Coding Agents

Setup, configuration, and environment for AI coding agents: Claude Code, Pi, and others.

Topics

• Agent Instructions: How to author a repo's AGENTS.md and its CLAUDE.md pointer: what goes in each, how to organize the rules as a sequential workflow, source them from real friction, and what belongs as prose vs. a runnable check. Also when to read them: at the start of every session and after every compaction. Use when setting up a new repo or maintaining standing agent instructions.
• Claude Code Settings: Configure and sync Claude Code settings (default model). Use when setting up Claude Code or syncing recommended settings to the current machine.
• Claude Code Permissions: Configure and sync Claude Code tool permissions (allow, ask, deny). Use when setting up permissions or syncing the recommended permission lists to the current machine.
• Claude Code Status Line: Custom status bar showing context usage, model, directory, and repo@branch. Use when setting up or syncing the status line script.
• Claude Code Remote Workflows: Control Claude Code sessions remotely from the iPhone Claude app — check in on sessions, spawn new ones, or run an always-on daemon. Use when setting up or using remote control for Claude Code.
• Claude Code Done Marker: Mark sessions as complete with /done so tooling can detect finished sessions via a single grep against the JSONL transcript. Use when you want to identify completed sessions without relying on natural-language gratitude.
• Update AI Agent CLIs: Update Claude, Codex, Copilot, Antigravity, and Pi CLI tools to their latest versions via Volta. Use when the user wants to update their AI coding tools.
• LLM Environment Tools: CLI tools that improve LLM coding agent performance (ripgrep, fd, fzf, DuckDB, git-delta, xh, watchexec, just, semgrep). Use when setting up or auditing a development environment for AI agent effectiveness.
• Pi Coding Agent: Setup, CLI commands, extensions, and workflows for Pi, the open-source minimal terminal coding agent. Use when working with Pi or comparing it to other coding agents.
• Pi Permissions: Configure Pi tool permissions via the guardrails extension (confirm, deny, file policies). Use when setting up permissions or syncing the recommended permission config to a machine.

See also

• Evals: Testing agent behaviors with simulated user conversations, structured grading, and eval reports.
• Begin Work: Craft a prompt that gives an agent everything it needs to start or continue work on a GitHub issue.
• Claude GitHub Action: Run Claude inside a GitHub repo via the official Action — @claude mentions and the claude label trigger workflow runs that respond, implement changes, and open PRs.
• Prompt Engineering: Patterns for structuring prompts and workflows that keep AI agents focused and verifiable.
• Agent Instructions: Mid-session directives that correct, redirect, or enhance agent behavior.
• Browser Automation: Automate web workflows via Playwright — CLI commands, site interaction, authentication, and file handling.