The ranking
25 rowsRanked on the score, not on stars and not on anyone’s opinion. To appear at all a subagent must load cleanly, score at least 78, sit in a repository with at least 1,000 stars, and have been pushed to within 45 days. One row per repository, the best one it publishes.
The 197 qualifying repositories publish 2,714 qualifying subagents between them. Every row below opens on the four components and the arithmetic that produced its score.
The order and the scores come from two different reads, and on one row they disagree. The ranking is the view this page is built from, which holds its own copy of each score. The Score column is the listing's own, which is what its four components add up to and what its page states. The view ranked code-reviewer at 95 and its listing scores 93.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
92/100
- no commits in the last 12 weeks
73/100
- 52,651 stars on the source repo.
81/100
- 1,314 words with worked examples.
100 × 0.40 = 40.0+ 92 × 0.25 = 23.0+ 73 × 0.20 = 14.6+ 81 × 0.15 = 12.15= 89.75+ 3 for an official publisher, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
82/100
- 257,226 stars on the source repo.
81/100
- 957 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 82 × 0.20 = 16.4+ 81 × 0.15 = 12.15= 93.55, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
82/100
- 250,096 stars on the source repo.
74/100
- 665 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 82 × 0.20 = 16.4+ 74 × 0.15 = 11.1= 92.5, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
70/100
- 36,185 stars on the source repo.
73/100
- 821 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 70 × 0.20 = 14.0+ 73 × 0.15 = 10.95= 89.95+ 3 for an official publisher, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
68/100
- 23,995 stars on the source repo.
73/100
- 964 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 68 × 0.20 = 13.6+ 73 × 0.15 = 10.95= 89.55+ 3 for an official publisher, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
94/100
- no commits in the last 12 weeks
- no license file
79/100
- 144,875 stars on the source repo.
66/100
- 666 words with worked examples.
100 × 0.40 = 40.0+ 94 × 0.25 = 23.5+ 79 × 0.20 = 15.8+ 66 × 0.15 = 9.9= 89.2+ 3 for an official publisher, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
75/100
- 82,346 stars on the source repo.
81/100
- 2,919 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 75 × 0.20 = 15.0+ 81 × 0.15 = 12.15= 92.15, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
75/100
- 80,112 stars on the source repo.
81/100
- 1,949 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 75 × 0.20 = 15.0+ 81 × 0.15 = 12.15= 92.15, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
75/100
- 72,251 stars on the source repo.
81/100
- 1,068 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 75 × 0.20 = 15.0+ 81 × 0.15 = 12.15= 92.15, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
74/100
- 67,661 stars on the source repo.
81/100
- 996 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 74 × 0.20 = 14.8+ 81 × 0.15 = 12.15= 91.95, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
72/100
- 43,972 stars on the source repo.
81/100
- 4,251 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 72 × 0.20 = 14.4+ 81 × 0.15 = 12.15= 91.55, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
94/100
- no commits in the last 12 weeks
- no license file
80/100
- 170,171 stars on the source repo.
74/100
- 408 words with worked examples.
100 × 0.40 = 40.0+ 94 × 0.25 = 23.5+ 80 × 0.20 = 16.0+ 74 × 0.15 = 11.1= 90.6, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
78/100
- 127,200 stars on the source repo.
66/100
- 613 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 78 × 0.20 = 15.6+ 66 × 0.15 = 9.9= 90.5, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
76/100
- 93,775 stars on the source repo.
73/100
- 1,770 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 76 × 0.20 = 15.2+ 73 × 0.15 = 10.95= 91.15, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
68/100
- 90,195 stars on the source repo.
81/100
- 1,567 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 68 × 0.20 = 13.6+ 81 × 0.15 = 12.15= 90.75, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
74/100
- 65,879 stars on the source repo.
74/100
- 562 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 74 × 0.20 = 14.8+ 74 × 0.15 = 11.1= 90.9, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
71/100
- 39,603 stars on the source repo.
81/100
- 1,532 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 71 × 0.20 = 14.2+ 81 × 0.15 = 12.15= 91.35, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
71/100
- 38,945 stars on the source repo.
81/100
- 976 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 71 × 0.20 = 14.2+ 81 × 0.15 = 12.15= 91.35, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
69/100
- 30,647 stars on the source repo.
81/100
- 1,248 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 69 × 0.20 = 13.8+ 81 × 0.15 = 12.15= 90.95, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
69/100
- 30,677 stars on the source repo.
81/100
- 1,920 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 69 × 0.20 = 13.8+ 81 × 0.15 = 12.15= 90.95, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
92/100
- no commits in the last 12 weeks
62/100
- 9,436 stars on the source repo.
81/100
- 1,083 words with worked examples.
100 × 0.40 = 40.0+ 92 × 0.25 = 23.0+ 62 × 0.20 = 12.4+ 81 × 0.15 = 12.15= 87.55+ 3 for an official publisher, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
64/100
- 45,632 stars on the source repo.
81/100
- 2,013 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 64 × 0.20 = 12.8+ 81 × 0.15 = 12.15= 89.95, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
94/100
- no commits in the last 12 weeks
- no license file
71/100
- 39,755 stars on the source repo.
81/100
- 1,542 words with worked examples.
100 × 0.40 = 40.0+ 94 × 0.25 = 23.5+ 71 × 0.20 = 14.2+ 81 × 0.15 = 12.15= 89.85, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
100/100
- no commits in the last 12 weeks
62/100
- 9,047 stars on the source repo.
81/100
- 1,730 words with worked examples.
100 × 0.40 = 40.0+ 100 × 0.25 = 25.0+ 62 × 0.20 = 12.4+ 81 × 0.15 = 12.15= 89.55, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
100/100
- Loads cleanly: valid frontmatter, required fields present, no dangling references.
94/100
- no commits in the last 12 weeks
- no license file
74/100
- 68,988 stars on the source repo.
73/100
- 918 words with worked examples.
100 × 0.40 = 40.0+ 94 × 0.25 = 23.5+ 74 × 0.20 = 14.8+ 73 × 0.15 = 10.95= 89.25, which rounds to
Loads. Every structural check Claude Code needs to register this passed.
What makes a Claude Code subagent good?
This is a floor, not a coronation. A high score means a subagent parses, declares real tools, and is still maintained, not that it is good at the job its description claims. Nothing here has been executed. Read how the score is built before treating a number as a recommendation.
A good Claude Code subagent starts by being a subagent Claude Code will actually register, and that is a documented, checkable property rather than a matter of taste. Anthropic’s subagent reference states that of the YAML frontmatter fields, “only name and description are required”, name being a “unique identifier using lowercase letters and hyphens” and description being when Claude should delegate to it. A file missing either is not a weak subagent; it is a file Claude Code never loads, however good the prose beneath the frontmatter is.
Above that floor, the four things worth reading are the ones this index scores: whether every tool in the allow-list is a tool Claude Code actually ships, whether the repository has been touched recently, whether anyone else uses it, and whether the instructions are long enough and specific enough to change behaviour. The description matters more than its length suggests, because it is the only thing Claude sees when deciding whether to delegate. A subagent with a vague description is one that never gets invoked.
Why do 25,450 of 120,796 published subagents fail to load?
21.1% of them21.1% of the Claude Code subagents this index can find on GitHub fail a structural check, and almost all of it is frontmatter. Broken or absent YAML frontmatter accounts for 1,001 of them; a further 6,610 parse but declare no name and 2,616 declare no description. Because Anthropic's reference makes both fields required on a subagent, each of those is a file that sits in .claude/agents/ looking correct and is never registered.
This is the gap a README list cannot show you. A repository can advertise forty subagents, collect thirty thousand stars, and ship files that Claude Code silently skips. Nothing in the GitHub interface reads the frontmatter and tells you. Every subagent that fails a check is listed in full on the will-not-load page, named, with the check it failed.
Where do Claude Code subagents live?
A Claude Code subagent is a markdown file in .claude/agents/ for one project, or ~/.claude/agents/ to make it available in every project on the machine. Anthropic’s documentation states that Claude Code scans both directories recursively, “so you can organize definitions into subfolders such as agents/review/ or agents/research/”.
The subdirectory path does not affect how a subagent is identified, and neither does the filename: identity “comes only from the name frontmatter field”, and “the filename doesn’t have to match”. That is why this index never deducts points from a subagent whose filename and name differ, and does apply that check to skills, where the directory name is what supplies the command.
Are Claude Code subagents the same as skills?
No, and the frontmatter rules differ in a way that matters when comparing them. A skill is a SKILL.md file in a directory under .claude/skills/, and on a skill the name field is not required at all. Anthropic’s reference marks it “Required: No” and says it “defaults to the directory name”. description is only “Recommended”: if omitted, a skill “uses the first paragraph of markdown content”. On a subagent both are mandatory. The same missing field is fatal on one and merely a deduction on the other.
Skills are also not a Claude-only format. Anthropic states that “Claude Code skills follow the Agent Skills open standard, which works across multiple AI tools”, with Claude Code extending it with invocation control, subagent execution and dynamic context injection. Subagents, by contrast, are a Claude Code construct. If portability across tools is a requirement, that difference decides which of the two you should be writing.
Why only one subagent per repository?
Because every artifact in a monorepo inherits that repository's maintenance, adoption and trust scores, so they land on the same number and a straight top-25 stops being a ranking. Ranked on score alone, the top eighteen subagents in this index are eighteen files from a single repository: a correct answer to the query as asked, and a useless answer to the question behind it.
Collapsing to the best row per repository trades a little precision for a list that actually spans the ecosystem. It also means this page is a starting point rather than a verdict. The repositories here publish 2,714 qualifying subagents between them, and the full scored directory is where to go when you know which job you are hiring for.
Sources
Frontmatter requirements, file locations and the filename rule are quoted from Anthropic’s Create custom subagents reference. The skill comparison is quoted from Extend Claude with skills. Both were re-read against the live documentation on 9 August 2026, and every sentence quoted above was still on the page word for word. Every count, score and star figure on this page is read out of this index’s own nightly rebuild rather than quoted from a repository’s README.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
69/100
- 28,315 stars on the source repo.
73/100
- 1,219 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 69 × 0.20 = 13.8+ 73 × 0.15 = 10.95= 77.75, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
59/100
- 5,108 stars on the source repo.
81/100
- 1,905 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 59 × 0.20 = 11.8+ 81 × 0.15 = 12.15= 76.95, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
59/100
- 5,108 stars on the source repo.
81/100
- 2,647 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 59 × 0.20 = 11.8+ 81 × 0.15 = 12.15= 76.95, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
59/100
- 5,108 stars on the source repo.
81/100
- 2,664 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 59 × 0.20 = 11.8+ 81 × 0.15 = 12.15= 76.95, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
59/100
- 5,108 stars on the source repo.
81/100
- 1,981 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 59 × 0.20 = 11.8+ 81 × 0.15 = 12.15= 76.95, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
59/100
- 5,108 stars on the source repo.
81/100
- 1,341 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 59 × 0.20 = 11.8+ 81 × 0.15 = 12.15= 76.95, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
57/100
- 3,735 stars on the source repo.
81/100
- 877 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 57 × 0.20 = 11.4+ 81 × 0.15 = 12.15= 76.55, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
62/100
- 9,242 stars on the source repo.
74/100
- 365 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 62 × 0.20 = 12.4+ 74 × 0.15 = 11.1= 76.5, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
100/100
- no commits in the last 12 weeks
62/100
- 9,242 stars on the source repo.
74/100
- 369 words with worked examples.
70 × 0.40 = 28.0+ 100 × 0.25 = 25.0+ 62 × 0.20 = 12.4+ 74 × 0.15 = 11.1= 76.5, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
92/100
- no commits in the last 12 weeks
68/100
- 25,892 stars on the source repo.
81/100
- 1,272 words with worked examples.
70 × 0.40 = 28.0+ 92 × 0.25 = 23.0+ 68 × 0.20 = 13.6+ 81 × 0.15 = 12.15= 76.75, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
92/100
- no commits in the last 12 weeks
68/100
- 25,892 stars on the source repo.
81/100
- 1,217 words with worked examples.
70 × 0.40 = 28.0+ 92 × 0.25 = 23.0+ 68 × 0.20 = 13.6+ 81 × 0.15 = 12.15= 76.75, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.
70/100
- No `name` in frontmatter, which a subagent requires to be registered at all.
92/100
- no commits in the last 12 weeks
68/100
- 25,892 stars on the source repo.
81/100
- 971 words with worked examples.
70 × 0.40 = 28.0+ 92 × 0.25 = 23.0+ 68 × 0.20 = 13.6+ 81 × 0.15 = 12.15= 76.75, which rounds to
Does not load. The frontmatter has no name field. A subagent requires one to be registered at all; a skill falls back to its directory name.