Proof
Proof.
What we've counted, how, and when. Where a result doesn't exist yet, this page says so.
01 · Method
How a number gets on this page.
A number appears here only if it's a count taken from files on disk, or a fact that a test in the repository proves about how a mechanism behaves. Each row below says how it was counted and on what date.
- CountedSkills, agents, commands and workflows are files counted under each plugin's directory.
- EstimatedAlways-on weight is the summed length of every skill, agent and command description, because Claude Code keeps those in context all the time, plus the protocol text dojo-protocol injects at the start of each session or fresh context. Tokens are characters divided by four, so they're an estimate.
/dojo-doctor:checkreports the description part per installed plugin as its always-on weight, and the session-start injection under injection weight, measured from your transcripts. It rounds tokens to the nearest whole number, so its figures can be one higher. - EnforcedThe release lint fails any plugin whose always-on descriptions reach 1,000 characters.
02 · Measured
What's counted.
| Plugin | Skills | Agents | Commands | Workflows | Description characters |
|---|---|---|---|---|---|
| dojo-protocol | 5 | 4 | 0 | 0 | 885 |
| dojo-gates | 1 | 0 | 0 | 0 | 206 |
| dojo-router | 1 | 0 | 0 | 0 | 192 |
| dojo-meter | 0 | 0 | 2 | 0 | 169 |
| dojo-verify | 1 | 0 | 0 | 0 | 173 |
| dojo-flow | 1 | 0 | 2 | 2 | 491 |
| dojo-doctor | 1 | 0 | 1 | 0 | 335 |
| dojo-settle | 3 | 0 | 0 | 0 | 587 |
| Suite total (8) | 13 | 4 | 5 | 2 | 3038 |
| Always-on source | Characters | Estimate, tokens |
|---|---|---|
| Descriptions of skills, agents and commands, kept in context | 3038 | 759 |
| dojo-protocol PROTOCOL.md, injected at session start | 1987 | 496 |
| Always-on total | 5025 | 1255 |
03 · Open
What hasn't been measured.
- Not yet measuredWhether a session with the suite costs less than one without it. A with/without eval is how to find out; the scaffolds ship in dojo-settle, and we haven't run them.
- Not yet measuredWhether it makes Claude Code more accurate.
- Not yet measuredHow often each guard fires in real use.
- Not yet measuredHow the mods behave on Claude Code versions other than the one we tested.
04 · Limits
What we won't claim.
No spend figures, no benchmark numbers, no percentage claims. A number will appear here only after a pre-registered with/without run (the method is in dojo-settle) or a count from disk backs it.