Cascade already reads a root AGENTS.md. Give it one that stops overpaying.

How this maps onto Windsurf

A multi-model IDE. AGENTS.md is the confirmed way in.

Windsurf was acquired by Cognition, and its documentation now lives under docs.devin.ai, branded as part of Devin Desktop — docs.windsurf.com redirects there. Per that documentation, Windsurf's Cascade agent now prefers workspace rules from .devin/rules/*.md, with .windsurf/rules/*.md and the legacy single-file .windsurfrules kept as fallbacks for backward compatibility. Global rules live at ~/.codeium/windsurf/memories/global_rules.md (capped at 6,000 characters; workspace rules files cap at 12,000). Separately, Cascade also reads a root-level AGENTS.md automatically — documented as location-scoped rules with zero configuration required, which is the most portable of the confirmed paths and doesn't depend on which of the rules-file conventions your workspace already uses.

Windsurf doesn't have a model family of its own the way Claude Code or Codex do — it's a multi-model IDE, wired up to whichever provider(s) you've configured in Cascade's model picker. Undercut's tier names — cheap, standard, frontier, apex — are generic on purpose for exactly this case. models.md gives the rule of thumb rather than a fixed mapping: cheap resolves to the smallest/fastest model you have access to in Cascade, standard to your default work model, frontier to the highest-reasoning model on offer, and apex to a single batched tie-break call on the strongest model available.

Nothing about this sits between Cascade and its model backend. It's a Markdown file Cascade reads at dispatch time — no proxy, no network hop, nothing to route around if you delete the block.

Install for Windsurf

Free. MIT. Nothing installs, nothing proxies your traffic.

Option 1 — append to AGENTS.md (most portable, confirmed path)

Cascade reads a root-level AGENTS.md automatically, with no rules-file setup required — this is the path we're most confident about, and it doesn't depend on which of Windsurf's own rules conventions your workspace uses.

Option 2 — Windsurf's own workspace rules

If you'd rather keep it inside Windsurf's own convention, drop it under .devin/rules/ — the preferred path per current docs since the Devin Desktop rebrand — or the legacy fallback location, .windsurf/rules/. Both are confirmed in Windsurf's documentation.

Want it applied to every workspace instead of one repo? Copy the same content into ~/.codeium/windsurf/memories/global_rules.md — note that file caps out at 6,000 characters (workspace rules files cap at 12,000), so check the raw SKILL.md fits before relying on the global copy.

Validate it yourself

1. 1Read the published results — testing/README.md and the raw JSON in testing/results/
2. 2Reproduce for free, no API key, no spend: node src/main.js --mock
3. 3Open Cascade on a small task with the policy installed and confirm in the model picker/response which tier it actually dispatched to

Zero lock-in. It's Markdown text Cascade reads — delete the file or the block and you're back to status quo.

Six flags assign a tier. Three triggers are the only way up.

This is a named, versioned spec — not internal logic we could quietly change. When the rubric changes (a new flag, a different cap), the version number moves and the changelog says why.

Every unit of work gets scored against six flags, then assigned the cheapest tier that can plausibly handle it. It only moves up a tier when something objective proves it needs to — a failed check, a measured disagreement, an explicit uncertainty flag. It never moves up on a guess, and it never moves back down mid-task.

 unit of work
      |
  6-flag rubric
      |
  +---+----+
  | 0 flags|--> cheap
  | 1-2    |--> standard
  | 3+/own |--> frontier
  +---+----+
      | fail x2 / disagree / uncertain
      v
  escalate +1 tier
  (residue only)
      |
  still unresolved?
      v
  single batched
  apex tie-break

Why it holds up

Flags steer. Verification and escalation decide.

Even when the rubric flags are scored imperfectly, the escalator still routes work to the right tier almost every time — because a wrong flag only ever costs one extra cheap attempt, never a wrong answer or a big bill.

Honest limits

What this doesn't do.

Not a proxy, not a gateway, doesn't auto-flag in production — and on Windsurf specifically, Cascade has its own confirmed skill-discovery path (.windsurf/skills/, distinct from Devin CLI's .devin/skills/ — see Install above), but that's a documented path, not a session we ran ourselves, so treat the live behavior as unverified until tested. Cascade's own reliability at scoring the flags hasn't been separately measured the way Claude Code's has (see above); treat that as unverified until tested. And it doesn't promise a dollar figure for your workload. Full list of general limits, with the finance/procurement note and the paid-tier disclosure, is on the main page.