Housekeeping
Agent hosts leave state on a machine: native memory, sessions, caches, plugins and storage. This skill leads the housekeeping family. It holds the rules every agent host shares, and each runtime skill depends on it and adds only its own host's surfaces and controls. The housekeeping standard holds the shared rules.
Runtime skills
ki-env-housekeeping-claudefor Claude Desktop, Cowork, Claude Code and VSCode state.ki-env-housekeeping-chatgptfor the ChatGPT local store and Codex memory and sessions.
Boundaries
ki-acquireowns capturing a session's knowledge; full-session import is an owner-selected exception there, never routine housekeeping.ki-env-tokenomicsowns context budgets;ki-env-bindingowns MCP configuration.ki-envleads the Environment guild this family belongs to.
Operating modes
For help, -h, ? or no recognised mode, explain the shared rules and route each agent host's state to its runtime skill, without reading or changing any state; offer a mode choice only in an interactive session.
Mode AUDIT
Run ki repo audit --skill ki-env-housekeeping-claude or ki repo audit --skill ki-env-housekeeping-chatgpt for the host in use; each composes this skill first. Then review the evidence against the shared rules.
Mode CONFORM
There is no shared conform action: each runtime skill proposes only the safe repairs its host allows, and deletion always needs separate, explicit authority. The host may regenerate this skill's rubric publication through ki repo conform --skill ki-env-housekeeping --repo <repo> --dry-run.
Mode EDUCATE
Explain the native-memory policy values, where durable learning goes, and why session sources stay read-only, then name the runtime skill for the host in use.
Mode REFRESH
REFRESH writes only in the canonical ki-agent-backpack repository; from an installed copy, stop and redirect there. Read the sources, compare the shared rules with both runtime skills' standards, move a rule here only when both hold it, then make and verify the smallest correction.