| name | maintain |
| description | Run note quality maintenance — invoke /inspect to diagnose vault metadata, then /repair to fix queued issues. Use when the user wants weekly or periodic Rill vault health maintenance. |
/maintain — Quality Maintenance
Conduct all conversation with the user — and write all generated output — in the language defined by .claude/rules/personal-language.md (or the user's input language if that file is absent). Follow the language rules in full — exceptions and translation quality are defined in the Language Rules of .claude/rules/rill-core.md and the vault's personal-*.md overrides, never restated per skill. The English instructions below are for skill clarity, not for output style.
Tool references in this skill (sub-agent, "skill invocation") describe intent, not Claude-specific tool calls. Each harness should map them to its native equivalent — Claude Code uses its built-in Skill / Agent tools as named; Codex CLI uses its own skill invocation mechanism (mention / implicit / /skills) and sub-agent mechanism as appropriate.
Runs note quality inspection and repair sequentially. Diagnoses issues with /inspect, then fixes them with /repair if any are found.
Arguments
$ARGUMENTS — None
Steps
Step 1: Quality Inspection
Invoke the inspect skill (via the harness's skill mechanism).
/inspect diagnoses taxonomy health + metadata accuracy + consistency, and adds problematic files to knowledge/.refresh-queue.
Step 2: Repair Decision
Read knowledge/.refresh-queue and check the count.
- If count is 0: Display "No repairs needed. Quality is good." and finish
- If count is greater than 0: Proceed to Step 3
Step 3: Metadata Repair
Invoke the repair skill (via the harness's skill mechanism).
/repair reads files from .refresh-queue and batch-normalizes tags/mentions/type.
Step 4: Completion Summary
- /inspect diagnostic results (files inspected, issues found)
- /repair results (files repaired, files skipped)
Why /inspect and /repair Are Invoked Inline
/inspect and /repair use sub-agents internally, but are not as heavy as /distill.
The combined context consumption of the 2 skills is manageable (~50-80K tokens).
To avoid the process startup overhead of spawning a new harness session, they are run sequentially inline within the current session.
Rules
- If /inspect fails, do not run /repair (cannot repair without diagnostic results)
- After everything completes, display a summary and finish (do not transition to assistant mode)
