bytedance/deer-flow

Agents & automation
Data & research

Graded C on context cost and craft. 2,682 tokens of this skill ride in the system prompt of every message, whether or not it fires.

Ranked #115 of 124 on the leaderboard →

Score: 1.80 / 4
Resident: 2,682 tokens
Footprint: heavy
Per activation: 9,031 tokens

Share this grade on X ↗Re-grade now →View on GitHub ↗Grade another skillLast graded Oct 18, 2026 from HEAD

Add the badge to your README

It shows the current letter grade and links back to this page. Improve the skill, hit Re-grade now, and the badge follows — GitHub may keep the old image for up to an hour.

Markdown

[![Skill Grader: C](https://seoagent.com/skill-grader/badge/bytedance/deer-flow.svg)](https://seoagent.com/skill-grader/bytedance/deer-flow)

HTML

<a href="https://seoagent.com/skill-grader/bytedance/deer-flow"><img src="https://seoagent.com/skill-grader/badge/bytedance/deer-flow.svg" alt="Skill Grader: C" height="20"></a>

Report card

C overall · 1.80/4

github.com/bytedance/deer-flow@HEAD · 33 skills · 2,682t resident (heavy) · 9,031t per activation · 107,800t bundle

Resident footprint

2,682t of description text ride in the system prompt of every message across 33 skills (paper's observed per-skill range: 50–280t).

Description honesty

Defensive padding detected: 207 comma-separated trigger keywords; 1440 words (a quiet description is ~15–40). Padding is individually rational and collectively ruinous — it dilutes every co-installed skill.

Body size

Largest body is 5,600 words (~9,031t), loaded on every trigger. Corpus median 921, p90 2207.

Progressive disclosure

Body routes to 13 reference file(s); 51 auxiliary .md files are fetched at point of use instead of riding in the body.

Factoring

Scoped to a coherent procedure set.

CLI leverage

Drives a CLI (`make`, 14 invocations + 55 shell blocks) — verification and state live in code, not prompt.

C overall (1.80/4). 33 skill(s), 2,682t resident on every message (heavy), 9,031t loaded per activation. Weakest axis: resident footprint (F).

Six deterministic axes, no LLM, same result every run. Tokens are approximated as chars/4. Read the methodology, or see how SEO skills score on task outcomes in the open SEO skill benchmark.