Table of Contents

Bindings - example override

Copy this table into your repository's two agent override files (.github/agents/backlog-worker.agent.md and .github/agents/backlog-pm.agent.md) and fill in your own values. Nothing else in the template needs editing.

Binding Value
{repoId} my-repo
{owner}/{repo} my-org/my-repo
{ghAccount} my-github-account
{homeRegion} local (derive it - see below; do not copy this)
{conventionsDoc} .github/copilot-instructions.md
{implementationAgent} feature-dev, or omit if you have no implementation agent

How to get each value

  • {repoId} is what repocontext_list_repos reports, which defaults to the final path segment of the indexed path. It is not your current working directory. In a git worktree it is the base repository's id, so repocontext_list_repos is the only reliable way to read it. Guessing it from your directory is the single most common adoption mistake.

  • {owner}/{repo} is the GitHub repository that mirrors backlog items as issues. It does not have to be the repository the code lives in, but the agents assume one of each.

  • {ghAccount} is the account every gh call authenticates as. Set it explicitly rather than relying on an ambient token, which is often not the identity you intend.

  • {homeRegion} is the region claims are taken in. Derive it from your own cluster; do not copy a value from this table. Call lattice_list_regions to see the regions the server routes to, and repocontext_claim_status on any claimed item to see the region a claim actually records - that recorded value is what the tag must match. On a single-region deployment these commonly differ in a way that catches people out: the only routable id is current, while claims record local.

    Region scoping is a property of your deployment, not of this tag. The intent is that a claim taken in one region refuses a write from another, and that does hold where the cluster genuinely spans regions. Where it does not, a geographic value such as uksouth is not enforced at all - it is not a region the cluster routes, every worker's region check passes vacuously, and a claim from anywhere succeeds. That failure is worse than a no-op, because the protocol documents the guarantee as enforced and workers rely on it. Treat a value your cluster does not route as a defect in the binding and fix it here.

Why bindings rather than editing the template

The template files are the source of truth and are consumed unmodified. Keeping every repository-specific value in one small override means:

  • upgrading to a newer template is a file replace, not a merge;
  • a diff of your override shows your entire deviation from the base;
  • the template cannot accumulate one repository's assumptions.

If a binding is unavailable at runtime the agents stop and report rather than guessing. That is deliberate: a gh call under the wrong identity and a claim taken in the wrong region are both expensive to unpick after the fact.