Configuring GitHub For Fullsend

The goal of this document is that you configure Fullsend for your GitHub repository. GitLab repositories use a different command (fullsend repos install --forge gitlab). See Configuring GitLab for that flow.

Prerequisites

• For Vertex agents, you have your WIF provider URL from Getting Inference. OpenAI-only setup does not need it.
• Download the latest fullsend CLI.
• Download the latest gh CLI and authenticate with it.

Note: If your organization restricts classic personal access tokens, gh auth login may produce a token that GitHub rejects with a 403. Create a fine-grained PAT scoped to the target repository with Contents, Workflows, Secrets, Variables (read/write), Pull requests (read/write — not needed with --direct), and Metadata (read-only), then export GH_TOKEN=github_pat_... before running setup. The CLI checks GH_TOKEN → GITHUB_TOKEN → gh auth token, in that order.

Fork limitation: Fine-grained PATs are scoped to a single GitHub organization and cannot create forks across org boundaries. When the CLI detects a fine-grained token, the fork delivery option ([f]) is unavailable. The CLI will offer the upstream option ([u]), which pushes a branch to the target repository and creates a PR containing the fullsend scaffolding files. No changes are made to the default branch until the PR is merged. This limitation only applies to the initial setup — after setup, fullsend agents use their own GitHub App tokens and do not require a fork.

Installing GitHub Applications

Install the following agent applications to your organization and provide them permissions to the repository you want to install Fullsend to.

Role	Installation URL
triage	<https://github.com/apps/fullsend-ai-triage/installations/new>
coder	<https://github.com/apps/fullsend-ai-coder/installations/new>
review	<https://github.com/apps/fullsend-ai-review/installations/new>
retro	<https://github.com/apps/fullsend-ai-retro/installations/new>
prioritize	<https://github.com/apps/fullsend-ai-prioritize/installations/new>

Note: You do not need the fullsend dispatch app (fullsend-ai-fullsend). Each repository dispatches through its own shim workflow.

Note: Installing a subset of GitHub Apps does not automatically limit which agents are active. You must also pass the --agents flag (see below) to match the set of apps you installed. For example, if you only install the triage and review apps, pass --agents triage,review when running setup.

Configuring GitHub

Run the command:

fullsend github setup <org>/<repo> \
  --inference-project "<gcp-project>" \
  --inference-wif-provider "<wif-provider-url>"

Where <org>/<repo> refers to the GitHub organization and repository you want to enable inference for, <gcp-project> is your GCP project name, and <wif-provider-url> is the WIF Provider URL created at Getting Inference.

The command creates files, secrets and variables in your repository. If you will use only OpenAI agents, omit both --inference-project and --inference-wif-provider. Setup then writes no GCP inference secrets; a later Vertex run requires both credentials. Configure each enabled agent's runtime and model for OpenAI before it runs. For OpenAI credentials, see OpenAI Workload Identity.

Enabling a subset of agents

By default, the setup command configures all available agent roles. To enable only specific agents, pass the --agents flag with a comma-separated list of roles:

fullsend github setup <org>/<repo> \
  --inference-project "<gcp-project>" \
  --inference-wif-provider "<wif-provider-url>" \
  --agents triage,review

Only the listed agents will be configured. Make sure you have installed the corresponding GitHub Apps for each agent you enable (see the table above).

For the full list of setup flags, see the CLI reference.

Using a vendor preset

If your platform operator provides a curated preset configuration, you can install it directly instead of relying on per-flag generation:

fullsend github setup <org>/<repo> \
  --inference-project "<gcp-project>" \
  --inference-wif-provider "<wif-provider-url>" \
  --config "<path-or-url>" \
  --config-hash "<sha256-hex>"

When --config is provided, the preset content is committed unchanged as .fullsend/config.base.yaml. Explicit persistent setup flags (--runtime, --agents, --mint-url, --inference-*) are written to the .fullsend/config.yaml overlay and override the same values from the preset. Omitted flags inherit from the preset, then from compiled-in defaults. If you pass a persistent flag whose value already matches the preset or compiled default, setup still writes it into the overlay and warns that the field is now pinned locally — later updates to the base layer or default will not apply until you remove that key from .fullsend/config.yaml. Per-run flags such as --dry-run never pin. The --config-hash flag is optional but recommended for remote URLs — it verifies the SHA-256 digest of the fetched content before committing. --config is only valid for per-repo mode.

This is where the agent runtime is selected: on a terminal, fullsend github setup asks once (press Enter to keep claude, the stable default); --runtime sets it explicitly. pi and codex are experimental and meant for opt-in pilots — see Choose a Runtime for what the runtimes are and how to change the selection after setup.

To apply the same preset across many repositories, declare defaults.config_base (and optionally defaults.config_base.sha256) in repos.yaml and run fullsend repos install. See Repo Management — Configuration presets.

Testing Fullsend

After installing open a new issue or comment /fs-triage in an open issue. Then visit the Actions tab to see the Fullsend workflow in action. In some minutes the fullsend-ai-triage bot should post a comment in the issue.

Next steps

• Read Repo Management to learn how to install and manage Fullsend across many repositories, including sharing configuration presets between them.
• Read the Agents section to learn about the default agents Fullsend ships with.
• Explore other sections of this documentation for more information.