AgentRigs (@AgentRigs) header
AgentRigs (@AgentRigs) profile picture

The leaderboard for AI coding agent setups. 7,800+ rigs scored. Steal the best one. by @bleedhigher

Joined October 2026
Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)Photo by AgentRigs (@AgentRigs)
rig of the day: ZacheryGlass/.claude what i like: the auto mode environment block. it tells the classifier which repos are private and which files hold secrets one tip: WebFetch(domain:*) is an allow rule, and allow rules resolve before the classifier, so fetches skip it
1
3
on models with a native 1M window, claude code auto-compacts at about 967k tokens by default to compact earlier, run /autocompact 500k. it gets saved for the current model in user settings one launch only: claude --autocompact 500k back to the default: /autocompact auto
1
7
commit is the git skill agent setups ship most 339 rigs i index have a skill named exactly commit, the #10 skill overall after it: pr 180, git-workflow 110, pr-review 105, using-git-worktrees 104 exact names only, so git-commit (74) and commit-message (42) count separately
1
2
rig of the day: citypaul/.dotfiles what i like: skill evals. requests that never name a skill check that claude loads the right one and its neighbours stay quiet one tip: the format hook always exits 0, so lint errors --fix can't fix never reach claude. exit 2 would show them
2
12
claude code skills can take arguments write $ARGUMENTS in the skill and /fix-issue 123 puts 123 in its place for more than one, use $0 $1 $2 by position and quote multi-word values no placeholder at all? claude code appends ARGUMENTS: <what you typed> to the end
1
8
30% of the personal setups i index that commit mcp config give the agent a docs server 161 of 536, and 140 of those use context7 project repos: 188 of 2,341 (8%) counted: context7, deepwiki, and the aws, microsoft learn and openai docs servers
1
14
rig of the day: obra/dotfiles what i like: scripts need a clear name, help text, errors written for an agent, and short output that says where the full logs are one tip: windows-vm grants bare Bash in allowed-tools. rules like Bash(docker *) would scope it to what it runs
1
4
claude code's lsp plugins don't ship the language server itself install the binary first (gopls, rust-analyzer, pyright-langserver...) and make sure it's on the PATH you start claude from then claude sees type errors from its own edits without running a build
1
15
103 setups i index score 5/5 on my guardrail checklist. their median quality score is 41/100 the checklist asks whether each guardrail exists. quality mostly replays 22 risky commands and 6 kinds of secret files against the rules only 14 of the 103 clear 60
1
3
rig of the day: lucasfcosta/dotfiles by @thewizardlucas what i like: plans go in phases that each end in a working state, with how to validate it one tip: a PreToolUse hook on Bash(git commit *) that exits 2 if lint or tests fail would enforce the no-commit-until-green rule
1
6