How to Automate a Weekly SEO Review with Codex and Search Console
Install a repository Codex skill, connect read-only Search Console data, and run a weekly one-page SEO review with a clear no-change outcome.
42 primary guides
Coding & Automation
Practical coding, scripting, agents, and workflow automation with AI assistance.
Start with a job
Guided collections in this topic
Guided collection
Use Codex for Office Files
Choose a checked workflow for spreadsheets, presentations, reports, and PDFs without confusing an editable file with a verified result.
Guided collectionFind, Audit, and Build Agent Skills
Move from a concrete job to package review, a small controlled test, and a reusable Skill with a clear activation boundary.
Main subject
Coding & Automation guides
These guides have Coding & Automation as their primary topic.
GPT-6 Sol and Luna vs Claude Opus 5.5: Which Costs Less to Finish Real Tasks?
From two independent task benchmarks, Luna had the lowest calculated model spend per pass. Opus beat Sol on coding cost per pass; Sol beat Opus on workflows at similar completion rates.
GPT-6 vs GPT-5.6 in Codex: Six Models, Credits, and Token Costs
Compare all six GPT-6 and GPT-5.6 models, see what Codex credits and API tokens cost, and choose a model for coding, agent tasks, and repeated work.
Can One Agent Skill Work in Both Codex and Grok Build?
One SKILL.md can work in Codex and Grok Build. Download the exact Markdown-link checker used in both hosts, compare the complete files, and see which discovery and permission behaviors still need separate tests.
Can an AI Agent Edit an Excel Workbook Without Breaking Formulas? A Univer CLI Test
A Univer CLI test preserved eight formula strings and correct caches through five typed edits, but the export lost the source workbook's checkbox cell-control metadata.
Grok 4.7 vs GPT-6 Astra: Which Coding Agent Finishes the Job?
Codex with GPT-6 Astra and Grok Build with Grok 4.7 fixed the same cross-file Python task. See exact times, tests, diffs, route warnings, and limits.
How to Build and Verify a Python AI Endpoint on Cloudflare's Free Plan
Create a Python Workers AI endpoint on Cloudflare Free, test its route and inference, estimate neuron use, and diagnose quota and runtime failures.
How to Hand Off a Coding Task from Claude Code to Codex with ai-memory
Connect both CLIs to ai-memory, end the Claude Code session, and open Codex in the same checkout. Check the handoff against Git, diagnose missing context, and choose between hooks and managed workstreams.
How to Search Local Markdown Notes with QMD and Verify the Results
Set up a QMD collection, compare keyword and semantic hits, inspect the full source note, and reproduce stale-index and deleted-file checks with a downloadable fixture.
How to Use ChatGPT Voice to Manage Codex Tasks on Desktop
Use ChatGPT Voice to start and follow Codex desktop tasks. Learn how to target the right project, verify a test run, switch tasks, and avoid permission and screen-context surprises.
CLAUDE.md vs AGENTS.md in Claude Code: Which Instructions Actually Load?
Claude Code can read AGENTS.md directly, but a project CLAUDE.md or CLAUDE.local.md changes the default. Check the loader setting, version, and session output before changing your repository.
What Is Jev AI? How to Use TypeSafe's Decision Model
Jev answers predefined questions about text with typed choices, scores, or yes probabilities. Learn when that helps, how to make a first call, and how to check a routing workflow against your own cases.
Claude Opus 4.6 vs Gemini 3.8 Flash: Which Wins for Coding Agents?
Choose between Claude Opus 4.6 and Gemini 3.8 Flash with current prices, IDE billing considerations, six coding scenarios, repository checks, and private-code data policies.
Grok 4.6 vs Gemini 3.8 Flash for Building an Interactive Web App
Compare Grok 4.6 and Gemini 3.8 Flash for one interactive web app using current WebDev results, API limits, response speed, cost, and a matched acceptance test.
How to Install and Test DeepSeek Harness Safely
Install DeepSeek Harness in a disposable VM, pin the reviewed dsh version, disable telemetry, start read-only, and verify both a denied mutation and one controlled workspace write.
GLM 5.3 Flash Free vs Paid: Current Routes, API Price, and Cost per Accepted Task
Compare the GLM-5.3-Flash trial, Coding Plan, metered API, and open weights. Check ZCode workspace privacy before a private trial and calculate cost per accepted task.
How to Save Codex Usage by Planning in ChatGPT First
Save Codex usage by doing evidence gathering and planning in ChatGPT first, then handing Codex a narrow, testable implementation brief. Check your usage boundary before relying on the method.
Gemini 3.8 Flash vs 3.7 Flash: A Coding Agent Test
A careful reading of Gemini 3.8 Flash versus 3.7 Flash on long-horizon coding agents, including pass rate, tool steps, output tokens, cost per solved task, and a practical migration test.
Claude Fable 5.1 vs GPT-6 Astra for Long-Running Work
Build a fair Claude Fable 5.1 versus GPT-6 Astra evaluation for long-running work, with a representative task, hard failure gates, cost math, and an evidence-based decision rule.
How to Connect Codex to Grok Bot with a Safe Webhook Bridge
Send tasks from Codex to a Grok Bot Routine, collect replies manually, and check the returned work. Includes task cards, six use cases, and timeout recovery.
16 AI Coding Models Compared: Accuracy, Speed, and the Cost of Being Wrong
A practical reading of a 16-model coding benchmark that separates solve rate, latency, list price, realized cost, and the downstream cost of a wrong answer.
How to Benchmark GPT-Live-1 vs GPT-Realtime-2.1 for Voice Interruptions
A developer guide to paired voice-interruption tests: capture one pilot pair, review eight conversation cases, and check playback, understanding, and pending actions.
A Screen Recording Is Not a Spec: Rebuilding an Interface State by State
A five-second interface demo hid more than it showed. Here is how we extracted six states, constrained a coding agent, caught a plausible wrong turn, and tested the result.
We Built a Tiny Browser Game With AI—Then Tested What “Playable” Means
Learn how to make a browser game with AI through a real small build: choose the stack, define the loop, prompt one milestone, test difficult states, and publish carefully.
Also relevant
Related guides from other main topics
These overlap with Coding & Automation, but their main subject belongs elsewhere.
How to Audit Agent Permissions Using OpenAI's Model Misalignment Reports
Audit an AI agent's effective authority by turning OpenAI's misalignment reports into staging tests with observable pass conditions and external evidence.
Why it fits here: The audit turns permission concerns into agent staging tests.
A Practical GPT-6 Astra Prompting Guide for Long, Tool-Heavy Work
Use one worked release example to brief GPT-6 Astra for long-running work, set budgets and approval limits, handle changes, and verify the handoff.
Why it fits here: The prompt patterns target long tasks and tool-heavy agent work.
When to Use One AI Agent vs Multiple Agents
A practical decision guide for knowing when subagents create useful parallel work and when one agent will be faster, cheaper, more coherent, and easier to verify.
Why it fits here: The choice changes the architecture of an automated workflow.
How to Test an AI Browser Agent Safely
Learn how to test a browser agent with isolated state, effect-based action boundaries, adversarial pages, evidence-matched assertions, and a release gate that preserves failures.
Why it fits here: The guide sets a controlled test boundary for an automated agent.
How to Evaluate RAG for Document Question Answering
Evaluate document Q&A RAG with a runnable kit, a retained model trace, citation review, and questions the corpus cannot answer.
Why it fits here: The evaluation concerns a retrieval-augmented question-answering system.
How to Test an AI Workflow for Indirect Prompt Injection
Test indirect prompt injection with a clean baseline, delivered hostile source, synthetic canary, independent action trace, and explicit pass, fail, and inconclusive decisions.
Why it fits here: The test exercises an application that consumes untrusted input.
How to Build an AI Evaluation Set That Catches Regressions
Build a compact AI evaluation set with explicit labels, must-preserve details, forbidden claims, deterministic grading, and a release decision that exposes failures instead of averaging them away.
Why it fits here: The regression set is used to test an AI workflow during development.