How do I install the R-cleanup-context skill?

Run npx skillmds@latest add roxabi/r-cleanup-context in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is R-cleanup-context safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with R-cleanup-context?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is R-cleanup-context free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published R-cleanup-context?

Roxabi (@roxabi) published this skill. Their other Agent Skills are listed on their SkillMD profile.

R-cleanup-context by roxabi — Agent Skill

Name: R-cleanup-context
Author: Roxabi

R-cleanup-context

Roxabi 43c55e6 5 files · 14.5 KB Updated Sep 17, 2026

File contents

Context Cleanup

Goal: every ε resolved. Context files = inbox — stale rules, contradictions, memory entries → permanent home ∨ deleted.

Addresses "rule accumulation decay": rules↑ → performance↑ → too many → contradictions → performance↓ silently.

Let: ε := finding (contradiction, stale ref, redundancy, bloat, memory entry) μ := MEMORY.md (first κ lines injected every session) τ := memory/.md (topic files, loaded on demand) α := .claude/agent-memory//MEMORY.md (per-agent) κ := 200 (MEMORY.md line cap) λ := .claude/context-audit-log.md (append-only audit log) Π := placement targets (auto-detected per project)

/R-cleanup-context                    Audit all context areas
/R-cleanup-context --scope claude-md  Only audit CLAUDE.md files
/R-cleanup-context --scope skills     Only audit skill files
/R-cleanup-context --scope memory     Only audit auto-memory (μ + τ + α)
/R-cleanup-context --dry-run          Show findings without proposing fixes

Resolutions

∀ ε → exactly one resolution:

Resolution	When	Action
Fix	Root cause = bug/wrong config/design flaw	Fix code/config/workflow, delete ε
Promote	Durable insight needed by multiple agents	Move to permanent target (see Π), delete ε
Relocate	Domain/agent-specific knowledge in wrong scope	Move to narrower target, delete ε
Delete	Ephemeral, stale, already covered, resolved	Delete ε

Dispatch

Phase 1 — Discovery → Read ${CLAUDE_SKILL_DIR}/cookbooks/discovery.md, execute. Phase 2 — Analysis → Read ${CLAUDE_SKILL_DIR}/cookbooks/analysis.md, execute. Phase 3–5 — Resolution → Read ${CLAUDE_SKILL_DIR}/cookbooks/resolution.md, execute.

Safety

Never delete entire files — only remove/edit specific lines
Never modify files outside the project without explicit approval
Always show the exact diff before applying any change
Memory files (~/.claude/projects/*/memory/) — extra caution, ask before every edit
CLAUDE.md changes — show full before/after of modified sections
Installed skill files — read-only (report issues, don't modify cache copies)
Back up large changes: copy original to {file}.bak before multi-line edits

Edge Cases

Scenario	Behavior
No context files found	"No context files discovered. Nothing to audit."
All checks pass	"Context is healthy. No issues found."
CLAUDE.md imports (@file)	Follow imports, audit imported files too
Monorepo with nested CLAUDE.md	Audit each independently, check cross-file consistency
User declines all fixes	"No changes applied. Re-run when ready."
Agent memory but no agent defs	Report as bloat — orphaned memory with no consumer

When to Run

/R-cleanup-context or /R-dev --cleanup-context anytime
After shipping a feature (ephemeral context likely stale)
After cleaning branches (worktree refs likely stale)
When memory files > 5 entries or μ > 50 lines
Periodically after every N completed issues

$ARGUMENTS

---name: r-cleanup-contextdescription: Audit and clean CLAUDE.md, memory, skills, and rules — resolve every finding (fix/promote/relocate/delete), track recurrences, target bloat=0. Triggers: "cleanup context" | "context audit" | "clean memory" | "drain memory" | "prune memory" | "audit memory" | "consolidate rules" | "spa day" | "memory audit".---# Context Cleanup**Goal: every ε resolved.** Context files = inbox — stale rules, contradictions, memory entries → permanent home ∨ deleted.Addresses "rule accumulation decay": rules↑ → performance↑ → too many → contradictions → performance↓ silently.Let: ε := finding (contradiction, stale ref, redundancy, bloat, memory entry) μ := MEMORY.md (first κ lines injected every session) τ := memory/*.md (topic files, loaded on demand) α := .claude/agent-memory/*/MEMORY.md (per-agent) κ := 200 (MEMORY.md line cap) λ := .claude/context-audit-log.md (append-only audit log) Π := placement targets (auto-detected per project)```/R-cleanup-context Audit all context areas/R-cleanup-context --scope claude-md Only audit CLAUDE.md files/R-cleanup-context --scope skills Only audit skill files/R-cleanup-context --scope memory Only audit auto-memory (μ + τ + α)/R-cleanup-context --dry-run Show findings without proposing fixes```## Resolutions∀ ε → exactly one resolution:| Resolution | When | Action ||-----------|------|--------|| **Fix** | Root cause = bug/wrong config/design flaw | Fix code/config/workflow, delete ε || **Promote** | Durable insight needed by multiple agents | Move to permanent target (see Π), delete ε || **Relocate** | Domain/agent-specific knowledge in wrong scope | Move to narrower target, delete ε || **Delete** | Ephemeral, stale, already covered, resolved | Delete ε |## DispatchPhase 1 — Discovery → Read `${CLAUDE_SKILL_DIR}/cookbooks/discovery.md`, execute.Phase 2 — Analysis → Read `${CLAUDE_SKILL_DIR}/cookbooks/analysis.md`, execute.Phase 3–5 — Resolution → Read `${CLAUDE_SKILL_DIR}/cookbooks/resolution.md`, execute.## Safety1. Never delete entire files — only remove/edit specific lines2. Never modify files outside the project without explicit approval3. Always show the exact diff before applying any change4. Memory files (`~/.claude/projects/*/memory/`) — extra caution, ask before every edit5. CLAUDE.md changes — show full before/after of modified sections6. Installed skill files — read-only (report issues, don't modify cache copies)7. Back up large changes: copy original to `{file}.bak` before multi-line edits## Edge Cases| Scenario | Behavior ||----------|----------|| No context files found | "No context files discovered. Nothing to audit." || All checks pass | "Context is healthy. No issues found." || CLAUDE.md imports (@file) | Follow imports, audit imported files too || Monorepo with nested CLAUDE.md | Audit each independently, check cross-file consistency || User declines all fixes | "No changes applied. Re-run when ready." || Agent memory but no agent defs | Report as bloat — orphaned memory with no consumer |## When to Run- `/R-cleanup-context` or `/R-dev --cleanup-context` anytime- After shipping a feature (ephemeral context likely stale)- After cleaning branches (worktree refs likely stale)- When memory files > 5 entries or μ > 50 lines- Periodically after every N completed issues$ARGUMENTS

Frequently asked questions