Agentic Configuration

Personal instructions, skills, and harness configuration shared across coding agents.

Layout

.agents/
  AGENTS.md          # canonical, harness-neutral global instructions
  skills/            # portable Agent Skills (`<name>/SKILL.md`)
.claude/
  CLAUDE.md          # imports ../.agents/AGENTS.md
  skills -> ../.agents/skills
  settings.json      # Claude-only model, permissions, hooks, UI, and worktree settings
  remote-settings.json

~/.agents points to this repository's .agents directory and ~/.claude points to its .claude directory. Portable behavior belongs in .agents; harness-specific configuration belongs in the corresponding harness directory.

Install

git clone git@github.com:wbbradley/agentic.git ~/src/agentic
~/src/agentic/install.sh

The repository name matches its canonical checkout at ~/src/agentic; the hidden subdirectories provide each harness's home-directory configuration.

Codex discovers skills from ~/.agents/skills. Its global instruction file is normally ~/.codex/AGENTS.md; link that file to the canonical source when needed:

ln -s ~/.agents/AGENTS.md ~/.codex/AGENTS.md

Portability Rules

• Skill frontmatter contains only name and description.
• Skill bodies describe capabilities, not harness-specific tool names or model tiers.
• The current user request replaces slash-command-only placeholders such as $ARGUMENTS.
• Delegation and structured questions use the active harness's available capability, with a local fallback when the capability is absent.
• Shared workflows are read explicitly by path; they do not rely on Claude's @file expansion.

Claude-Only Settings

The following remain in .claude/settings.json because they are not portable instruction content:

• Claude model and effort selection
• permission allow/deny lists
• lifecycle hooks (shellsafety, tmux notifications, resume, and cleanup)
• worktree, status-line, TUI, voice, and experimental feature settings

Configure equivalents in each harness rather than translating them into AGENTS.md or skills.

Task Skills

• todo: research and add work to the front of PLAN.md
• later: research and add work to the back of PLAN.md
• list-tasks: summarize remaining plan items
• next-task: plan, implement, verify, commit, and archive the next item
• find-bugs: audit a codebase and add verified findings to the plan