body-tokens

Severity: warning above about 6,000 tokens, a note from 5,000 to 6,000.

What it catches#

SKILL.md body is over about 5,000 tokens.

A SKILL.md whose body, after the frontmatter, is over about 5,000 tokens. The Agent Skills spec recommends staying under that and moving detail into files the skill references, which load only when needed. Token counts are local estimates, so a body of 5,000 to 6,000 tokens is a note and the warning starts above 6,000. When the file is also over 500 lines, this one finding gives both counts instead of a second length finding.

Why it matters#

The Agent Skills spec recommends a body under 5,000 tokens, with detail moved to files the skill links to. Token counts are estimates, so a body up to 6,000 is a note. A body over 500 lines is counted in the same finding.

---
name: aws-cost-report
description: Build the monthly AWS cost report. Use when asked for cloud spend or a cost breakdown.
---

# AWS cost report

## Steps
...

## Appendix A: every Cost Explorer dimension and its filter syntax
...3,000 words...

## Appendix B: sample JSON responses
...2,500 words...

---
name: aws-cost-report
description: Build the monthly AWS cost report. Use when asked for cloud spend or a cost breakdown.
---

# AWS cost report

## Steps
...

Dimension and filter syntax: [references/dimensions.md](references/dimensions.md).
Sample responses: [references/samples.md](references/samples.md).

How to fix#

Keep the steps and the decisions in SKILL.md. Move tables, long examples and API detail into references/ and link them with relative paths, one level deep.

Suppress or configure#

For one finding, put a comment on the line above it. The reason after -- is optional but worth writing:

<!-- asitis-disable-next-line body-tokens -- why this one is fine -->

For a whole file:

<!-- asitis-disable-file body-tokens -- why this file is an exception -->

For every file, turn it off or change its severity in .asitisrc.json:

{ "rules": { "body-tokens": "off" } }

A .asitisrc.json overrides entry limits a change to some paths, and a baseline (asitis lint --format json --write-baseline) accepts today's findings so only new ones fail the build.

Source#

Agent Skills specification, read on 14 October 2026.