Copilot already reads a repo instructions file. Give it one that stops overpaying.

How this maps onto Copilot

No native skill loader. Repo custom instructions are the injection point.

Copilot doesn't have a first-class "skills" directory the way Claude Code does — there's nothing that auto-discovers skills/firstpass/SKILL.md on its own. What GitHub Copilot does read automatically is repository custom instructions: a single .github/copilot-instructions.md file at the repo root, written in plain Markdown, that's applied in full to Copilot Chat, code review, and the Copilot coding agent for every request in that repository. As of August 2025, the Copilot coding agent also reads AGENTS.md (root or nested — the nearest one in the directory tree wins), alongside the GitHub-native format. That means Undercut's policy has to be delivered as instruction content Copilot reads whole, not as a skill it discovers — see Install below for the exact mechanics and what's still unverified.

The tier names — cheap, standard, frontier, apex — stay generic on purpose. models.md gives the rule of thumb rather than a fixed mapping for Copilot specifically: Copilot's own model picker exposes several vendors' models (varies by plan), so cheap resolves to the smallest/fastest model you have access to, standard to your default work model, frontier to the highest-reasoning model, and apex to a single tie-break call on the strongest model available. Nothing here sits between Copilot and its model backend — it's instruction text Copilot reads at request time, no proxy, no network hop.

Install for Copilot

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

Option 1 — inline it into repo custom instructions (recommended)

.github/copilot-instructions.md is read in full by Copilot Chat, code review, and the coding agent on every request in the repo — so appending the policy text there is the most reliable way to get it in front of Copilot, since it doesn't depend on Copilot separately choosing to open another file.

Option 2 — scope it with a path-specific instructions file

Copilot also supports .github/instructions/*.instructions.md files with a YAML applyTo glob in frontmatter, so you can scope the policy (or exclude it from code review with excludeAgent: code-review) instead of applying it repo-wide:

You'll need to add an applyTo: "**" frontmatter block yourself — the raw SKILL.md doesn't ship with one, since it's written to be portable across agents that don't use this convention.

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 a small PR against a repo with the instructions installed and watch which model tier Copilot's coding agent session log reports it used

Zero lock-in. It's instruction text Copilot reads — delete 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 Copilot specifically, it's not a "skill" Copilot discovers on its own: it only takes effect if you've put it in an instructions file Copilot reads (see Install above). Copilot'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.