Set it up · Why it's different · How it works · Networks · Grading · Showcase · Docs

Warning

Alpha — expect bugs and breaking changes. standdown is pre-1.0 and under active development: the API may shift between minor versions and edge cases are still being found. Because it makes revenue-affecting decisions, pin your version, verify it against your own integration with the conformance grader, and please report anything that misbehaves.

Your extension shouldn't steal the sale. standdown detects existing affiliate attribution, suppresses competing activation, and proves the decision was made locally — never on a server.

standdown is a zero-runtime-dependency TypeScript library for extension developers who need to detect existing affiliate attribution, suppress competing activation, and prove that suppression decisions were made locally and deterministically. Built and maintained by Dupe.

Set it up

Hand the whole integration to your coding agent — that's the supported path. standdown makes revenue-affecting decisions across your extension's code, so the fastest and safest way in is to let a coding agent do the work: it reads your codebase, gates every place attribution fires, bundles, and grades the result. Paste this into your agent — Claude Code, Codex, Cursor, opencode, Copilot — pointed at your extension's repo:

Set up standdown — the open-source affiliate stand-down library (npm package
`standdown`, source https://github.com/dupe-com/standdown) — in this browser
extension, so it stops hijacking affiliate attribution a partner already owns.
Then grade and verify it. Read and follow, end to end, the official guide at
https://raw.githubusercontent.com/dupe-com/standdown/main/AGENTS.md — it detects
whether this is a fresh install or a migration of existing stand-down logic and
branches accordingly. When you're done, report my conformance grade and keep
fixing until it's an A or better.

That single prompt runs the whole loop — install, adapter choice, gating, bundling, and grading — and auto-switches to the shadow-mode migration path if your extension already has its own stand-down logic. The steps live in one place, AGENTS.md, so every agent follows the same playbook.

Use a capable model. This reasons about revenue-affecting control flow across an unfamiliar codebase — run it on a frontier coding model (Claude Opus, GPT-5.5, or equivalent). Lighter models miss firing points and mis-handle the degraded gate; if you must use one, review the gating diff by hand and always run the grader.

On Claude Code? Install the plugin once — no checkout — and get typed commands:

/plugin marketplace add dupe-com/standdown
/plugin install standdown@dupe

Then run /standdown:setup in your extension's repo (or /standdown:adopt for a migration, /standdown:showcase to publish your grade). The plugin also auto-triggers when you just say "add standdown to my extension."

Prefer a single skill, without the plugin?

Curl the skill straight into your personal skills — the slash command matches the folder you install it into:

mkdir -p ~/.claude/skills/standdown && curl -fsSL \
  https://raw.githubusercontent.com/dupe-com/standdown/main/skills/setup/SKILL.md \
  -o ~/.claude/skills/standdown/SKILL.md

Restart Claude Code, then run /standdown. (Swap skills/setup for skills/adopt or skills/showcase to grab the others.) The skill pulls the live AGENTS.md playbook itself, so it stays current.

Wiring it by hand instead? The full manual walkthrough and complete API reference — the five published surfaces, self-exemption, per-host disable, signed refresh, the standdown/url helper, interop — live in INSTALL.md.

What makes it different

Affiliate stand-down is easy to claim and hard to prove. standdown is built so the guarantees are structural — enforced by the type system and the architecture, not by a promise in a blog post.

• 🔒 Decisions never leave the device. No network call participates in a stand-down decision, ever. The decision path is a pure function of local signals and bundled policies.
• 🛡️ User data can't leak into a decision — by construction. Signals is a closed type: identity, accounts, balances, email, and login state are structurally unable to enter it.
• ⚖️ Fails toward standing down. Ambiguity, storage errors, or a malformed policy all resolve to suppress. The library never hijacks a sale by accident.
• 🎯 Detects attribution the way networks actually set it. Landing params, redirect-chain hops, first-party cookie names (never values), and referrer/initiator classification — across eight verified network packs.
• 🅰️ It grades itself, F→A+, and hands you a shareable card when you pass — with an inert-code guard so "disciplined" can't be faked by shipping nothing.
• 📦 Zero runtime dependencies. Ships ESM + CJS + types. MV3, with a content-script path for Safari and reduced-permission contexts.

How it works

standdown answers one question, locally: does someone already own the affiliate attribution on this page, so my extension should stay quiet? It returns a Decision — it never redirects, blocks, or fires anything itself. You own the side effects; you gate them on decision.standDown.

For each navigation it inspects local signals only and matches them against bundled, per-network policy packs:

• Landing params — the query params networks stamp on a click-through (cjevent, irgwc, ranSiteID, awc, sscid, eBay's mkcid/campid, …).
• Redirect-chain hops — tracker domains a click passed through (LinkSynergy, prf.hn, awin1.com, the universal fingerprint set), when the adapter can see them.
• First-party cookie names — attribution cookies a prior click dropped (lsclick_mid, sscid, im_ref, …). Names only, never values.
• Referrer / initiator — whether the navigation came from your own site (a self-owned click) or a third party.

A small state machine turns those into a decision: if a competitor's attribution is present, it stands down for the policy's duration; if the signal is yours (a declared self-param), it doesn't stand you down against your own click. Every decision is deterministic, appended to a local audit log, and made without a network call.

Two integration shapes:

	When	Playbook
Greenfield	Your extension has no stand-down logic yet	Gate each affiliate firing point on decision.standDown. Follow AGENTS.md.
Brownfield	Your extension already has homegrown attribution/stand-down logic (params, cookies, a disable list)	Migrate that decision path onto standdown in shadow mode first — run both in parallel and prove they agree before cutover, so no live commission is risked. Follow ADOPTING.md.

Works across the major affiliate networks

Each bundled pack implements a network's stand-down expectations — the detection signals it sets, the suppression behavior it asks for, and how long it lasts — so you don't have to reverse-engineer them. Eight verified packs ship enabled by default: seven named networks, plus a universal redirect-fingerprint set.

The eighth verified pack is a universal set of publisher-contributed redirect fingerprints (piedotorg/standdown-domains). Two more experimental packs (Sovrn/Skimlinks, Partnerize) are inferred from domain knowledge and stay opt-in until you verify them. See the full pack table and citations in POLICIES.md.

Network names and logos identify the stand-down policies each pack implements. They don't imply endorsement, partnership, or certification by these networks.

How it's graded

Unit tests prove the library's decisions in isolation. The audit/ harness proves the thing they can't: whether a real integration actually stands down instead of hijacking existing attribution. Two graders:

• conformanceGrade (start here) — deterministic and browser-free. It drives the policy set your extension bundles through the real decision engine over every network's attribution and control scenarios and scores F→A+. This is the number to report, and the correct sensor for any real host extension.

  bun install && bun run build   # build the lib the grader imports (dist/ is gitignored)
  cd audit && npm install
  DISABLE_HOSTS="ebay.com,homedepot.com" npx tsx grade/conformance.ts
  #   standdown conformance grade: A+  (100/100)

  The bare command grades the bundled allPolicies; if you ship a custom or subset set, add POLICY_PACK=/abs/path/to/your/policies.ts so it grades what you actually bundle.

• grade.ts (in-browser) — loads an unpacked extension into a real browser and watches for the testexts' /aff/:net?actor= redirect. A real host extension that activates by painting UI usually scores C (inert) here — that means "wrong sensor," not "dead code." Adapt grade/host-extension-probe.ts to sense your extension's own surface.

The rubric has an inert cap: an extension that never activates even when activation is allowed can't score above a C, so "disciplined stand-down" can't be faked with dead code. The harness is opt-in — not part of the npm package, not on the required CI path. See audit/README.md.

Prove it, and show it off

On a passing grade, both graders print a shareable card — a terminal card, a copy-paste social snippet, and a self-contained standdown-grade.svg written to your working directory that you can post anywhere:

Graded A/A+? Add your extension to the showcase — a wall of fame where every badge is reproduced by CI (a submission declares its policy inputs; CI re-runs the grader and regenerates the card, so nothing can be faked). The badge is tiered: a config-verified entry earns an A, and verifying the live published extension unlocks A+. One prompt opens the PR — see showcase/README.md.

Public commitments

The seven invariants standdown holds — the structural guarantees behind the pitch.

• I1: Client-side decisions only. No network call participates in a stand-down decision. Refresh may update the already-applied local policy bundle asynchronously, but decisions use local state synchronously.
• I2: No user profiling. Signals are a closed type and exclude user identity, accounts, balances, emails, login state, and tester-differentiating data.
• I3: Fail toward standing down. Unknown, ambiguous, malformed, or storage error states suppress activation.
• I4: Monotone remote updates. Signed refresh bundles may only broaden detection or lengthen durations and may not edit activation rules.
• I5: Audit log on by default. Decisions and refresh outcomes are locally auditable.
• I6: No remote code. Policies are data. No eval, remote scripts, or dynamic code loading.
• I7: Deterministic and loggable. Given the same local signals, policies, state, and clock, decisions are reproducible.

Documentation

Doc	What's in it
INSTALL.md	Manual install + full API: adapters, quickstarts, self-exemption, per-host disable, signed refresh, interop
AGENTS.md	The consumer playbook an AI agent follows (greenfield)
ADOPTING.md	Brownfield migration: move existing stand-down logic onto the library in shadow mode
POLICIES.md	Every network pack, its signals, and citations
SPEC.md	The behavioral spec and invariants
audit/	The conformance graders
SHOWCASE.md · showcase/	"Graded with standdown" wall of fame + how to add yours (CI-verified)
examples/mv3-extension · examples/content-extension	Minimal working integrations

Releasing

Releases are cut by tagging a version with npm run release; CI then publishes over npm Trusted Publishing (OIDC), with no stored token. See RELEASING.md.

License

MIT — a project by Dupe.com.