Agent onboarding
The copy-paste discovery snippet that lets any coding agent find and use Kibi's interfaces.
This repository uses Kibi.
Initialize repository infrastructure with kibi init. Then ask the agent to
“Bootstrap Kibi for this repository.” The agent first inspects
kb_status.bootstrap and follows its typed nextAction, then routes the
request to kibi-bootstrap. That skill first interviews the human about
knowledge sources outside the code, then calls kb_plan_bootstrap with the
declared sources and cited intent claims; further questions come only from a
needs_context result. It shows the hash-bound plan, obtains
approval, passes the unchanged plan to kb_apply_plan, and finishes with
kb_check and kb_status.
On a repository whose knowledge base was written by an older Kibi (a
.kb/config.json and no .kb/manifest.json), the typed next action is
kibi migrate --yes, the same step kibi doctor names: do not run
kibi init first, because it would claim the latest schema and skip the
legacy migration steps.
The stable onboarding contract is typed status plus the named workflow skill.
Skill manifests and resources are host plumbing for advanced integrations;
they are not a prerequisite for ordinary bootstrap requests.
Advanced hosts may call kb_skills with action: "list" and action: "load"
(CLI: kibi skill) to discover that plumbing; ordinary onboarding follows
typed status and the named skill.
For general Kibi work, load the bundled kibi-usage skill as the shared
source-first guidance.
Follow the loaded skill guidance and current operation schemas.
Do not treat a package skills/ directory as implicitly loaded by the host.
Do not bypass Kibi's normal authorization or mutation safeguards.
Skills are bundled only. Remote install, marketplace install, and script execution are not supported.