Wiring agents

Run anb setup in the project root after installing anb on your PATH. Setup installs instructions and skills so agents can use the notebook, plus session-start hooks for Claude Code and Codex. Run it again after an upgrade to refresh the managed files.

anb setup

The reply lists each file and whether it was written, already present, or left alone. Setup does not create the notebook itself; the first anb add does that.

What setup writes

File	Contents
AGENTS.md	A marked instruction block pointing to anb status and anb --help
CLAUDE.md	The same block, unless the file already links to or imports AGENTS.md
.claude/settings.json	A SessionStart hook running anb status --hook with a 15-second timeout
.codex/hooks.json	The corresponding Codex hook
.claude/skills/anb/, .agents/skills/anb/	The workflow skill, command reference, worked session and refusal reference
.claude/skills/anb-atlas/, .agents/skills/anb-atlas/	The skill for drawing and reviewing the notebook

Setup adds its hook beside existing hook groups. It validates the files before writing: invalid JSON or an unmatched instruction marker causes a refusal before those changes are applied. Symlinks are reported and left alone, including an existing CLAUDE.md link to AGENTS.md.

Enable the session hook

For Codex, trust the project in its configuration and review the installed hook with /hooks. Setup prints a reminder. Claude Code uses the hook in .claude/settings.json.

The hook gives the model Status as additional context. If the notebook root cannot be read, it returns an empty payload and exits successfully. Hook failure therefore does not prevent the session from starting. To inspect the notebook directly, run anb status.

For an agent without a compatible hook, make anb status the opening command in its instructions. An agent that reads AGENTS.md can find that command there. Point it to the installed anb skill for the workflow; the CLI itself needs only shell access.

The supplied workflow

The anb skill teaches agents to resume the active Task or select ready work, log progress, record Decisions and Questions, and close with proof before archiving. It keeps one Task in flight, groups larger work into epics, and requires a reason for a hold. By default, the agent checks the notebook before stopping and commits it with the code it describes. Maintaining the notebook is part of the agent’s work. Detailed command and refusal references are separate files, so the agent can load them when needed.

The binary generates the workflow skill and references. Command definitions also supply --help, and worked examples run against a scratch notebook during generation. CI compares the committed skill with that output. This checks command and example drift; the authored workflow guidance still needs review when behavior changes.

anb skill                       # print the skill
anb skill path/to/skill          # write it to a directory
anb skill path/to/skill --check  # fail if the directory differs

The anb-atlas skill is authored separately and bundled with the binary. It uses graph data to build an interactive page and routes review comments back through the CLI. See Drawing the notebook.

Customize or remove the integration

Make the workflow yours gives complete recipes for a private notebook in .tmp/xxx, another storage location, custom skills and a review stage. Remove managed-by: anb from each skill file you maintain yourself; setup then reports it as yours, left alone during updates and removal.

anb setup --remove removes managed instructions, hooks and skill files. Other tools’ instruction text and hook groups remain, and a skill directory is removed only if it is empty. The notebook records are not removed.