SKM-B02 · Warning
Body is around or over 5,000 tokens
The specification recommends keeping the activated body under roughly 5,000 tokens.
What the rule is
The specification's activation-stage guidance recommends staying under roughly 5,000 tokens for the body content that loads when a skill fires. This tool estimates tokens as characters divided by four — a rough approximation, not a real tokenizer result, since no tokenizer ships with this checker.
This measures the same underlying cost as SKM-B01 (line count) from a different angle — a file can pass one and fail the other depending on line length and content density.
Where it comes from
- Agent Skills — SpecificationSpecification
- SKILL.md, the format hubSpecification
Which runtimes enforce it
| Runtime | Effect | Note |
|---|---|---|
| Claude Code | Degrades | Same mechanism as SKM-B01 — a larger activation payload costs more context on every runtime, independent of the line-count check. |
| Codex | Degrades | Same mechanism as SKM-B01 — a larger activation payload costs more context on every runtime, independent of the line-count check. |
| Cursor | Degrades | Same mechanism as SKM-B01 — a larger activation payload costs more context on every runtime, independent of the line-count check. |
| OpenClaw | Degrades | Same mechanism as SKM-B01 — a larger activation payload costs more context on every runtime, independent of the line-count check. |
| Hermes Agent | Degrades | Same mechanism as SKM-B01 — a larger activation payload costs more context on every runtime, independent of the line-count check. |
How to fix it
- Move detail out of the activated body and into `references/`, the same fix as SKM-B01. Treat the token estimate as directional, not exact.
Elsewhere
- SKILL.md, the formatThe full frontmatter field table this rule is drawn from.
- Agent Skills Not WorkingThe diagnostic reference this rule’s portability and failure evidence comes from.
- All rulesEvery check the validator runs, grouped by class.