| name | rule-manager |
| description | Audit and iteratively optimize .claude/rules/ files for token efficiency. Each invocation: audit all rules, pick the single highest-priority optimization, apply it, and log the change. Run when the SessionStart hook reports issues, when context seems bloated, or periodically for maintenance. Uses bash scripts for fast data collection.
|
| user-invocable | true |
Rule Manager
Iteratively optimizes .claude/rules/ for token efficiency. Each run: audit -> pick ONE optimization -> apply -> log.
Step 1: Read Framework Reference
Read the memory framework doc for context on priorities and standards:
${CLAUDE_PLUGIN_ROOT}/memory-framework.md
Also fetch/refresh the official Claude Code memory docs (cached locally):
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-fetch-docs.sh"
Step 2: Audit
Run the audit script to collect current state:
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-audit.sh" --json
Parse the JSON output. Report to user:
- Total files, lines, estimated tokens
- Files with/without paths: frontmatter
- Files with @ references
- Health indicators (oversized, unscoped, refs)
Also read the optimization log to see what's been done (per-day files):
.claude/memory/ (read the latest file, or all files for full history)
Step 3: Identify ONE Optimization
Pick the HIGHEST priority that applies (do NOT do multiple):
Priority 1: @ References Between Rules
If any rule file has @ references to other rule files:
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-strip-refs.sh" <filename> --dry-run
Show the preview to user. If approved:
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-strip-refs.sh" <filename>
Priority 2: Unscoped Domain Rule
If a rule lacks paths: frontmatter but is domain-specific:
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-suggest-paths.sh" <filename>
Review the suggestions. For complex cases, use an Explore agent to grep the codebase for the rule's key terms. Show proposal to user. If approved:
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-apply-paths.sh" <filename> "pattern1" "pattern2"
Priority 3: Oversized Rule (>400 LOC)
If a rule exceeds 400 lines:
- Read the file and identify distinct sections
- Propose splitting into 2-3 focused files with paths: frontmatter
- Create new files, delete original
- This is a manual operation (no script -- requires judgment)
Priority 4: Stale/Orphaned Rule
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-find-orphans.sh"
If orphans found, propose archiving or deletion.
Priority 5: All Clean
Report "Rules are healthy. No optimization needed." and STOP — do NOT proceed to Step 4 (nothing to log).
Step 4: Log (only when a change was applied)
Skip this step if Priority 5 was reached (no change made).
After applying a change:
"${CLAUDE_PLUGIN_ROOT}/skills/rule-manager/scripts/rule-log.sh" "<action>" "<details>"
Report: "Done. Run /rule-manager again for next optimization."
