# Built-in Reefine: reef serve --recipe reefine --model ollama/gemma4:26b
schema-version: 2
reef:
  host: 127.0.0.1
  port: 8901
  # Unset REEF_TOKEN leaves the loopback service without authentication; set it to require a Bearer token.
  token: ${REEF_TOKEN}
  run-dir: .reef/reefine/stack

recipe:
  implementation: reef.recipe.reefine:ReefineRecipe
  # The recipe uses the upstream runtime's model for evaluation as well as serving.

  config:
    evolution:
      adapter: pi
      # A coding agent (pi) answers a request: it edits the tree, runs the changed harness to prove it, and
      # hands the entries back. Jailed with bubblewrap and pasta where the host has them; without them the text
      # proposer answers, unless proposer_agent.sandbox (or REEF_PROPOSER_SANDBOX) is e2b, an E2B cloud sandbox
      # (E2B_API_KEY), or none, which runs the agent unisolated (trusted clients only).
      propose: reef.recipe.reefine.agent:propose
      proposer_agent:
        timeout_s: 1800
        trial_timeout_s: 300
      # Images, embeddings, speech and decisions: Reef relays the harness's calls to this gateway with its key,
      # unrecorded. The key is REEF_MULTIMODAL_API_KEY or --recipe.config.evolution.multimodal.api_key, else the
      # upstream's when the upstream is this gateway; without one those routes answer 501. For another gateway:
      # preset: openai-compatible and its url.
      multimodal:
        preset: openrouter
        api_key: ${REEF_MULTIMODAL_API_KEY}
      evaluate: reef.recipe.reefine.evolution:evaluate
      # The seed carries the /reefine command and the pi extension API skill the proposer reads.
      requests: true
      # The notice at session start offers the update once the head moved.
      version_check: true
      # An evolved extension runs in pi's process with your privileges, so it waits for a promote.
      review_kinds: [code_extension]
      # The floor runs the candidate alone on the health task and publishes it when the task passes: it checks
      # that the tree still works (the model binding, the tools, the extensions load), not that the requested
      # change does; the step page's design and review notes and the person judge the change itself.
      selection: floor
      # Every step's proposer calls, parsed proposal and evaluation episodes, so a row can be read back.
      step_record_dir: .reef/reefine/steps
      # The proposal inbox of POST /reef/harness/proposals; the default is .reef/proposals under the service's directory.
      proposals_dir: .reef/reefine/proposals
      # No binary: at startup reef installs the descriptor's pinned pi under
      # ~/.local/share/reef-harness/pi, shared with the client install script.
      # Set evolution.binary to a path to use an install of your own instead.
      # On an adapter whose prompt is a Harbor task directory (terminus), the recipe runs the health task as
      # the directory reef/recipe/reefine/health instead, scored by its verifier's reward.
      tasks:
        - '[health] Run the shell command `echo reef-ok` with your shell tool and reply with its exact output
          as a plain word alone on the last line.'
      seed:
        - id: answer-style
          name: skill
          config:
            name: answer-style
            text: '# answer-style


              Starter skill. The evolution loop replaces this placeholder with

              concrete guidance learned from failing tasks.

              '
    batch-size: 1
    # One step per instruction and no failure driven step between them: the measurement counts a step per request.
    training-mode: manual

inference:
  upstream-url: ${REEF_UPSTREAM_URL:?}
  upstream-model: ${REEF_UPSTREAM_MODEL:?}
  upstream-api-key: ${REEF_UPSTREAM_API_KEY}
storage:
  agent-record-dir: .reef/reefine/agent-record
  artifact-repository: .reef/reefine/artifacts.git
  artifact-work-dir: .reef/reefine/artifact-work
  artifact-cache-dir: .reef/reefine/artifact-cache