level-up
Skill3 starsChanged 5 months agoMITView on GitHub
Use weekly to find and ship one new automation. Walks the 3Ms interview — Mindset (find the candidate) → Method (scope one) → Machine (build it). Trigger on "let's level up", "what should I automate next", "find me leverage this week", or as a Friday ritual. One run = one shipped artifact, or a logged decision to stop.
18 other projects use a version of this, 8 of them edited.
Adopted by
Most recently taken up by
- vikneshwaran16032005-alt/claude-skillsedited, first seen Nov 1, 2026
- yowpi-tech/unbubbletheir own version, first seen Nov 1, 2026
- kristian-presso/claude-operator-skillstheir own version, first seen Oct 20, 2026
- emiljuchnikowski/ai-os-templatetheir own version, first seen Oct 20, 2026
- holt-os/holttheir own version, first seen Oct 17, 2026
- sameerch1b/asi-osedited, first seen Oct 9, 2026
Used with
Projects with it also use explain and build.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
sunnyclinches44Skillexplain
Explains code using real-world analogies and ASCII diagrams
Also in 3 projects that use this
Microsoft AzureSkillbuild
Build the TypeSpec Rust emitter from source. Run this after cloning the repo, after pulling changes, or when TypeScript source files have been modified.
Also in 3 projects that use this
Other versions18 projects use a version of this.
Word for word · 0
None found.
Edited · 8
- nateherkai/AIS-OS★ 1.4k63% same
- alirezarezvani/gaios★ 5081% same
- vikneshwaran16032005-alt/claude-skills★ 362% same
- boothy1021-bot/aios-microsoft★ 171% same
- ced4568/Synthos-AIOS★ 181% same
- sameerch1b/asi-os★ 168% same
- berch-t/Kopern-OS★ 072% same
- GerardoRdz96/claude-skills★ 059% same
Wrote their own version · 10
- holt-os/holt★ 7same name
- daveebbelaar/project-template★ 4same name
- yowpi-tech/unbubble★ 4same name
- sohaibdevv/project-template★ 2same name
- cacardinal/laila★ 1same name
- Nythbat/zpd-learning★ 1same name
- SkillPilotCH/skillpilot-aios-starter★ 1same name
- emiljuchnikowski/ai-os-template★ 0same name
- kristian-presso/claude-operator-skills★ 0same name
- nsbradley88/level-up★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository15 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-builderSkill
- agents-team-builderSkill
- aios-auditSkill
- grill-meSkill
- hooks-builderSkill
- multi-brainSkill
- onboardSkill
- plugin-builderSkill
- printing-pressSkill
- routines-builderSkill
- session-handoffSkill
- skill-builderSkill
- workflow-builderSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 8 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 16 edited versions we read keep these.
8
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
Not a multi-candidate planner.
- Distinctive: line 18
Not a coach. The user does the thinking.
- Very specific: line 66
Delegate third: if too complex/variable/judgment-heavy → suggest a person.
- Distinctive: line 75
If the user can't articulate any of the five: *"If you can't explain it to a person, you can't explain it to an AI.
- Distinctive: line 104
Ask: *"How do you want to ship this?"* Options ordered by Boring-is-Beautiful default:
- Very specific: line 137
Every /level-up run produces:
- Distinctive: line 147
EAD enforces "eliminate first.": If the answer is Eliminate, exit cheerfully — that's a win, not a failure.
- Distinctive: line 149
Boring-is-Beautiful default in Machine handoff: Default = highest non-AI option.
- Distinctive: line 150
Tie-to-KPI is mandatory: If the user can't name bucket + metric, skill stops.
- Distinctive: line 153
Attribution on output: Every report and every scaffolded artifact references the framework.
- Very specific: line 158
Fresh-clone test: Run on a brand-new kit with an empty decisions/log.md and no audits/.
“Common” is measured: kept by at least 6 in 10 of the 16 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
The Three Ms of AI is a framework by Nate Herk. See
references/3ms-framework.md.
What this skill does
Walks the user through the 3Ms each week to surface and ship one new automation. One interview = one outcome: one shipped artifact, or one logged decision to eliminate, delegate, or stop. It also installs the 3Ms framework into the user's head over time — after 4-6 runs, the user starts spotting opportunities mid-week without prompting because the questions have become internal defaults.
This is the brain-rewire mechanism. The kit doesn't need cron jobs to anchor behavior; it needs /level-up running every Friday.
What /level-up is NOT
- Not
/aios-audit./aios-auditis structural ("is the AIOS built right?")./level-upis functional ("what leverage am I missing?"). Run/aios-auditfirst if structure is messy. - Not a multi-candidate planner. One run scopes one candidate, ending in one artifact or one logged early-exit decision.
- Not a coach. The user does the thinking. The skill conducts the interview.
When /level-up runs
- First run: after setup is real. Once the user has connected ≥1 MCP/script and run
/aios-auditonce. Earlier yields trivial output. - Cadence: weekly, Friday afternoon. Review the week, surface one automation, ship Monday.
- On-demand any time. Mid-week if a manual task itches.
Inputs the skill reads
context/priorities.md— what the user said matterscontext/about-me.md— top_pain, roleconnections.md— what's reachable, by what mechanismreferences/3ms-framework.md— the framework (used to quote principles back)decisions/log.md— recent decisions (what's already shipped or considered).claude/skills/*/SKILL.mdfrontmatter — what capabilities exist- Recent
audits/audit-{date}.mdif present
Execution — three phases
Phase 1 — Mindset interview (find the candidate)
Fresh clone, no history? If decisions/log.md is empty and there's no audits/ yet (a brand-new kit), don't stall. Skip the "what did you do in the repo this week" mining and run the interview straight from what the user tells you, plus top_pain in context/about-me.md and context/priorities.md. Suggest running /aios-audit once so future runs have signal, but you can still ship one small automation today.
Surface 1-3 candidates ranked by leverage. Ask these in order, conversationally:
- "Walk me through your week. What did you do 3+ times?" (frequency)
- "Anything that felt manual, boring, or copy-paste?" (drudgery)
- "Anything where you thought 'a smart intern could handle this'?" (delegation)
- "If your workload doubled tomorrow, what would break first?" (constraint)
- "What would let you take on twice the work without breaking?" (growth lever)
Quote relevant Mindset principles when they fit:
- "Sounds like the Default Shift applies — to what extent could AI be leveraged here?"
- "This is the Function Breakdown — you're not automating the whole job, just this one piece."
- "AI is better than you think and improving faster than you think. If it couldn't do this last quarter, it might be ready now."
Output of Phase 1: numbered list of 1-3 candidate opportunities, one-line "why this is leverage" per candidate. Ask: "Pick one to scope."
Phase 2 — Method interview (scope one)
User picks one candidate. Walk the Method pipeline:
Step 1 — Find the constraint. Which bottleneck does this solve, or which growth lever does it open? Tie back to Phase 1 answers.
Step 2 — EAD: Eliminate / Automate / Delegate.
- Eliminate first: "What happens if we just stop doing this?" If the answer is "nothing breaks" → skill exits cheerfully. "Don't automate waste." This is a win, log to
decisions/log.mdand stop. - Automate second: apply 60/30/10 framing. ~60% deterministic, ~30% AI-assisted, ~10% manual.
- Delegate third: if too complex/variable/judgment-heavy → suggest a person. Skill exits with a delegation suggestion, log it.
Step 3 — Map the process. Five elements:
- Trigger (what kicks it off)
- Data sources (where info comes from)
- Data transformations (how data changes shape)
- Decision points (where it branches)
- Destination (where output goes)
If the user can't articulate any of the five: "If you can't explain it to a person, you can't explain it to an AI. Sketch it on paper first, then come back." Skill stops.
Step 4 — Pick the autonomy level.
| Level | Name | What happens |
| L0 | Manual | No AI |
| L1 | Suggested | AI suggests, human decides every step |
| L2 | Drafted | AI drafts, human reviews and edits |
| L3 | Supervised | AI runs, human validates periodically |
| L4 | Autonomous | AI handles end-to-end |
Default = lowest level that solves the problem. Push back on L4 unless the user has explicitly run lower levels first. "Workflows beat agents. If a decision doesn't HAVE to be made by AI, don't let AI make it."
Side-effect gate (overrides the autonomy level). Before locking the level, ask: does any step send, delete, publish, pay, or deploy? If yes, that step is a mandatory human-gate regardless of the autonomy level chosen — the AI may prepare the action, a human approves the irreversible part. This carries into the scaffolded artifact. (Loop-engineering doctrine: references/agent-loops.md.)
Step 5 — Tie to a KPI. Which of the Three Buckets does this move?
- More output / more customers
- More value per unit of work
- Less cost
Plus a specific metric (response time, error rate, conversion rate, time-to-completion). If the user can't name a bucket and a metric, skill stops. "If your automation doesn't move a number, why are you building it?"
Step 6 — Define "done = right." How will the user know the output is correct? Pick the cheapest check that fits: a boolean test (pass/fail), a quick look, a rubric, or your eyes. Default to a functional test or a manual check. For content-quality judgments, route the check to a different model lineage — don't let the model that produced the work grade its own work (references/agent-loops.md). This line carries into the scaffolded artifact.
Output of Phase 2: scoped automation spec written to decisions/log.md as a dated entry with the Method answers + autonomy level + KPI + the "done = right" check. Durable record of what was decided and why.
Phase 3 — Machine handoff (build it)
Ask: "How do you want to ship this?" Options ordered by Boring-is-Beautiful default:
- Prompt-only — saved prompt template the user runs by hand. Zero infrastructure. Highest manual involvement.
- Deterministic skill — SKILL.md that runs a script (no AI step). Best for transformations with clear rules.
- AI-assisted skill — SKILL.md with one AI call inside. Drafts, classifies, summarizes.
- Sub-agent — multi-step agent. Last resort. Only if the work genuinely needs reasoning + tool use.
Default selected = highest non-AI option that solves the problem. User has to explicitly choose more autonomy.
Once chosen, route to the appropriate scaffolder:
skill-creatorif available globally (Anthropic-shipped)skill-builderif the user has it locally (ships with this kit)- For a sub-agent, route to
agent-builder - Otherwise write a SKILL.md / agent file inline with frontmatter, location, and contents
Every scaffolded artifact keeps standard frontmatter (name + description, only the fields skill-builder allows — don't add fields just because you can) and carries a short note block in the body, right under the frontmatter:
> **Bike Method — Phase 1 (training wheels).** Run this by hand first; advance only after you've validated it manually.
> **Side-effect gate:** any step that sends / deletes / publishes / pays / deploys stays human-gated, whatever the autonomy level.
> **Done = right when:** <boolean test / a look / a rubric / your eyes>. Route content-quality checks to a second model (different lineage).
> *The Three Ms of AI is a framework by Nate Herk.*This is a convention, not enforcement — nothing in the kit reads a bike-method-phase field, so the reminder lives in the body where the user actually sees it, instead of an invented frontmatter key that breaks the skill-builder rule. It nudges the user to validate manually first and to keep irreversible steps gated; it doesn't mechanically block skipping. The loop-engineering doctrine behind training-wheels phases, side-effect gates, and verification lives in references/agent-loops.md.
Surface the Machine principles when scaffolding:
- Lego Principle — smallest steps, zero-AI first if possible
- Validation Chain — test each step before chaining
- Iteration Mindset — ship the POC, expand from real usage
Output contract
Every /level-up run produces:
- One
decisions/log.mdentry — dated, with the Method spec - Either one scaffolded artifact (prompt, skill, or agent file) or one logged early-exit decision — eliminate, delegate, or stop. Early exits are wins, not failures, and still get logged.
- A one-screen close — what was scoped, and either what was built (with the Bike Method Phase 1 reminder) or why the run stopped early.
Critical implementation rules
- One interview = one outcome (one artifact, or one logged early-exit decision). No multi-candidate parallel scoping.
- Mindset phase always runs first. Even if the user comes in with a pre-formed idea.
- EAD enforces "eliminate first." If the answer is Eliminate, exit cheerfully — that's a win, not a failure.
- Default to the lowest autonomy level that works. Push back on L4.
- Boring-is-Beautiful default in Machine handoff. Default = highest non-AI option.
- Tie-to-KPI is mandatory. If the user can't name bucket + metric, skill stops.
- Bike Method ships into every artifact as a body note (a convention, not a frontmatter field — nothing reads it).
- Read-only on user files except
decisions/log.mdand the new artifact. Don't modify other existing files. - Attribution on output. Every report and every scaffolded artifact references the framework.
Verification (for the implementer)
- Dry run on a populated profile with no prompt. Expected: skill surfaces 2-3 candidates pulled from recent activity, priorities, and top_pain. Generic output ("you should build a brief") = fail.
- Fresh-clone test. Run on a brand-new kit with an empty
decisions/log.mdand noaudits/. Expected: skill doesn't stall or error on missing history — it interviews fromtop_pain+ priorities + what the user says, and suggests/aios-auditfor next time. - Eliminate-first test. Feed an obviously eliminate-able candidate. Expected: skill suggests Eliminate, exits, logs the win.
- L4 push-back test. User asks for an autonomous email-replier on first build. Expected: skill insists on L1/L2 first, won't ship L4 without explicit override.
- Boring-is-Beautiful test. Candidate solvable with deterministic Python. Expected: skill recommends
(2) deterministic skillas default. - Bike Method anti-skip. User scaffolds, asks to advance to Phase 4 immediately. Expected: skill conversationally walks them through what each phase means and asks them to confirm they've validated the lower phases (a nudge — the body note is a convention, not a hard lock).
The Three Ms of AI is a framework by Nate Herk. See
references/3ms-framework.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

