Skill
ledger
Use ONLY when the project's own instructions (its CLAUDE.md) name a task ledger or session files. The short form of that…
@maximocorrea1Install SkillAgent Memory
Skill
/usage-check
Use when the user asks where the tokens went, why a usage limit was hit, or for the weekly usage review. Three existing tools, cheapest first, and how to read them.
Install
$ npx -y skills add maximocorrea1/lean-harness --skill usage-check --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
- You can call itInvoke it directly when you want it.
- Slash command
/usage-check
Context preview
The summary Claude sees to decide when to auto-load this skill.
Use when the user asks where the tokens went, why a usage limit was hit, or for the weekly usage review. Three existing tools, cheapest first, and how to read them.
SKILL.md
usage-check.SKILL.mdname: usage-check
description: Use when the user asks where the tokens went, why a usage limit was hit, or for the weekly usage review. Three existing tools, cheapest first, and how to read them.
Where the tokens went
Run the cheapest tool that answers the question. All three read the local transcripts under `~/.claude/projects` and send nothing anywhere.
1. `bunx ccusage@latest daily --since YYYYMMDD` (also `weekly` and `session`). Tokens and list-price cost by day. An existing tool with nothing to maintain. Its dates are local, not UTC. 2. Anthropic's `session-report` plugin (marketplace `claude-plugins-official`). An HTML report of tokens, cache, subagents, skills and the most expensive prompts. 3. For CAUSES, the probe in the checkout that holds this harness (run `claude plugin marketplace list` and take the folder shown for `flowy-harness`; its repository root is the checkout). From that root: `bun docs/measurements/instruments-2026-10-11/claude-usage.probe.cjs --days 7`. Prompt size per call, the calls that rebuilt the cache and what came before them, lead against agents, agents by model, and what re-reading loaded skill text cost. `--from` and `--to` (ISO instants) replace `--days` and measure one local day.
Read the result as a shape, never as the plan's own meter (that one is not published):
- Which sessions grew large. A call costs its whole prompt, so the median prompt of a lead call is
the number to watch. The handoff line (two thirds of the compaction window, as the context line and the CLAUDE.md card state it) is the line.
- Which calls rebuilt the cache. A pause over an hour before a large prompt is the dearest call.
- Which agents ran on the lead's model.
Then name ONE change, make it, and measure again in a week. A rule that does not move a number is not a rule.
Read more
name: usage-check description: Use when the user asks where the tokens went, why a usage limit was hit, or for the weekly usage review. Three existing tools, cheapest first, and how to read them.
Where the tokens went
Run the cheapest tool that answers the question. All three read the local transcripts under `~/.claude/projects` and send nothing anywhere.
1. `bunx ccusage@latest daily --since YYYYMMDD` (also `weekly` and `session`). Tokens and list-price cost by day. An existing tool with nothing to maintain. Its dates are local, not UTC. 2. Anthropic's `session-report` plugin (marketplace `claude-plugins-official`). An HTML report of tokens, cache, subagents, skills and the most expensive prompts. 3. For CAUSES, the probe in the checkout that holds this harness (run `claude plugin marketplace list` and take the folder shown for `flowy-harness`; its repository root is the checkout). From that root: `bun docs/measurements/instruments-2026-10-11/claude-usage.probe.cjs --days 7`. Prompt size per call, the calls that rebuilt the cache and what came before them, lead against agents, agents by model, and what re-reading loaded skill text cost. `--from` and `--to` (ISO instants) replace `--days` and measure one local day.
Read the result as a shape, never as the plan's own meter (that one is not published):
- Which sessions grew large. A call costs its whole prompt, so the median prompt of a lead call is
the number to watch. The handoff line (two thirds of the compaction window, as the context line and the CLAUDE.md card state it) is the line.
- Which calls rebuilt the cache. A pause over an hour before a large prompt is the dearest call.
- Which agents ran on the lead's model.
Then name ONE change, make it, and measure again in a week. A rule that does not move a number is not a rule.
Ships withlean-harness
A Claude Code harness that keeps sessions small and smart: a context line that orders the handoff before a session gets expensive, Sonnet agents for bounded work, a short task-ledger skill and a usage check. Built at Flowy and measured on real sessions there.
Get the whole plugin
Stats
0
Stars
0
Forks
Active
Maintenance
JavaScript
Language
1d ago
Last commit
1d ago
Created
44m ago
Added
Other skills on lean-harness.
@maximocorrea1Install Skill