Ollama Agent GitHub Action Example

This repo contains a GitHub Actions workflow that uses the ai-action/ollama-action to run an agent which can suggest commands to run against this repository. For demo purposes the agent will return dummy upgrade commands defined in agent-config.json or in its generated response.

Usage:

• Trigger the workflow via Actions -> Ollama Agent Runner -> Run workflow (workflow_dispatch), or send a repository_dispatch event.
• The action will call the Ollama model, save the JSON response to agent_response.json, and run scripts/execute_commands.py to execute suggested commands.

Files added:

• .github/workflows/ollama-agent.yml - workflow that runs the agent.
• agent-config.json - configuration with working_dir and commands list.
• scripts/execute_commands.py - reads the agent response and executes commands, writing agent_run_results.json.

You may edit agent-config.json to point to a different folder or change the dummy commands.

Triggering via repository_dispatch

• A repository_dispatch event can start the workflow. Example methods:

• Using the gh CLI (recommended if authenticated):

gh api --method POST /repos/scottscott/agent-ollama/dispatches \
	-f event_type=run-ollama-agent \
	-f client_payload='{ "reason": "manual-test" }'

• Using curl with a token in GITHUB_TOKEN or PERSONAL_TOKEN env var:

export PERSONAL_TOKEN="<your_token_with_repo_scope>"
export OWNER=scottscott
export REPO=agent-ollama
curl -X POST "https://api.github.com/repos/$OWNER/$REPO/dispatches" \
	-H "Authorization: token $PERSONAL_TOKEN" \
	-H "Accept: application/vnd.github+json" \
	-d '{"event_type":"run-ollama-agent","client_payload":{"reason":"test"}}'

• I added a helper script scripts/trigger_dispatch.sh that will use gh if available or curl with GITHUB_TOKEN/PERSONAL_TOKEN.

Storing and using your token (quick steps)

• After creating the fine-grained token in GitHub, store it as a repository secret:

  1. Go to https://github.com/scottscott/agent-ollama -> Settings -> Secrets and variables -> Actions -> New repository secret.
  2. Name it PERSONAL_TOKEN and paste the token value, then save.

• To run the trigger locally using that token (one-off):

# export the token into your shell for the current session
export PERSONAL_TOKEN="$PERSONAL_TOKEN"
# run the helper (script will pick up PERSONAL_TOKEN)
bash scripts/trigger_dispatch.sh

• To let a workflow step use the secret, reference it in the workflow step's env:

env:
	PERSONAL_TOKEN: ${{ secrets.PERSONAL_TOKEN }}
run: bash scripts/trigger_dispatch.sh

Security note: Keep tokens in repository secrets (or GitHub's secret manager), not in plaintext files or command history. Use a fine-grained token limited to this repository where possible.

Run Ollama locally (self-hosted runner)

If you want the model to run on your local machine (so it doesn't spin up on GitHub-hosted runners), use a self-hosted runner with Ollama installed.

1. Install Ollama on your machine following https://ollama.com/docs.
2. Register a self-hosted runner for this repository: GitHub -> Settings -> Actions -> Runners -> Add runner. During registration add the label ollama (or edit the runner labels later).
3. On your machine, ensure the runner service is running and that Ollama can be invoked by the runner user.
4. Trigger the workflow Ollama Agent (Self-hosted) from the Actions tab or via repository_dispatch. It will run on your self-hosted runner (label ollama) and use the local Ollama installation.

Notes:

• The repository includes .github/workflows/ollama-agent-selfhosted.yml which is configured to run on runs-on: [self-hosted, linux, ollama].
• Use a smaller model for quick tests (change model: tinyllama in the workflow) while verifying behavior.

Quick notes on safety, PR flow, and CI

• Allowlist & validation: agent-config.json may include an allowlist array of regex patterns. The runner script scripts/execute_commands.py supports two modes:

  • validate — checks agent-suggested commands against the allowlist and writes agent_run_results.json with the validation details.
  • execute — runs only the commands that match the allowlist and marks skipped commands.

• Approval-gated self-hosted run: the self-hosted workflow now separates prepare (run agent + validate) and execute (post-approval) jobs. To require manual approval create a repository environment named agent-approval (Settings → Environments) and add required reviewers or protection rules. When a run reaches the execute job, GitHub will pause and request approval from the configured reviewers.

• PR-based proposal workflow: instead of executing changes, the workflow .github/workflows/ollama-agent-pr.yml can run the agent and open a proposal PR with AGENT_PROPOSAL.md so humans can review and merge changes.

• CI for the repo: a CI workflow runs unit tests under .github/workflows/ci.yml. To run tests locally:

python3 -m unittest discover -v tests

AI Notice:

The contents of this Repository were built with AI Coding assistants.