context-engineeringlisted

# Context Engineering I'll audit your current context load, identify what is inflating it unnecessarily, and restructure the 5-layer hierarchy so the agent stays sharp for the rest of the session. Arguments: `$ARGUMENTS` — optional focus (e.g., "optimize for implementation", "optimize for debugging") ## Token Optimization **Expected range**: 200–600 tokens (audit report + recommendations) **Early exit**: If context appears healthy (no redundant files loaded, no stale error traces), report "Context healthy — no changes needed" **Patterns used**: Bash for file size checks, progressive disclosure (summary first, details on request) ## The 5-Layer Context Hierarchy ``` Layer 1 — Rules files CLAUDE.md, .claude/rules/*.md Layer 2 — Specs / designs docs/specs/, PRD, ADRs Layer 3 — Source files only the files currently being changed Layer 4 — Error output compiler errors, test failures, logs Layer 5 — Conversation current session history ``` Each layer has an appropriate size budget. Layers 3–5 are the most common source of bloat. ## Step 1 — Audit Current Context ```bash # Estimate context pressure indicators wc -l .claude/rules/*.md 2>/dev/null | tail -1 find . -name "CLAUDE.md" | xargs wc -l 2>/dev/null | tail -1 ls -la docs/specs/ 2>/dev/null | head -10 ``` Check for common inflation sources: - Entire files loaded when only a function is needed - Stale error traces from previous attempts (no longer relevant) - Spec documents for features not cur