codeburn-optimize
Use when the user asks about CodeBurn, LLM spend, token usage, rate-limit pressure, model efficiency, Claude vs Codex usage, subscription value, or where coding-agent usage can be optimized.
- License 许可
- MIT
- 最近更新
- 2026-11-07
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- codeburn-optimize
- description
- Use when the user asks about CodeBurn, LLM spend, token usage, rate-limit pressure, model efficiency, Claude vs Codex usage, subscription value, or where coding-agent usage can be optimized.
03 SKILL.md 正文
CodeBurn Optimize
Use codeburn to turn coding-agent logs into spend, usage, efficiency, and optimization advice.
Steps
-
Confirm the CLI exists.
command -v codeburn codeburn --versionDone when the executable path and version are known. If missing, say CodeBurn is unavailable and stop.
-
Pick the smallest useful window.
- Today:
codeburn today --timezone America/Denver - Yesterday:
codeburn models --from YYYY-MM-DD --to YYYY-MM-DD --provider all --format json - This month:
codeburn month --timezone America/Denver - Last 30 days:
codeburn models --period 30days --provider all - All visible logs:
codeburn models --period all --provider all
Done when the command matches the user's timeframe. Use exact dates for "today" and "yesterday".
- Today:
-
For spend and model mix, prefer non-interactive model tables.
codeburn models --period 30days --provider all codeburn models --period all --provider all --format jsonDone when totals are separated by provider and model.
-
For waste, run optimize once per relevant provider.
codeburn optimize --provider all codeburn optimize --provider claude codeburn optimize --provider codexDone when savings, waste type, and exact fix are identified. Skip provider-specific runs if the all-provider result already answers the question.
-
For Claude vs Codex decisions, compare efficiency, not only dollars. Report:
- API-equivalent cost
- calls and sessions
- one-shot rate
- retries/edit when visible
- top projects and activities
- cache hit rate
Done when the recommendation says which plan/model should be the workhorse and why.
-
Keep the money caveat explicit. CodeBurn cost is API-equivalent token value unless the data proves actual billing. Subscription-covered usage is not the same as card spend. Done when the answer separates "API-equivalent" from "cash paid".
Output
Lead with the answer, then the evidence.
Use this shape:
Total: $X API-equivalent.
Provider split:
- Claude: $X
- Codex: $X
Optimization read:
- Biggest waste: ...
- Best fix: ...
- Subscription read: ...
Verified with: `command`
Guardrails
- Do not use
codeburn reportunless the user asks for the interactive dashboard. - Do not paste full JSON. Aggregate with
jq, tables, or a short summary. - Do not treat API-equivalent value as actual subscription billing.
- Do not recommend upgrading a plan from raw spend alone. Use one-shot rate and retry pressure too.
04 同仓库更多 Skills
writing-for-agents
Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.
wait-what
Stop. That last message did not land — re-pitch it.
ponytail
Lazy senior dev mode for any coding task (write, refactor, fix, review): YAGNI, stdlib first, no unrequested abstractions. Not for non-coding requests.
i-have-adhd
Shape output for a reader with ADHD. Use this skill whenever responding to ANY user message including coding tasks, debugging, explanations, planning, and casual conversation. Output should lead with concrete next actions, number multi-step work, externalize state across turns, suppress tangents, give specific time estimates, and make wins visible. Trigger even on casual messages and even when the user did not explicitly ask for brevity.
x-research
Use when the task needs live posts from X/Twitter, or when deep-research needs its X pass.
slop
Answer an empirical question, or explore/test something about real code, by building and running throwaway code in ~/slop/ — harnesses, lint rules, load tests, ports, API probes. Never merged. Use whenever disposable code would answer, explore, or test something faster or more reliably than reasoning about it.