attaxr-seo-pipeline: Instructions file for Codex

SEO Content Pipeline Agent Instructions

Platform-agnostic workflow — works with Hermes Agent, Claude Code, Codex, OpenCode, or any LLM-based agent framework that supports tool calling (web search, file I/O, terminal, browser automation).

A 4-stage pipeline for systematic SEO content research, analysis, and creation. Outputs local files — no CMS integration.

How This Works

Runs as a stage-by-stage loop, one keyword cluster at a time. Each stage feeds the next. If a stage fails on a specific item, log the error and continue.

Agent Framework Requirements

Capability	Used In	Purpose
Web search API	Stage 1-2	Keyword research, SERP extraction
Browser automation / MCP	Stage 2	Article content scraping
File system read/write	All stages	Data persistence between stages
LLM text generation	Stage 3-4	SEO analysis, content creation
Python execution	Stage 2	Scraping

---

Stage 1 — Keyword Research

1. Use web search to research keywords related to the business domain.
2. For each keyword, capture: phrase, estimated search volume, relevance (1-5), commercial intent, cluster name.
3. Prioritize long-tail keywords with clear commercial intent.
4. Save to keywords.json.

Schema:

[{"term":"keyword","search_volume":3200,"difficulty":72,"relevance":5,"intent":"commercial","cluster":"cluster-name"}]

---

Stage 2 — Content Collection

⚠️ Step 0 — Credential Check (REQUIRED)

Before attempting any scraping, the agent MUST verify that BROWSER_USE_API_KEY is configured:

1. Check if the environment variable BROWSER_USE_API_KEY is set.
2. If not set and a .env file exists, load it via:

   python -c "from pathlib import Path; import os; exec(open(Path('.env')).read().replace('export ',''))"
   

   Then re-check the variable.
3. If still not set, the agent MUST prompt the user interactively:

   "BROWSER_USE_API_KEY is not configured. This is required for article extraction.
    Enter your key (starts with 'bu_'): "
   

4. Security rules for the key input:
   • The key must be captured via a redacted input method — never via plain input() or echo-enabled prompts.
   • The python scripts/setup.py script uses getpass.getpass() which masks input; pass control to it when feasible.
   • The key value must never appear in logs, terminal scrollback, conversation history, or LLM output.
   • If you write the key to .env on the user's behalf, do so with set +o history and sanitize any terminal output.
   • After capturing, write to .env and source it: set -a; source .env; set +a
   • Do not proceed without a valid key.