A Composite GitHub Action that runs fount characters with a real OpenAI-compatible API to perform engineering tasks in CI.

Add these repository secrets (Settings → Secrets):

permissions:
  contents: write # required for the built-in push step after success
  actions: write # required for cache delete via github-token

jobs:
  ai-task:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: steve02081504/fount-CI@master
        with:
          task: |
            Fix spelling mistakes in README.
          max-steps: '200'
          max-time-seconds: '1800'
        env:
          OPENAI_BASE_URL: ${{ secrets.OPENAI_BASE_URL }}
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          OPENAI_MODEL: ${{ secrets.OPENAI_MODEL }}

Checkout your repository before this action (e.g. actions/checkout@v4). The agent runs in GITHUB_WORKSPACE and edits files via code-execution and file-operations plugins. On success, when no-push is false and the ref is not a tag, it runs git pull --rebase --autostash then commits and pushes via actions-go/push.

• Character fount-CI-agent: simplified ZL-31 with code-execution + file-operations only
• AI source fount-CI: proxy generator; url, apikey, model from secrets
• Plugin CI-settlement: agent exits with <CI-exit status="success|failure" reason="..."/>

1. char-path — which character loadPart loads (builtin or a folder in your repo).
2. char-config-path — JSON read after load and passed to char.interfaces.config.SetData(...) to bind the AI source and plugins. Unrelated to where the character files live. The path is resolved against GITHUB_WORKSPACE first; if the file is missing there, the action falls back to the same path inside the action package.

Default SetData template default_data/char-config.json:

{
  "AIsource": "${AIsource}",
  "plugins": ["code-execution", "file-operations", "${CIPlugin}"]
}

Placeholders: ${AIsource}, ${CIPlugin}, ${charname}, ${username}.

Both are checked in the CI-settlement plugin GetPrompt (on every buildPromptStruct call, including tool regen loops).

The plugin prompt shows remaining steps and remaining seconds. Tool log entries count toward step length. Commit/push runs only when outputs.status is success.

Each agent round logs a collapsible ::group:: block to the console (tools, reply, exit signal) and an Agent steps section in GITHUB_STEP_SUMMARY.

export FOUNT_CI_TASK="List files in the current directory"
export OPENAI_BASE_URL="https://api.openai.com/v1/chat/completions"
export OPENAI_API_KEY="sk-..."
export OPENAI_MODEL="gpt-4o"
export GITHUB_WORKSPACE="$(pwd)"
# optional — same env vars the action passes to index.mjs
# export FOUNT_CI_CHAR_CONFIG_PATH="default_data/char-config.json"
# export FOUNT_CI_MAX_STEPS="200"
# export FOUNT_CI_MAX_TIME_SECONDS="1800"

./run.sh

Requires fount and Deno.

This action runs AI-generated code on the runner and reads/writes repo files. It also calls external AI APIs (cost applies). Use only on trusted repos and branches, with sensible max-steps and max-time-seconds.