish would break Execution discipline and brand QA.

• Never invent keyword priorities not in brief JSON.
• Never skip credit check before content create or analyst-equivalent batch jobs.
• Never bypass human review on money pages.
• Re-pull brief after major sync, not hourly without shipping.

Example agent prompts that work

Good prompt: Given this Mission Brief JSON, implement order one technical acceptance criteria in /docs, then request a Content Operations draft for the refresh URL with diagnosis notes from the brief. Bad prompt: Improve our SEO this week. The first binds evidence. The second hallucinates a backlog.

For striking-distance orders, agent expands outline sections named in brief evidence. For cannibalization orders, agent drafts redirect map and internal link diff before copy changes. Codex terminal SEO workflow shares JSON patterns for OpenAI Codex users.

Knowledge Base and URL Library grounding

Agents without memory produce interchangeable SaaS slop. Populate Knowledge Base with ICP, pricing boundaries, and voice rules. Paste relevant vault excerpts into Claude Code context or maintain repo copies synced with your vault.

Internal link orders must use URL Library targets, not improvised exact-match anchors. Knowledge base advantage article explains why memory beats longer prompts.

Asset Yield and growth orders in agent context

When revenue connects, CLI asset-yield commands return summary JSON for commercial triage. Growth orders materialize from Mission Brief top actions. Agents can prioritize money URLs when brief includes revenue-weighted Impact.

Revenue attribution for SEO work explains investor-grade framing. Agents should cite orders, not invent ROI guarantees.

Stakeholder communication without dashboard theater

Leadership asks why traffic dropped. Dashboards invite a sixty-minute chart walk. Intelligence answers with three diagnosed URLs, root cause tags, and the order shipping this week. AI Analyst in Learn Domains cites connected queries when you need a memo with evidence. Keep memos short: drop type, top losers, action taken, next measurement date.

Investors and clients respect closure language. Orders shipped beats impressions explained. Organic traffic dropping playbook pairs with revenue attribution for SEO work when money pages are in the loser list.

Multi-repo and monorepo patterns

Marketing sites in monorepos can wire Claude Code to pull brief JSON then open PRs on content paths directly. Separate marketing repos should store .growth logs beside MDX or CMS export folders. Keep website id or domain consistent in CLI flags across repos to avoid wrong-asset orders.

Technical SEO orders often touch app router config, redirects, or sitemap generators in code. Content orders stay in Content Operations. Mixed orders should split into two agent tasks with explicit boundaries so code agents do not rewrite blog prose in middleware files.

Security and token hygiene for agents

API tokens belong in environment variables, never in committed repos. Rotate tokens when contractors roll off. Scope tokens to the minimum website access required. Agents should read tokens from shell env, not from markdown files checked into git.

Rate limits apply to REST API the same as the app. Batch jobs should backoff on 429 responses. Log token id prefix in execution JSON for audit, never the secret itself.

Troubleshooting agent sessions

When agents ignore brief JSON, tighten project instructions: mandatory file read of .growth/brief-latest.json before edits, cite order id in PR title, refuse tasks without evidence block. When credits fail unexpectedly, verify token org matches website org. When drafts sound generic, paste Knowledge Base excerpts into the session and regenerate.

Autonomous website AI agents article sets category expectations. This workflow stays human-in-the-loop by design.

Limits and operator contract

Claude Code plus Learn Domains is not autonomous SEO. No silent cron publishes. No third-party keyword database in product today. You own keeper URL decisions and credit budget.

Start with one agent task per week tied to order one. Expand automation only after execution logs show consistent publish and measurement discipline. Agents amplify good ops. They multiply chaos when the queue is unranked.

Full REST API and CLI reference lives at /docs/api and /docs/cli. Mission Brief Method defines ICEE vocabulary agents should echo back in commit messages and PR descriptions.