mimic-build
Skill1 starsChanged 22 days agoMITView on GitHub
Builds your better version slice by slice from the architecture and the recon, clean room: every screen with all its states, or for a service or local business every step the customer goes through (order page, forms, messages, checklists). Writes every word fresh, uses open or owned assets, builds the review fixes into the slice they belong to, and keeps features.csv honest. Use when the user says "build it", "start building", "build the first slice", "make the booking page", "code the core flow", "build the order page", "set up the customer steps", "next slice", or after /mimic-architect or /replica-design. Refuses to copy a competitor's code, text, images or logo.
In this repository12 other files it gives its agents.
- competitorsSkill
- launch-videoSkill
- market-researchSkill
- mimic-architectSkill
- mimic-backendSkill
- mimic-deploySkill
- mimic-launchSkill
- mimic-reconSkill
- mimic-reviewsSkill
- mimic-runSkill
- mimic-testSkill
- validateSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. About 0 are common, 10 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 41
Copy what it does, never what it owns: No code, HTML, CSS, scripts, images, icons, logos, fonts, photos, menu text, or marketing copy from …
- Very specific: line 51
Licensed content and networks are out: If a feature depends on what the competitor licenses or owns (a catalogue, their users, partner …
- Distinctive: line 58
No design tokens? Use a plain neutral theme with clear roles (background, surface, text, muted text, accent, danger) so replica-brand can …
- Distinctive: line 66
every state each one needs: empty, loading, filled, error, no permission, small screen
- Distinctive: line 86
Run the "done when" check yourself where you can: start the app, run the type checker or linter, hit the route.
- Distinctive: line 90
Look for the states you listed.
- Very specific: line 95
Update mimic/features.csv: yes only when it runs, partial with a note on what is missing, no otherwise.
- Very specific: line 98
Add an entry to mimic/build-log.md:
- Distinctive: line 122
Use the tools picked in the architecture (a booking tool, a form, a sheet).
- Distinctive: line 132
Every choice you make that is not in the inputs (a time window, a default, a price) is written as "assumed" in the build log, so the user …
- Very specific: line 135
Never mark a feature yes that does not run.
- Very specific: line 136
Never paste a competitor's text, even "for now".
- Distinctive: line 137
Keep secrets out of the code.
- Distinctive: line 145
If a slice is too big for one pass, say so up front, finish part of it well, and log the rest.
- Distinctive: line 149
Every file and reply is read by a busy founder.
- Distinctive: line 152
If the humanize skill is installed, run copy that a person will publish or send through it.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
mimic-build
The aim is a working product the user can click through after every slice, not a big reveal at the end. Small, checked steps catch wrong turns while they are cheap.
Reads and writes
Reads, in the user's project:
mimic/architecture.md(stack, schema, build order with "done when" checks)mimic/recon.md(screens or customer steps, flows, states seen)mimic/features.csv(what to build and its priority)mimic/fixes.md(the fixes that make it better)mimic/design/(tokens and component specs, if replica-design ran)- the business context (founder-context) for tone and who it serves
Writes:
- code or assets in the project
mimic/features.csv, theclonecolumn, filled honestlymimic/build-log.md, one entry per slice (template below)
The clean room
The product does the same jobs as the competitor's, in its own way. That protects the user from takedowns and gives customers a reason to switch.
- Copy what it does, never what it owns. No code, HTML, CSS, scripts, images, icons,
logos, fonts, photos, menu text, or marketing copy from a competitor. Screenshots in
mimic/screens/are for looking at, never for tracing or shipping. - If asked to copy any of it, say no to that part and build an own version of the same function. Explain in one line why: it is their property, and it is easy to spot.
- Write every word fresh: labels, buttons, empty states, errors, emails. Use the brand
voice if
mimic/brand.mdexists, else the tone in the business context. - Assets: an open icon set (Lucide, Phosphor, Heroicons, Tabler), open fonts (Inter,
Source Sans, IBM Plex and similar), and the user's own photos or placeholders marked
TODO: own photo. - Licensed content and networks are out. If a feature depends on what the competitor
licenses or owns (a catalogue, their users, partner deals), mark it
skipwith a reason.
Step 0: check the ground
- No
mimic/architecture.md? Build only the core loop as a first slice, write down the stack you assumed at the top of the build log, and suggest/mimic-architectnext. - No design tokens? Use a plain neutral theme with clear roles (background, surface, text, muted text, accent, danger) so replica-brand can swap the values later.
Step 1: one slice at a time
Take the next slice from the build order. Before writing code, list:
- the screens or steps it covers, from
recon.md - every state each one needs: empty, loading, filled, error, no permission, small screen
- the fixes from
fixes.mdthat belong in this slice - the "done when" check from the architecture
Then build it. Keep the project runnable after each slice. If it is a new project, set up the skeleton first with the stack's standard starter, and say the commands.
Build the fixes inside the slice they belong to. A reminder fix belongs with booking, not in a "later" pile. Fixes are why this product exists.
Build the whole slice, not its easy half. Every screen and state the architecture and recon list for it gets built, including the sign-in and "connect" steps those screens depend on. When a step needs an outside service (OAuth sign-in, a calendar connect), build the working flow through the official SDK with keys from env vars. mimic-backend then checks and hardens it: scopes, webhooks, token storage, the security list. Why: a button that does nothing is a dead end the user clicks on day one, and the slice is not done. Defer a piece only when it truly cannot work yet, and then say so in the reply and the log.
Step 2: check the slice
Run the "done when" check yourself where you can: start the app, run the type checker or linter, hit the route. If you cannot run it here, say so and give the user the exact steps to check it, with what they should see.
Look for the states you listed. An empty state you did not build is a blank screen a real customer will see on day one.
Step 3: record it honestly
Update mimic/features.csv: yes only when it runs, partial with a note on what is
missing, no otherwise. The parity score later depends on this being true.
Add an entry to mimic/build-log.md:
## Slice 1: Core loop, guest books a slot (YYYY-MM-DD)
built: S07 booking page, S08 pick a time, S09 details, S10 confirmed
states: empty (no slots), loading, filled, error (slot taken), mobile
fixes in this slice: time zone confirm step (fixes.md #3)
how to run: npm run dev, open /book/demo
done when: passed (second booking for the same slot refused with a clear message)
left: reschedule link (next slice)Services, local businesses and physical products
The slices are the steps of the customer path in recon.md (C01, C02...). Build what each
step needs, with every "state" of that step: sold out, late, wrong order, refund.
- Online parts: a landing or order page, an order or booking form, confirmation and reminder messages, a small admin view or a shared sheet for incoming orders.
- Offline parts: a written checklist or SOP for each step (prep, pickup, delivery, handover), the message templates staff send, a partner or supplier sign-up form.
- The "done when" check is a dry run: place a test order, follow the checklist, receive the messages.
Use the tools picked in the architecture (a booking tool, a form, a sheet). Do not build custom software where a tool does the job.
Where files go
Code goes where the stack puts it. Plans, specs, checklists and message templates for a
service or local business go in mimic/build/ so the other skills can find them.
Rules
- Every choice you make that is not in the inputs (a time window, a default, a price) is written as "assumed" in the build log, so the user can change it.
- Never mark a feature
yesthat does not run. - Never paste a competitor's text, even "for now". Placeholders are fine:
{{headline}}. - Keep secrets out of the code. Use env vars; mimic-backend sets them up properly.
- Lock each user's data to that user from the first slice: access rules on every table (row-level security policies, or the stack's own version), and requests that run as the signed-in user. Keep the service role or admin key for server jobs only. Keep OAuth tokens where the browser cannot read them. Why: an open table leaks every customer at once, and it is much harder to add later.
- A save that replaces the user's data must not lose it if it fails halfway. Write the new rows before removing the old ones, or do both in one transaction or upsert.
- If a slice is too big for one pass, say so up front, finish part of it well, and log the rest.
Writing
Every file and reply is read by a busy founder. Plain words, sentences under 25 words, no
em dashes, no hype. Skip these words in everything you write:
delve, leverage, seamless, streamline, comprehensive, journey, unlock, elevate, empower, foster, robust, crucial, utilize, simple. <!-- allow-words -->
If the humanize skill is installed, run copy that a person will publish or send through it.
Output
The slice built, features.csv and build-log.md updated, then a short summary: what
works now, how to see it, what is next.
End the reply with: "Next: the next slice (name it), then /mimic-backend, then
/mimic-test." Run /mimic-run to see where you are.
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.

