Skip to content
Agentmart
Submit a tool

suede-parity-contract

Suede AI cross-surface canon discipline: hold one canonical answer across every surface that states it (web, iOS, Android, docs, a second service) by…

Skill by JasonColapietro

IndexedMIT★ 125
Write a review

Chalk Labs

Need help deploying suede-parity-contract?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/jasoncolapietro-suede-parity-contract

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Suede AI cross-surface canon discipline: hold one canonical answer across every surface that states it (web, iOS, Android, docs, a second service) by generating a contract from the reference surface and making the others assert against it, so a divergence fails a test instead of reaching a user or an answer engine. Consistent canon is what makes a product citable: generative search quotes sources that do not contradict themselves, and a number that differs between your site and your app is a contradiction a model can see. Use when the same rule, threshold, ladder, or stated fact lives on two or more surfaces; when values are copied by hand out of a handoff; when two surfaces already disagree; or when auditing where they drifted. NOT FOR: deciding which surface is correct (a product call this skill records rather than makes); reviewing a diff (use suede-code); wiring merge gates (use suede-ci-gate); a generative-search audit of a page (use suede-seo-audit).

Use cases: documents, testing, devops, security

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/suede-parity-contract

What's inside

Skill name
suede-parity-contract
Description
Suede AI cross-surface canon discipline: hold one canonical answer across every surface that states it (web, iOS, Android, docs, a second service) by generating a contract from the reference surface and making the others assert against it, so a divergence fails a test instead of reaching a user or an answer engine. Consistent canon is what makes a product citable: generative search quotes sources that do not contradict themselves, and a number that differs between your site and your app is a contradiction a model can see. Use when the same rule, threshold, ladder, or stated fact lives on two or more surfaces; when values are copied by hand out of a handoff; when two surfaces already disagree; or when auditing where they drifted. NOT FOR: deciding which surface is correct (a product call this skill records rather than makes); reviewing a diff (use suede-code); wiring merge gates (use suede-ci-gate); a generative-search audit of a page (use suede-seo-audit).
License
MIT
Instructions
172 lines in SKILL.md · view on GitHub

Files (3)

  • CARD.md
  • SKILL.md
  • openai.yaml · agents/

SKILL.md (first lines)

# Suede AI Parity Contract Skills

## Compatibility

Requires source access to one reference and each follower plus their build/test runtimes. Contract generation must import live constants; cross-repository updates require write authority to each repository.

For a concrete input, expected output, recovery path, and validation check, read
[the worked contract](references/worked-contract.md). Use it to check the handoff
before treating the task as complete.

```text
Iron Law: generate the contract from the reference surface's live constants.
A contract you transcribe is a second copy, and second copies drift.
A contract you generate cannot disagree with the code it came from.
```

Silent drift is the failure this prevents. Nothing crashes when a threshold is
copied wrong into a second language. The two surfaces simply start answering the
same question differently, and a user finds out before you do.

## Step 1: Name one reference surface

Exactly one surface is the reference. Every other surface is a follower that
asserts against it. Two references is two sources of truth with extra steps.

Pick the surface where the domain is most complete and most exercised, and write
that choice into the contract's `reference` field so nobody re-litigates it.

## Step 2: Build the contract by importing, never retyping

The builder imports the constants from the modules the application actually runs
on and serializes them. Retyping a value into the builder reintroduces exactly
the copy this skill exists to delete.

When a needed value is private, export it rather than duplicating it. That is a
one-line change and it keeps the count of definitions at one.

Serialize with stable key order and a trailing newline, because followers vendor
the file verbatim and a reformat shows up as a spurious diff.

## Step 3: Make staleness fail on the reference side

One test rebuilds the contract in memory and compares it to the committed file.
Without it the JSON is a snapshot someone took once.

Give it a write mode so regeneration is one command, and put that command in the
contract's README and in the file's own `note` field:

```bash
CONTRACT_WRITE=1 <test runner> <contract test>
```

Compare parsed objects first so the failure names the key that moved, then
compare raw bytes so a reformat is caught too.

## Step 4: Vendor byte-identical to each follower

Followers keep a copy at the same relative path. Prove it matches rather than
assuming:

```bash
shasum -a 256 <reference>/contracts/<name>.json <follower>/contracts/<name>.json
```

Two identical hashes, or the copy is stale. Record the re-sync command in the
follower's README so the next person does not invent one.

## Step 5: Assert through public behavior, not private internals

A follower test that reimplements the reference formula proves only that you can
write the same bug twice.

Assert what a user experiences. For a ladder, feed the contract's rung values to
the public accessor and check the level it returns. For a threshold, check the
boundary **and one step below it**: a threshold asserted only from above still
passes after it moves down.

Start the follower suite with a test that the contract loaded and is non-empty.
A file that fails to load makes every assertion after it vacuously true.

## Step 6: Pin the divergences you already have

You will find existing differences. The contract's job on day one is to make
them visible and hold them still, not to erase them.

Record each under `knownDivergences` with the value on each surface and a
`reason` that survives you leaving. Then add a guard asserting the divergence
list is **exactly** the expected set, so an entry added elsewhere is a failure
rather than a silence.

Each pinned divergence gets two assertions on the follower: that it still
differs by the recorded amount, and that it does **not** equal the reference. The
second one fires when someone closes the divergence and forgets the contract.

### Halt format: a divergence found mid-build

Changing a shipped constant changes behavior for real users, and which surface is
right is a product decision. At the moment you find one, stop and report:

```text
Divergence: <key>. <reference surface> says <a>, <follower> says <b>.
Both shipped. <one line on what a user feels>.
  1. Move <follower> to <a>
  2. Move <reference> to <b>
  3. Record it and decide later
```

Then wait. Do not pick for the user, and do not change either side as a side
effect of writing a contract.

## Step 7: Prove non-vacuity on both sides, in this run

A parity test that cannot fail is worse than none, because it reads as evidence.

Reference side: change one constant, watch the staleness test fail **naming that
key**, restore it, watch it pass. Follower side: edit one value in the vendored
copy, watch exactly the matching assertion fail, restore.

Paste both the failing and the passing output. "The tests pass" is not proof

From JasonColapietro/suede-creator-skills, MIT.

More from JasonColapietro/suede-creator-skills

Signals

Stars
125
Forks
12
Open issues
4
Contributors
3
Last commit (this folder)
Oct 24, 2026 (7 days ago)
Last release
Aug 4, 2026 (2 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from JasonColapietro

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get suede-parity-contract running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying suede-parity-contract?

CallGet help
Cite this page

suede-parity-contract skill. Agent Mart. https://agentmart.dev/skills/jasoncolapietro-suede-parity-contract (last updated 2026-10-24)