What level are you with agents?
100 tricks, from your first prompt to your own agent. Your agent checks which ones you use.
/100
Your agent noticed · only you see this
works inClaude CodeCursorCodex
Where you'd start
Try an agent in ten minutes
Install one, open a folder you care about, and paste the same line.
- Claude desktop app Claude Code is built in
- Codex from OpenAI, app or CLI
- Cursor an editor with an agent inside
The 100
Unlock next
Your card is saved anonymously at its link: the 100 answers and the agents you use. What your agent noticed stays in your own link. A name and photo go up only when you share.
0
First contact
It edits, it runs things, it answers. You stop copy-pasting.
How it's checkedAny agent config folder exists: ~/.claude, ~/.codex, ~/.cursor, ~/.config/opencode, ~/.gemini, ~/.factory, ~/.pi. You're reading this from inside one, so: yes.
“To find value, you must use an agent.”
Mitchell Hashimoto · source
How it's checkedIts file-edit tool shows up in session logs (Edit, Write, apply_patch).
How it's checked10+ prompts that give the reason, not just the task ("because", "so that", "the goal is"), or 20+ prompts over 40 words.
“Just talk to it. Play with it. Develop intuition.”
Peter Steinberger · source
How it's checked5+ prompts that are questions about the code (how, why, where, what does).
Ask it the questions you'd ask a senior engineer about the codebase.
Anthropic, Claude Code best practices, paraphrased · source
How it's checked3+ prompts with an image attached ([Image #…]).
Try⌘⇧⌃4 on a Mac copies it, then paste
“I’d say at least 50% of my prompts contain a screenshot.”
Peter Steinberger · source
How it's checked5+ prompts with pasted blocks or raw logs ([Pasted text …], stack traces).
How it's checkedA dictation app is installed (Wispr Flow, superwhisper, MacWhisper, VoiceInk, OpenWhispr, Aqua, Spokenly, Handy), /voice is on in Claude Code, or prompts read like speech (um, uh, long run-ons).
TryOpenWhispr free, open source · Wispr Flow · superwhisper · /voice built into Claude Code
“Also I just talk to Composer with SuperWhisper so I barely even touch the keyboard.”
Andrej Karpathy · source
How it's checked5+ interrupted turns in session logs ("[Request interrupted by user]").
“Don't be afraid of stopping models mid-way.”
Peter Steinberger · source
How it's checked/clear, /new or a fresh session after a failed attempt, 3+ times.
After two failed corrections, /clear and write a better first prompt.
Anthropic, Claude Code best practices, paraphrased · source
How it's checked1+ session in a folder that isn't a git repo (Downloads, Documents, home).
“I’ve been saying for a while now that Claude Code is a “general agent” disguised as a developer tool.”
Simon Willison · source
Context
It knows what you know, because you told it.
How it's checkedAn AGENTS.md, CLAUDE.md, GEMINI.md or .cursor/rules in any repo you work in.
“I started with zero custom rules. I only added a rules file later after the model got a few things wrong”
Lee Robinson · source
How it's checked~/.claude/CLAUDE.md or ~/.codex/AGENTS.md exists and isn't empty.
How it's checked10+ prompts with @file mentions or file paths, or prompts that name a file to copy ("like X", "same as Y").
How it's checked10+ prompts with URLs, or web fetches in session logs.
How it's checkedPlan mode used, or 5+ prompts asking for a plan before code.
“Plan mode in Cursor is one of the most underrated features.”
Lee Robinson · source
Pros disagree“I can just write “let’s discuss” or “give me options” and it will diligently wait until I approve it. No harness charade needed. Just talk to it.”
Peter Steinberger · source
How it's checked5+ prompts asking for variants, options or versions.
How it's checkedIts ask-the-user tool used, a grill-me style skill installed, or prompts like "ask me" / "interview me".
grill-me
Matt Pocock · source
How it's checked5+ prompts asking for a decision or a critique ("should I", "help me decide", "what am I missing", "push back", "poke holes", "devil's advocate").
“AI can provide good strategic or entrepreneurial advice if you are capable of executing on it.”
Ethan Mollick · source
How it's checkedA SPEC.md, PLAN.md or plans/ folder in a repo you work in.
“if you want more than a prototype, you need a real spec.”
Indragie Karunaratne · source
How it's checkedA rules file changed 3+ times, agent memory notes saved, or prompts like "add this to AGENTS.md".
“Anytime we see Claude do something incorrectly we add it to the CLAUDE.md”
Boris Cherny · source
Checks
It proves the thing works before it says done, and you stop babysitting it.
How it's checkedPrompts that state a finish line ("done when", "until", "should show").
Goal, Context, Constraints, Done when: their prompt template.
OpenAI, Codex best practices, paraphrased · source
How it's checked10+ prompts that ask it to verify, test, or make sure it works.
“give Claude a way to verify its work… 2-3x the quality”
Boris Cherny · source
How it's checked5+ prompts asking for proof, screenshots or recordings of the result.
“give me irrefutable proof that this works. Dazzle me. Give me a video as proof, or something.”
Thorsten Ball · source
How it's checkedA browser MCP or skill installed, or 10+ browser-automation calls.
How it's checkedPrompts about failing tests, TDD or red/green, or a TDD skill installed.
“Verify the same fixture fails before the fix and passes afterward; do not change expectations to match buggy behavior.”
Steph Ango · source
Pros disagree“I personally have stopped telling my coding agents to write tests first, let alone do TDD (which I never did, to be honest), until I see evals or other strong arguments that convince me otherwise.”
Birgitta Böckeler · source
How it's checkedIt ran git commit 5+ times in your sessions, or you use checkpoints.
How it's checked/rewind or a checkpoint restore used by you at least once (typed, not just mentioned).
How it's checkedPermission allow rules, Codex rules, or auto mode configured.
How it's checkedDeny rules that cover .env, keys or credential files, or a secrets manager (1Password CLI, Doppler, Infisical) that agents call instead of reading raw keys.
Try1Password CLI op run · Doppler
“A compromised agent process can’t leak keys it never held in the first place.”
Ryan Dahl · source
How it's checked20+ queued prompts in your agent's stats or logs, or /btw used.
Tools
It reaches past the repo: your CLIs, your computer, your inbox.
How it's checkedIt ran 3+ different third-party CLIs (gh, aws, gcloud, vercel, netlify, stripe, supabase, wrangler, fly…).
How it's checkedSessions about the machine itself: disk space, installs, PATH, a slow Mac (du, df, brew doctor, prompts like "why is my mac slow"), 3+.
“How can I clear some storage on my computer?”
Lenny Rachitsky · source
How it's checkedSmall single-purpose tools started in agent sessions: a single HTML file, a one-off script, a repo with a handful of commits.
TrySimon's 150+ tools for ideas
“I have built over 150 of these in the past two years, almost all of them written by LLMs.”
Simon Willison · source
How it's checked1+ skill folder in ~/.claude/skills, ~/.codex/skills, ~/.agents/skills or a repo's .claude/skills.
How it's checked1+ MCP server configured.
Pros disagreePrefers CLIs: the GitHub MCP alone ate 23k tokens of context.
Peter Steinberger, paraphrased · source
How it's checkedIt ran gh pr create (or similar) 3+ times.
How it's checkedIt ran a deploy CLI 3+ times.
How it's checked/effort or /model switched 3+ times, or reasoning effort set in config.
How it's checkedA mail, chat or calendar connector or MCP configured (names only).
“I go to my email client, and check the drafts. Edit and send most of them cuz they are great.”
Harper Reed · source
Your setup
Rules, skills, hooks and memory that you wrote yourself.
How it's checkedA skill or custom command you wrote yourself (no upstream source or lockfile entry, or it lives in your own repo); .claude/commands counts.
How it's checked10+ subagent runs in session logs.
“Then, when I need anything new, I say "using a subagent in parallel, do X"”
Alexander Embiricos · source
How it's checkedA post-edit hook that runs a formatter or linter.
“We use a PostToolUse hook to format Claude's code.”
Boris Cherny · source
How it's checkedA pre-tool hook that blocks or guards commands.
“Replaces the bash tool with a `uv`-aware version that injects Python command shims and blocks common non-`uv` workflows.”
Armin Ronacher · source
How it's checked5+ memory notes, Codex memories on, notes, PROGRESS or HANDOFF files in your repos, or a handoff skill.
“Create a notes.md file in that folder and append notes to it as you work, tracking what you tried and anything you learned along the way.”
Simon Willison · source
How it's checkedA CLAUDE.md that is a symlink to, or imports, an AGENTS.md (or the other way round).
“~/.codex/AGENTS.md -> ~/Projects/agent-scripts/AGENTS.MD ~/.claude/CLAUDE.md -> ~/Projects/agent-scripts/AGENTS.MD”
Peter Steinberger · source
How it's checkedThe rules file of the repo you work in most is under 150 lines and points to other docs for the rest.
“This file is included in full context for every agent conversation. Keep it tiny and stable.”
Kent C. Dodds · source
How it's checkedA hook or script that notifies you when it's done (sound, push, notification), or a rule telling it to.
Tryntfy push to your phone, open source
“when it's done work it runs a `notify msg` bash function I have”
Chris McCord · source
Pros disagreeTurns agent notifications off, so he decides when to switch context.
Mitchell Hashimoto, paraphrased · source
How it's checkedTwo sessions active in the same 10 minutes, on 5+ occasions (from prompt timestamps).
TryConductor parallel agents, each in its own worktree
How it's checkedgit worktree list shows 2+ in any repo.
“Worktrees are essential… dozens of Claudes”
Boris Cherny · source
Pros disagree“Why not worktrees? I run one dev server, as I evolve my project I click through it and test multiple changes at once.”
Peter Steinberger (2025, he uses them now) · source
Second opinions
Another agent reads it before you do.
How it's checkedA review command, review skill or review subagent used 3+ times, in a context that didn't write the code.
“making a human review a change before an LLM is disrespectful.”
Filippo Valsorda · source
How it's checkedAn explain-diff or quiz skill, or prompts asking it to explain or quiz you on a change.
“I only merge after I pass the quiz perfectly.”
Thariq Shihipar · source
How it's checked3+ prompts asking for a diagram, chart or visual explanation.
“/show-me prompts the agent to use concise visuals to explain what's happening, instead of walls of prose.”
Dex Horthy · source
How it's checkedIt checked CI or PR status (gh pr checks, gh run) 5+ times.
“Babysit a PR. Iterate with AI reviewers and CI. Get it ready and maybe merge. Send summary message.”
Kent C. Dodds · source
How it's checked3+ headless runs in shell history or session logs (codex exec, claude -p, gemini -p), including one agent launching another.
How it's checkedroborev, or a git post-commit hook that runs a reviewing agent (codex exec, claude -p).
Tryroborev
“I do not code with agents without a continuous reviewer running in the background.”
Wes McKinney · source
How it's checkedReviews run by 2+ competing or adversarial agents, or rounds of review and fix between two agents.
“Assign two subagents to review your work adversarially. Tell them that whoever finds the largest number of legitimate significant issues gets 5 points.”
Jesse Vincent · source
How it's checkedFindings from 2+ review tools merged into one triaged list.
TryCursor Bugbot · CodeRabbit
“Run a Claude sub-agent, Codex, and Cursor Bugbot to find bugs in this PR ranked by critical/high/medium/low.”
Nolan Lawson · source
How it's checkedYour rules or review setup name the paths that always get a human read.
“I read diffs when it's touching critical code, or code that's important to get right. I don't write the code.”
Boris Cherny · source
Pros disagree“First, I reviewed the entire diff (yes, all ~3,000 lines of it). Then I reviewed it again. I re-read the entire diff by hand three or four times.”
Mitchell Hashimoto · source
Off the desk
It works while you sleep, sit on the couch, or close the laptop.
How it's checkedPermission prompts off AND a sandbox (devcontainer, Docker, VM, /sandbox). Prompts off without a box doesn't count.
TryDev Containers · /sandbox built into Claude Code
“You have to turn off the sandbox, which means you have to provide your own sandbox. I have tried just about everything and I highly recommend: use a fresh VM.”
David Crawshaw · source
How it's checkedAgent network access restricted (a sandbox domain allowlist, or network off in Codex) on a machine where it can read private data or mail.
Try/sandbox limits its network to domains you allow
“The only way to stay safe there is to avoid that lethal trifecta combination entirely.”
Simon Willison · source
How it's checkedA scheduled or recurring agent run (loop/cron tools, a launchd or cron entry that calls an agent).
“One automation I have runs multiple times a day. It picks a random file and finds and fixes a subtle bug.”
Thibault Sottiaux · source
How it's checkedAgent activity between 1am and 6am while you typed nothing.
“This is over-hyped on Twitter, but I do really try and kick off 3-4 tasks in Codex before going to bed, so they'll be ready for review in the morning.”
Calvin French-Owen · source
How it's checkedRemote control set up, a mobile session, or mosh/Tailscale used to reach your agent box. Usually only you can tell, so you tick it.
TryRemote Control Claude Code · Termius · Tailscale
Claude Code is better on your phone
Harper Reed · source
How it's checkedCloud or remote agent tasks in your history. Usually only you can tell, so you tick it.
“I feed this thing every little random tedious OSS task that I'd otherwise neglect on my TODO list”
Adam Wathan · source
How it's checked3+ prompts like "don't stop until" / "keep going until", or a goal set with /goal.
How it's checkedA shell loop or script that re-runs an agent, or /loop until a condition.
“while :; do cat PROMPT.md | claude-code ; done”
Geoffrey Huntley · source
How it's checkedclaude-code-action, codex action or similar in a .github/workflows file.
Orchestra
Many agents, one plan, and a setup that gets sharper with every run.
How it's checkedThe same task run in 2+ parallel sessions or worktrees.
“So I might kind of throw throw codex at it but I might just do it four times in parallel instead.”
Alexander Embiricos · source
How it's checked5+ sessions or background agents running at the same time.
“I regularly have 40+ threads running concurrently.”
Theo Browne · source
How it's checkedOne agent messaging or steering others (agent teams, SendMessage, an orchestrator session), or agents reading each other's output without you relaying it.
“Tell coordinator to parallelize the work over 4 separate computers via SSH”
Andreas Kling · source
How it's checkedPer-agent model settings, or 3+ prompts assigning models to roles.
“Every implementation bead goes through this lifecycle: Fable design, Opus implementation, Fable review.”
Steve Yegge · source
How it's checked2+ custom agent definitions (.claude/agents/*.md or equivalent).
How it's checkedA workflow tool used, or a script that chains several agent runs.
“An agent skill /linear-project-loop which reads in a Linear project and starts by auditing that project’s goal definition on these dimensions”
Will Larson · source
How it's checkedRules, skills or memory edited by agent sessions 5+ times.
“Compound is still the most important step, because the whole point is that every feature should make the next feature easier.”
Kieran Klaassen · source
How it's checkedCustom lint rules, ast-grep rules or check scripts in a repo.
“encode as many rules as possible into deterministic checks (first) and adversarial reviews (second), instead of relying exclusively on a bunch of AGENTS.md files to do the job.”
José Valim · source
How it's checked3+ prompts asking for a codebase-wide change ("migrate", "convert all", "upgrade every", "across the codebase").
“Claude, buddy. You're gonna do the whole conversion in an hour, then we'll see if it works.”
Kenton Varda · source
How it's checkedYour own scripts or CLIs that agents call regularly.
“The CLI can offer both situational awareness for the human in supervised sessions, and a token-efficient and guidance-enriched summary to the agent.”
Birgitta Böckeler · source
Beyond the screen
Your agents reach the team, the release, the pager, and your oldest code.
How it's checkedEval files for a skill or plugin, or a plugin eval run.
How it's checkedYour skills live in a git repo with a remote.
agent-scripts
Peter Steinberger · source
How it's checkedA chat-integrated agent set up (Slack app, Claude Tag, Devin, a bot). Usually only you can tell, so you tick it.
Claude Tag lands 65% of the Claude Code team's product PRs.
Cat Wu and Thariq Shihipar, via Simon Willison, paraphrased · source
How it's checkedRemotion, Hyperframes or ffmpeg renders from a session, or a video skill installed.
TryRemotion
How it's checkedA personal context repo or folder that your agents load when relevant, through a rule or a skill.
How it's checkedComputer-use or desktop-control tools used: computer use, Peekaboo, Claude in Chrome, simulator control.
TryClaude in Chrome · Peekaboo
How it's checkedA QA or release checklist file, or a release skill.
QA_BEFORE_RELEASES.md
Salvatore Sanfilippo · source
How it's checkedAn error-tracking or alerting connector (Sentry, Datadog, PagerDuty, Grafana) or alerts sent into an agent session.
TrySentry MCP
How it's checkedA repo you're committing to again after a year or more without a commit.
“It is no exaggeration to say that this was brought to us by Claude.”
Bryan Cantrill · source
How it's checkedAn agent SDK in a project's dependencies (Claude Agent SDK, Codex SDK, OpenAI Agents).
Full circle
Fewer, sharper tools, and you understand them all the way down, to the metal.
How it's checkedYou install skills, plugins and MCP servers only from sources you trust, or read the code before you do. Usually only you can tell, so you tick it.
“Do an audit of the (open-source) agent code base first.”
Sebastian Raschka · source
How it's checkedA rule or hook that makes it stop and report when stuck or out of scope.
“after 8 tool-call rounds with no text response of 100+ chars, it injects a reminder to surface and reassess”
Jeremy Howard · source
How it's checkedCommits or PRs in your repos by a bot account you run, or a separate login the agent uses.
“Agents on behave of @antfu”
Anthony Fu · source
How it's checkedDevice or CAD tools driven by an agent (serial, adb, simctl, esptool, arduino-cli, OpenSCAD, CadQuery, build123d), or .stl, .3mf or .scad files it produced.
“I started the day with "I want to connect two pedals together and have them validate each other", and ended the day with three pedals, a signal generator, and a USB MIDI dongle all connected together as a validation bench with automated checks.”
Linus Torvalds · source
How it's checkedPi, OpenClaw, Hermes or a harness you wrote, installed and in use.
“I use OpenClaw (which uses Codex) and web sessions.”
Peter Steinberger · source
How it's checkedA recurring agent run that reviews past sessions and edits rules or skills, or /insights used to change your setup.
Schedule a review of recent chats to find friction, then fix prompts and setup.
OpenAI, Codex best practices, paraphrased · source
How it's checkedA project of yours that calls a model API with tools in a loop you wrote, not only an SDK or someone else's harness.
TryHow to build an agent Thorsten Ball's walkthrough
“It’s an LLM, a loop, and enough tokens.”
Thorsten Ball · source
Fossil record
The list is always 100. When a new level earns its place, an old one goes extinct and moves here.
Role-play preambles. Today's models don't need the pep talk; they need the context.
It sort of worked once. Nobody ever paid.
Reasoning models think before answering anyway.
From the era of lazy models and "// rest of the code stays the same".
One file in the repo root. Replaced by rule folders and AGENTS.md, which every agent reads.
Deprecated in favour of skills.
Thinking keywords. Effort is a setting now, and reports disagree on whether the word still does anything.
Retired as a benchmark. Still a good party trick.
More levels are coming
Agents change monthly, so the list will too. Leave your email and you'll hear from me each time 20 new levels are in, and only if they're worth your time. Nothing else.