How do 94 engineers actually use agents?

$autoreview, before every commit.

Showboat, agents demo their own work.

/go, test, /simplify, open PR.

autoresearch, overnight agent research.

Harness engineering, after every agent mistake.

"Review this with claude", adversarial review from Codex.

/answer, in place of plan mode.

T3 Code, his own GUI over every agent.

QA_BEFORE_RELEASES.md, an agent QA pass per release.

Field guide, agent writes its own lessons.

Seats, 18 long-lived Fable officers.

tm, a tmux session per site.

Daily code limit, sized to what he can review.

“Irrefutable proof”, video proof, not line reads.

/par, rival reviewers, 5 points.

"fresh eyes" review, a subagent rereads the code.

TCR skill, red tests revert the code.

/ce-compound, turns each fix into a lesson.

/grill-me, before every change.

Claudescope, transcript link in every PR.

/loop PR triage, on Agent Skills, every day.

The Ralph loop, an agent in a bash while loop.

"I'm feeling Pucky" button, starts half their work.

Voice prompting (Hex), instead of typing.

"Chief of staff" thread, spawns every project thread.

Every tedious OSS task, fed to cloud Claude Code.

antfubot, agents ship PRs as a bot.

Welfare checks, on workers across 4 machines.

This Old Repo, reviving a 1990s game.

sensors check, a sidecar the agent polls.

A lead agent per project, he just files issues.

"I wrote this PR with an LLM", flagged for careful review.

/pr-pass, loops until CI and bots pass.

Fresh config per model, reset at every model launch.

Cross-provider runner, Codex, Claude spawn each other.

`notify msg`, in CLAUDE.md, not a hook.

meat, strips diffs to what matters.

LGTM, the PR closer, a bot that clears her PR queue.

Named-persona review panels, with contrarian personas.

Plumb, reviews decisions at commit.

/show-me, diagrams, not walls of prose.

"Want me to compound?", asked after every sweep.

"Using a subagent in parallel, do X", from one always-on thread.

sl follow / sl adopt, restack parallel agents.

"Reload agent.md", when code quality drops.

filippo-agent, agent's own GitHub account.

/plan-ceo-review, finds the 10-star product.

AI Disclosure, README names the exact model.

/explain-diff, with a quiz before he ships.

(kimi) in the PR title, names the model per PR.

Adversarial review, 2 Claudes refute every change.

Capability Owners, standing agents per feature.

Adversarial review agents, on compiler & security work.

claude-air, nudges after 8 silent rounds.

$ultragoal, goal kept in an editable file.

/ship-pr, Bugbot loop, merge, Discord.

plans/ in the repo, checked in for code review.

Fail-first fixtures, before every extraction fix.

Plan mode, before every long agent run.

/linear-project-loop, loops a project to its goal.

maggie-design, her design taste, as a skill.

Hand-typed skeletons, Claude fills in the bodies.

"You are AGI-pilled.", in every agent's system prompt.

AgentIDE, his own vibe-built agent IDE.

iOS Claude pings Android Claude, to port each feature.

codex exec Luna subagents, 7–12 cheap critics per pass.

/gemini-feedback, Gemini critiques each diagram.

Be introspective, Claude fixes its own rules.

code-review-turbo, three-agent PR bug hunt.

/commit, human writes the headline.

paulirish-agent, a GitHub account for agents.

Audit the agent first, before installing any harness.

gdp-ts, auth checks as type proofs.

Claw Patrol, firewall between agents and prod.

30-second triage, rejects most agent attempts.

Fourth-wall check, enforced in CI on every PR.

!land playbook, refreshes skills on every PR.

rue-overnight-hunt, bugs filed only with repros.

$cassandra-creator, a standing dissent agent.

Five favorable conditions, before handing code to an agent.

Change quiz, must pass before merging.

AI Disclaimer, names agent and model in PRs.

Random-file bug hunt, automation, several times a day.

Pi on GLM 5.2, as the security bug hunter.

qmd, local search, agents too.

Live nREPL for the agent, instead of app restarts.

Pedals in a ring, a bench that tests itself.

swiftui-app template, copied into every new app.

uml-viewer, colored by CRAP + mutation.

Daily customer-report Devin, escalates top feature asks.

vite-plugin-killer-instincts, hands the LLM a kill command.

roborev, reviews every agent commit.

Assisted-by: footer, model + tool on every commit.

Scorer + self-improvement agents, agents that fix agents.