agentleFS
TomSzenessyTomSzenessy/AI_TEMPLATE/.agents/skills/product-recon/SKILL.md

product-recon

Skill0 starsChanged 11 days agoMITView on GitHub

Reverse-engineer a reference product into a recon map: screens, flows, components, inferred data model, feature matrix — from public pages, screenshots, listings, help docs and the user's own account. Use when the user says "clone this app", "reverse engineer X", "map out this product", "how does X work", "build our own version of X", or "competitive analysis".

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Public sources and the user's own account only: Never someone else's account, never a password, never past a paywall or a login by any …

    line 35
  2. Distinctive:

    Reading, not scraping: No crawlers, no bulk downloads, no loops.

    line 37
  3. Distinctive:

    No source code, no private APIs: No decompiling, no reading its bundles, no logging network calls to copy endpoints.

    line 39
  4. Distinctive:

    Ask three things, or propose answers and get a yes:

    line 48
  5. Distinctive:

    Every repeated UI part: buttons, inputs, date pickers, modals, tables, toasts, nav.

    line 87

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

product-recon

Everything downstream — the design tokens, the build, the parity check — builds from what this writes. A bad recon map means a bad build.

reference/recon.md        the recon map (template: recon-map.md in this folder)
reference/screens/        reference product's screenshots. Never shipped.
docs/product/features.csv the feature matrix, shared with product-kickoff,
                          review-mining, parity-check and `make next`

reference/ holds recon evidence only (screens, notes, the map); the matrix lives in docs/product/features.csv so make next sees it.

Once per project, add "reference" to infrastructure_paths in project.toml (or declare it a surface): make check flags unregistered top-level dirs. The other product skills write under reference/ too and rely on this step.

The rules, before anything else

This skill studies functionality and UX patterns, clean-room style: what the product does and how a user moves through it, taking nothing it owns.

  • Public sources and the user's own account only. Never someone else's account, never a password, never past a paywall or a login by any trick.
  • Reading, not scraping. No crawlers, no bulk downloads, no loops. With a browser tool on the user's own browser, read at human speed, user present.
  • No source code, no private APIs. No decompiling, no reading its bundles, no logging network calls to copy endpoints; public API docs are fine.
  • Check the terms. Some terms forbid using an account to build a competing product; under those, say so and work from public sources only.
  • Screenshots are reference. They live in reference/screens/, are used to compare layouts, and never go into the build.

Step 1: scope

Ask three things, or propose answers and get a yes:

  1. Which product, which platform. Web, iOS, Android, desktop.
  2. Which slice. "All of Notion" is not a project; "Notion's pages, blocks and sharing" is. Default to the core loop: the flow users pay for.
  3. What it is for. The user's own business, a niche, a competitive read, a product to sell.

Step 2: list the sources

Build a sources table first, a URL on every row, in order of value:

sourcewhat it gives you
help center / docsthe most complete feature list there is, and the settings
pricing pagewhich features matter (they gate them)
changelogwhat was added recently, what the team thinks is important
app store listingscreenshots of every key screen, the pitch, ratings
public walkthrough videosreal flows, click by click
marketing sitepositioning, the core loop in their words
the user's own accountthe real thing, every state, driven by the user
public API docsthe data model, almost for free

Step 3: screen inventory

One row per screen, in the Screens table of recon-map.md. IDs are stable: S01, S02... Every other file refers to them.

States matter: empty, loading, filled, error, permission denied, mobile. An empty state you did not record is an empty state you will not build.

Step 4: user flows

F01, F02... Each one is a goal and the screens it passes through (the Flows block of recon-map.md). Count the happy-path clicks — the number to beat — and the edge cases: no availability, time zones, the slot taken mid-form.

Step 5: components

Every repeated UI part: buttons, inputs, date pickers, modals, tables, toasts, nav. Name, variants, states, which screens use it.

Step 6: inferred data model

Entities, fields and relationships, each with its evidence (screen IDs, help articles) and a confidence: high, medium, guess. Mark guesses as guesses. The build turns this into a real schema.

Step 7: feature matrix

Append one row per feature the reference product has to docs/product/features.csv (columns: feature, area, priority, status, evidence, acceptance, source; template: features.csv in this folder; create the file with its header if kickoff has not). Priority is must / should / could. status is does this build have it, starting at no and filled in during the build; source is product-recon. A must row needs an acceptance criterion. Quote any value that contains a comma. parity-check and make next both read this file.

Step 8: what cannot be cloned

List it honestly, as status=skip rows with the reason in acceptance: licensed content (a music catalogue, a stock library), the network and its users, data the product owns, partner deals, hardware, regulated licences (banking, health). Learning from a product means the features and the flow, not what it owns.

Step 9: size it

Screens, flows, entities, and the hard parts (realtime, sync, payments, calendar or email integrations, offline). Give a size: S (a weekend), M (a few weeks), L (a quarter), XL (rescope it). No promises of a perfect clone.

Output

reference/recon.md and the new rows in docs/product/features.csv, then a five-line summary: the core loop, screen and flow counts, the three hardest parts, what is out of scope, and the next step: design-tokens, then the project's issue-backed build loop (AGENTS.md).

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.