CCDV-F Ep 8: Context Editing vs Compaction, Subagents and the Memory Tool

CCDV-F Exam Prep, Ep 08: context editing vs compaction, subagents and the memory tool. Domain 6, Prompt and Context Engineering (11.0% of the exam), and Domain 1, Agents and Workflows (14.7% of the exam). On turn 30 of a triage run, Claude rereads every old tool result on every request, so answers get worse and the bill climbs. Learn the rule: prune at the source, clear with context editing, summarize with compaction, isolate reading in a subagent, and keep lasting facts in the memory tool. ⏱️ Chapters 0:00 Turn 30: what leaves the window? 0:23 A bigger context window is not the fix 1:00 Symptoms: context bloat vs context drift 2:04 Prune tool output at the source 2:33 Context editing clears, compaction summarizes 3:09 context_management and clear_tool_uses 3:36 Threshold compaction: append the whole response.content 3:51 Cache rebuilds, trigger and clear_at_least 4:24 instructions replace the default summary prompt 4:39 Beta platform availability, including Amazon Bedrock 5:08 trigger and keep are example values 5:24 Isolate big reading jobs in a subagent 6:08 Outlive sessions with the memory tool 6:52 Decision rules cheat sheet 7:28 Next: Ep 9, MCP and Skills 📚 Course playlist:    • Claude Certified Developer (CCDV-F) Exam P...   🧪 Hands-on workshop: https://devops.gheware.com ⬅️ Previous: Ep 07 — One Tool Call End to End — strict, Parallel Results and is_error    • CCDV-F Ep 7: One Tool Call End to End — st...   ➡️ Next: Ep 09 — Build an MCP Server; Built-in Tool vs Custom Tool vs Skill vs MCP    • CCDV-F Ep 9: Build an MCP Server; Built-in...   Facts checked against the Claude docs on 2026-10-03. Context editing and compaction are still beta, so their beta headers, edit types, example trigger values and which platforms carry on-demand compaction can change. If a detail differs on exam day, reason from the principle. Independent exam-prep material. Not affiliated with or endorsed by Anthropic. Contains no live exam items. #CCDVF #ClaudeAI #ClaudeAPI #AnthropicCertification #AIAgents