# Context > Audit auto-loaded context budget, detect redundancy, and recommend optimizations. - Skill: `coreindustries/context` (Agent Skill) - Install (CLI): `npx skillmds@latest add coreindustries/context` - Raw SKILL.md: https://api.skillmd.com/api/skills/coreindustries/context/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Security - Author: coreindustries (https://skillmd.com/u/coreindustries) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/coreindustries/context --- # /context Audit auto-loaded context budget, detect redundancy, and recommend optimizations. ## Usage ``` /context [action] [--verbose] ``` ## Arguments - `action`: `audit`, `redundancy`, `recommendations`, `compare` (default: `audit`) - `--verbose`: Show per-line token estimates for each file ## Instructions When this skill is invoked: ### Agent Behavior **Autonomy:** - Run the full audit pipeline without prompting - Produce a formatted report, not just raw data - Prioritize actionable findings over informational noise **Token Estimation:** - Use the approximation: **1 token ~ 4 characters** (English prose/code average) - Read each file, count characters, divide by 4, round to nearest integer - This is an estimate — actual tokenizer counts vary by ~10% ### Actions #### Audit (`/context audit`) Measure all auto-loaded files and report token usage: 1. **Read `CLAUDE.md`** — count characters, estimate tokens 2. **Glob `.claude/rules/*.md`** — read each file, estimate tokens 3. **Check for enabled platform rules** — identify symlinks in `rules/` that point to `rules-available/` 4. **Sum totals** and calculate percentage of 200K context window **Report format:** ``` Context Budget Report ===================== Auto-loaded files: CLAUDE.md 1,265 tokens rules/ai-agent-patterns.md 1,212 tokens rules/code-quality.md 842 tokens rules/error-handling.md 534 tokens rules/git-workflow.md 687 tokens rules/quality-checks.md 612 tokens rules/security-core.md 1,705 tokens rules/task-management.md 786 tokens ───────────────────────────────────────────── Total: 7,643 tokens (3.8% of 200K) Platform rules (enabled): (none active) Available platform rules (not enabled): rules-available/android.md ~XXX tokens rules-available/docker.md ~XXX tokens rules-available/ios.md ~XXX tokens rules-available/nextjs.md ~XXX tokens rules-available/python.md ~XXX tokens rules-available/security-mobile.md ~XXX tokens rules-available/security-owasp.md ~XXX tokens rules-available/security-web.md ~XXX tokens ``` #### Redundancy Scan (`/context redundancy`) Cross-reference auto-loaded files for duplicated content: 1. **Extract all `{placeholder}` references** from each auto-loaded file 2. **Find placeholders that appear in multiple files** — these are potential redundancies 3. **Scan for repeated guidance** — same concept explained in both CLAUDE.md and a rule file 4. **Check CLAUDE.md against rules** — content in CLAUDE.md that duplicates what a rule already covers **Report format:** ``` Redundancy Findings =================== Shared placeholders (appear in 2+ auto-loaded files): {lint_fix} → CLAUDE.md:48, quality-checks.md:45 {type_check} → CLAUDE.md:49, quality-checks.md:39 {test_with_coverage} → CLAUDE.md:50, quality-checks.md:47 Duplicate guidance: - CLAUDE.md "Context Recovery" section overlaps with /resume skill (low priority — CLAUDE.md version is a brief pointer) Summary: 3 shared placeholders, 0 high-priority duplicates ``` #### Recommendations (`/context recommendations`) Produce prioritized optimization suggestions: 1. **Run audit and redundancy scan** (internally) 2. **Categorize findings** by priority: - **P1** (high impact): Files > 1,500 tokens that could be split or moved to references - **P2** (medium impact): Redundant content that could be consolidated - **P3** (low impact): Minor optimizations, formatting improvements **Report format:** ``` Context Optimization Recommendations ===================================== P1 - High Impact: - security-core.md (1,705 tokens): Consider moving detailed tables and checklists to .claude/references/security-checklist.md. Auto-loaded portion could be a ~400 token summary with "see references/security-checklist.md for details". Estimated savings: ~1,200 tokens P2 - Medium Impact: - {lint_fix} placeholder appears in 2 auto-loaded files. CLAUDE.md Commands section duplicates quality-checks.md. Consider removing command block from one location. Estimated savings: ~150 tokens P3 - Low Impact: - quality-checks.md checklist section could move to references. Estimated savings: ~80 tokens Total potential savings: ~1,430 tokens (18.7% of current budget) ``` #### Compare (`/context compare`) Show the budget impact of enabling/disabling platform rules: 1. **Read all files in `rules-available/`** — estimate tokens for each 2. **Check which are currently enabled** (symlinked into `rules/`) 3. **Show current vs. projected budget** for each combination **Report format:** ``` Platform Rule Budget Comparison =============================== Current baseline: 7,643 tokens (3.8% of 200K) If enabled: + python.md +892 tokens → 8,535 (4.3%) + security-owasp.md +1,105 tokens → 8,748 (4.4%) + nextjs.md +1,340 tokens → 8,983 (4.5%) + docker.md +654 tokens → 8,297 (4.1%) + ios.md +780 tokens → 8,423 (4.2%) + android.md +720 tokens → 8,363 (4.2%) + security-web.md +940 tokens → 8,583 (4.3%) + security-mobile.md +830 tokens → 8,473 (4.2%) Common combinations: make enable-web (nextjs + security-web + security-owasp) +3,385 tokens → 11,028 (5.5%) make enable-python (python + security-owasp) +1,997 tokens → 9,640 (4.8%) make enable-mobile (security-mobile + security-web + security-owasp) +2,875 tokens → 10,518 (5.3%) ``` ### Implementation Notes - **File reading**: Use Read tool for each file, count `content.length` characters - **Token estimation**: `Math.round(charCount / 4)` — acceptable approximation - **Symlink detection**: Use `ls -la .claude/rules/` via Bash to identify symlinks pointing to `rules-available/` - **Output**: Always produce the formatted report directly — do not write to a file unless asked - **No modifications**: This is a read-only diagnostic skill. Never modify files during audit ### Platform Rule Combinations Reference These mappings come from the Makefile `enable-*` targets: | Target | Rules Enabled | |--------|--------------| | `make enable-web` | nextjs, security-web, security-owasp | | `make enable-python` | python, security-owasp | | `make enable-api` | security-owasp | | `make enable-ios` | ios, security-owasp | | `make enable-android` | android, security-owasp | | `make enable-mobile` | security-mobile, security-web, security-owasp | | `make enable-docker` | docker, security-owasp | ## Example Output ``` $ /context Context Budget Report ===================== Auto-loaded files: CLAUDE.md 1,265 tokens rules/ai-agent-patterns.md 1,212 tokens rules/code-quality.md 842 tokens rules/error-handling.md 534 tokens rules/git-workflow.md 687 tokens rules/quality-checks.md 612 tokens rules/security-core.md 1,705 tokens rules/task-management.md 786 tokens ───────────────────────────────────────────── Total: 7,643 tokens (3.8% of 200K) Platform rules (enabled): (none active) Redundancy findings: 3 shared placeholders across auto-loaded files 0 high-priority content duplicates Top recommendation: P1: Move security-core.md details to references (~1,200 token savings) Run '/context recommendations' for full optimization plan. Run '/context compare' to see platform rule budget impact. ```