ki-env-housekeeping-claude
Standard: the hygiene of accumulated Claude state across the surfaces where it collects — the areas and the skill↔server pairing in the Claude-state standard; the memory area's file format in the auto-memory standard; line-by-line criteria in the rubric; tracked provenance in sources. The standards embed the two exact canonical shapes (directory/index and frontmatter), so this skill intentionally needs no separate exemplars file.
This skill composes ki-env-housekeeping, the family's lead, which holds the rules every agent host shares, and adds Claude's own surfaces and controls.
What this skill owns
The standard and judgment over the state Claude accumulates on a machine, across three surfaces — Claude Desktop / Cowork sessions, Claude Code (~/.claude/), and VSCode chat sessions — spanning the areas: sessions, artifacts / outputs, backups, plugins, project cache, and auto-memory. Full model in the Claude-state standard.
The mechanical arm is split by area:
- Native memory — KI declares
auto_memory = "transition" while a repository reconciles existing Claude notes, then "disabled" when review and routing finish; "enabled" requires a separate per-project Claude opt-in. Omission is a FAIL; only an explicit value resolves repository policy. This KI value is separate from Claude's boolean autoMemoryEnabled. The checker resolves user, project, and project-local settings plus the auto-memory environment override. Disabled KI policy skips index/file checks, warns whenever its selected memory directory exists, and fails selection if Claude memory is still on. Transition always warns until closed. Enabled memory is inspected only in its selected contained directory. Detail in the auto-memory standard.
- Headroom output and every other area — a
headroom:learn block is rendered-file evidence only; it does not prove a Headroom database, version, installation, or executed learn action. The paired mcp-housekeeping-claude server is a separate tool source: a source checkout or inventory declaration does not prove registration, access exposure, or an executed audit. The skill applies judgment only to independently obtained server audit evidence; it never re-implements those tools.
AI session acquisition
Readable Claude session capture belongs to ki-acquire's agent-session acquisition guidance. Full-session import is an owner-selected exception, not routine bulk housekeeping. The existing mcp-housekeeping-claude session-source surface stays read-only and shared; discovery or checkpoint evidence grants no capture, harvest or retirement authority. Keep this skill for content-minimised inventory, local memory and separately authorised storage cleanup.
Operating modes
Carries the universal AUDIT · CONFORM · EDUCATE · REFRESH. Invoked as help / -h / ?, it explains itself and stops — the generated HELP block (name, purpose, invocation, modes, off-ramps), taking no action. With no mode it does the same, then, in an interactive session only, offers the mode choice via AskUserQuestion, prompting for any argument-hint target the chosen mode shows.
| Mode |
What it does |
| AUDIT |
Composes ki-env-housekeeping first. Run ki repo audit --skill ki-env-housekeeping-claude for selected native-memory evidence and its bounded store. Report non-memory server state unavailable unless registration, access exposure, and an executed server audit are separately evidenced; then apply the judgment criteria in rubric.md. Procedure in mode-audit.md. |
| CONFORM |
Run ki repo conform --skill ki-env-housekeeping-claude; it proposes only safe transactional repairs in an already selected bounded memory directory. Any non-memory cleanup requires independently established server access and authority. Re-AUDIT until the local evidence is clean. Procedure in mode-conform.md. |
| EDUCATE |
Run ki repo educate --skill ki-env-housekeeping-claude to declare the repository's bounded user-home evidence; add it with ki skill add ki-env-housekeeping-claude when it should also be installed for every supported Claude Code agent. User activation installs a skill only — it does not audit or conform it. |
| REFRESH |
Re-check the standard against its sources: Headroom's memory-feature behavior for the memory format, and the mcp-housekeeping-claude server's tool surface for the other areas, per mode-refresh.md. |
Notes
- The state this skill governs lives outside the repo tree — under
~/.claude/, ~/Library/Application Support/Claude/, and VSCode's workspaceStorage/. Its structured session is restricted to the selected repository's Claude project memory; the session / artifact / storage areas are machine-level and audited directly through the server, not by repository audit.
- Disabled KI policy and an enabled selected directory with no
memory/ directory are NA for index/file criteria. An existing selected memory directory under omitted or disabled KI policy, or any open transition, is WARN at SELECT-2; omission is also FAIL at SELECT-1. Uncertain selection, mismatched KI/Claude state, and enabled memory without a project-scoped opt-in or transition are FAIL at SELECT-1.
- Route Markdown formatting to the separately applicable
ki-authoring standard. The local ki-skills:rubric shared module is compile-time packaging, not a governance dependency.
- Does not assume any particular personal
~/.claude/CLAUDE.md content — the doctrines checked are the systems' own universal instructions, not one user's private elaboration of them.
1---2name: ki-env-housekeeping-claude3ki-kind: governance4ki-applicability: declaration-only5ki-guild: environment6ki-depends-on: [ki-env-housekeeping]7ki-runtime-binding: true8ki-supported-runtimes: [claude-code]9ki-shared-dependencies: [ki-skills:rubric]10description: >11 Audit Claude Desktop, Cowork, Claude Code and VSCode state: memory, storage hygiene, plugins, caches12 and backups. Use for local Claude housekeeping; `ki-acquire` owns exception-based agent-session13 capture, `ki-repo-kb` KB structure and `ki-env-tokenomics` context budgets.14argument-hint: 'audit | conform | help | educate | refresh'15metadata:16 source: https://skillmd.com/skills/knowledgeislands/ki-env-housekeeping-claude17---1819# ki-env-housekeeping-claude2021**Standard:** the hygiene of accumulated Claude state across the surfaces where it collects — the areas and the skill↔server pairing in [the Claude-state standard](references/standards-claude-state.md); the memory area's file format in [the auto-memory standard](references/standards-auto-memory.md); line-by-line criteria in [the rubric](references/rubric.md); tracked provenance in [sources](references/sources.md). The standards embed the two exact canonical shapes (directory/index and frontmatter), so this skill intentionally needs no separate exemplars file.2223This skill composes `ki-env-housekeeping`, the family's lead, which holds the rules every agent host shares, and adds Claude's own surfaces and controls.2425## What this skill owns2627The **standard and judgment** over the state Claude accumulates on a machine, across three surfaces — Claude Desktop / Cowork sessions, Claude Code (`~/.claude/`), and VSCode chat sessions — spanning the areas: **sessions**, **artifacts / outputs**, **backups**, **plugins**, **project cache**, and **auto-memory**. Full model in [the Claude-state standard](references/standards-claude-state.md).2829The **mechanical arm** is split by area:30311. **Native memory** — KI declares `auto_memory = "transition"` while a repository reconciles existing Claude notes, then `"disabled"` when review and routing finish; `"enabled"` requires a separate per-project Claude opt-in. Omission is a FAIL; only an explicit value resolves repository policy. This KI value is separate from Claude's boolean `autoMemoryEnabled`. The checker resolves user, project, and project-local settings plus the auto-memory environment override. Disabled KI policy skips index/file checks, warns whenever its selected memory directory exists, and fails selection if Claude memory is still on. Transition always warns until closed. Enabled memory is inspected only in its selected contained directory. Detail in [the auto-memory standard](references/standards-auto-memory.md).322. **Headroom output and every other area** — a `headroom:learn` block is rendered-file evidence only; it does not prove a Headroom database, version, installation, or executed learn action. The paired **`mcp-housekeeping-claude`** server is a separate tool source: a source checkout or inventory declaration does not prove registration, access exposure, or an executed audit. The skill applies judgment only to independently obtained server audit evidence; it never re-implements those tools.3334## AI session acquisition3536Readable Claude session capture belongs to `ki-acquire`'s agent-session acquisition guidance. Full-session import is an owner-selected exception, not routine bulk housekeeping. The existing `mcp-housekeeping-claude` session-source surface stays read-only and shared; discovery or checkpoint evidence grants no capture, harvest or retirement authority. Keep this skill for content-minimised inventory, local memory and separately authorised storage cleanup.3738## Operating modes3940Carries the universal **AUDIT · CONFORM · EDUCATE · REFRESH**. Invoked as `help` / `-h` / `?`, it explains itself and stops — the generated HELP block (name, purpose, invocation, modes, off-ramps), taking no action. With no mode it does the same, then, in an interactive session only, offers the mode choice via `AskUserQuestion`, prompting for any `argument-hint` target the chosen mode shows.4142| Mode | What it does |43| --- | --- |44| AUDIT | Composes `ki-env-housekeeping` first. Run `ki repo audit --skill ki-env-housekeeping-claude` for selected native-memory evidence and its bounded store. Report non-memory server state unavailable unless registration, access exposure, and an executed server audit are separately evidenced; then apply the judgment criteria in [rubric.md](references/rubric.md). Procedure in [mode-audit.md](references/mode-audit.md). |45| CONFORM | Run `ki repo conform --skill ki-env-housekeeping-claude`; it proposes only safe transactional repairs in an already selected bounded memory directory. Any non-memory cleanup requires independently established server access and authority. Re-AUDIT until the local evidence is clean. Procedure in [mode-conform.md](references/mode-conform.md). |46| EDUCATE | Run `ki repo educate --skill ki-env-housekeeping-claude` to declare the repository's bounded user-home evidence; add it with `ki skill add ki-env-housekeeping-claude` when it should also be installed for every supported Claude Code agent. User activation installs a skill only — it does not audit or conform it. |47| REFRESH | Re-check the standard against its sources: Headroom's memory-feature behavior for the memory format, and the `mcp-housekeeping-claude` server's tool surface for the other areas, per [mode-refresh.md](references/mode-refresh.md). |4849## Notes5051- The state this skill governs lives **outside the repo tree** — under `~/.claude/`, `~/Library/Application Support/Claude/`, and VSCode's `workspaceStorage/`. Its structured session is restricted to the selected repository's Claude project memory; the session / artifact / storage areas are machine-level and audited directly through the server, not by repository audit.52- Disabled KI policy and an enabled selected directory with no `memory/` directory are **NA** for index/file criteria. An existing selected memory directory under omitted or disabled KI policy, or any open transition, is **WARN** at SELECT-2; omission is also **FAIL** at SELECT-1. Uncertain selection, mismatched KI/Claude state, and enabled memory without a project-scoped opt-in or transition are **FAIL** at SELECT-1.53- Route Markdown formatting to the separately applicable `ki-authoring` standard. The local `ki-skills:rubric` shared module is compile-time packaging, not a governance dependency.54- Does not assume any particular personal `~/.claude/CLAUDE.md` content — the doctrines checked are the systems' own universal instructions, not one user's private elaboration of them.