| name | zoo-init |
| description | Initialize repo-local .zoo customization files for Zoo from real repository research. Use when setting up Zoo customization for a repository. |
Follow .zoo/zoo.md if exists.
Create concise repo-local Zoo guidance. No placeholders.
Codex only: before continuing, ensure global Codex config ($CODEX_HOME/config.toml, usually ~/.codex/config.toml) contains default_mode_request_user_input = true under [features]. Required for request_user_input tool to work outside of planning mode. If added during running session and the tool stays unavailable, start a new turn/thread or ask user restart Codex. Bail without this tool available.
Research first: root agent instructions, docs dirs, layout, build/test/browser harness, commit/review/docs/proposal/release conventions.
After research, ask user with request_user_input / AskUserQuestion, give option for their existing value if different from recommended value:
- do they want to use terse messages style, recommend yes.
- spec (aka task) file path, recommend
.spec/YYYYMMDD-taskslug.md
- should spec files be committed or gitignored, recommend commit
- evidence dir (screenshots and example files), recommend
.tasks/YYYYMMDD-taskslug-evidence/
- should evidence be git tracked or gitignored, recommend ignore
- research path, recommend
.tasks/YYYYMMDD-taskslug-research.md
- should research be git tracked or gitignored, recommend ignore
- proposals path, recommend
.proposals/YYYYMMDD-title.md
- archived proposals path, recommend
archived/ under chosen proposals dir
- mistake log path, recommend
.ai-mistakes.md, give option to have none
NB: when asking for paths, give full example (.spec/YYYYMMDD-taskslug.md), not prefix (.spec/YYYYMMDD).
If user agreed to terse style, load terse skill immediately, and add at the start of CLAUDE.md and AGENTS.md (whichever exist): Load terse skill for voice in all messages incl between tool calls.
Create .zoo/, create new files under it; merge new research if files already exist;
.zoo/zoo.md: zoo dirs (specs, proposals, tasks), how to run tests and scripted checks, mistake log path, ticket tools
.zoo/planning.md: pointers to planning docs if any (e.g. product, architecture info)
.zoo/planreview.md: leave empty
.zoo/testing.md: pointers to test writing docs if any
.zoo/coding.md: pointers to coding/style docs if any
.zoo/codereview.md: leave empty
.zoo/browser.md: pointers to harness info, how to run app for browser testing, tunnel/callback rules
.zoo/docs.md: pointers to doc destinations, instructions, validation, terminology, audience
.zoo/subtask-start.md: leave empty
.zoo/task-finish.md: leave empty
.zoo/rebase.md: leave empty
.zoo/push.md: repo publish instructions (use git push, open PR, etc) if any
.zoo/proposals.md: proposal folder/naming/archive/statuses/rules
.../MISTAKES.md: initialize unless user refused
Update .gitignore to ignore/unignore according to user choices.
If any unknown files found under .zoo, list them and ask user if they want to delete.
Check if commit skill exists; if not, ask if they want to create it (recommended); if yes, run zoo-hr skill in subagent.