Do you actually know where your AI coding time goes? Which tools fail the most? Which tasks eat the most tokens? Which prompts you type every single day? Your agent already logged all of it. This prompt reads those logs locally and builds you a full dashboard. Then it writes the CLAUDE. md changes your data says you need. Paste it into your agent tonight --------------------------------------------- Build me a private dashboard of my AI coding sessions from the last 7 days. WHERE TO LOOK (read only, on this machine): Claude Code: ~/.claude/projects/**/*.jsonl Codex: ~/.codex/sessions/**/*.jsonl and ~/.codex/history.jsonl PRIVACY RULES: Never print API keys, secrets or full prompts. Any snippet max 10 words, with keys, paths, URLs and long ids removed. If something isn't in the logs, write "not logged". Never invent numbers. PART 1: CRUNCH THE NUMBERS Write a Python script using only the standard library. For each tool, calculate: sessions per day, prompts, tool calls by name, failed calls by tool and error type, models used, tokens per model per day (input, output, cache), typical and worst case tool latency, session length, biggest sessions by tokens, retries after errors, prompts grouped by task type (build, debug, deploy, research, writing, UI, review, other), sessions grouped by project, with subagent runs marked. Save it as harness_summary.json. PART 2: BUILD THE DASHBOARD One HTML file, dark theme, with: a top bar of headline numbers (tool calls, prompts, error rate, tokens, worst case latency), a searchable, sortable table of every session, a click through view of each session's steps with status and timing, charts for tokens over time by model and sessions by project, an error explorer grouped by tool and error type, health scores clearly labeled as estimates. Limit to the top 200 sessions by tokens if the file gets too big. PART 3: RECOMMEND CHANGES Write HARNESS.md proposing updates to my CLAUDE.md or AGENTS.md: which model to use for which task type, based on what worked and what failed, tools to keep, fix or drop, a skill for every task I repeated 3+ times, a guardrail for each of my top 3 failure patterns, one fix for each of my top 5 token sinks. Cite the sessions behind every recommendation. ----------------------------------- PART 4: CHECK WITH ME Show me the 5 headline numbers and the proposed changes as a diff. Change nothing until I say yes.