fabro/lib/components/fabro-workflow
Bryan Helmkamp ddcf86bcfb
fix(workflow): materialize only prompt values the selected fidelity renders
Prompt-value demotion previously demoted every prior outcome value and
visible context value wholesale, before the preamble builders decided
what the selected fidelity renders. That materialized blobs and sandbox
files no generated prompt referenced — e.g. a compact prompt created a
blob for a prior LLM response even though compact stage details never
include response.<node_id>.

Make the materialization set fidelity-aware:

- preamble::rendered_value_selection computes, per fidelity, exactly
  which context keys and (stage, key) outcome values the builders
  render, sharing the recency-window logic with the builders.
- demote_large_values_for_prompt takes a PromptValueSelection and
  demotes only the values it names.
- The lifecycle unions the node's selection with each parallel branch's
  effective fidelity selection, and skips demotion when the union is
  empty.

Durable context and node outcomes are unchanged: response.<node_id>
stays the exact model response and output.<node_id> the validated
structured value. A drift-guard test asserts, at every fidelity, that
each demoted path is referenced by the built preamble and no unselected
large value leaks inline.

Closes #800

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011bs8C1fjVCu8a8qDvFskYV
2026-08-25 07:54:14 -04:00
..
src fix(workflow): materialize only prompt values the selected fidelity renders 2026-08-25 07:54:14 -04:00
tests fix(workflow): materialize only prompt values the selected fidelity renders 2026-08-25 07:54:14 -04:00
Cargo.toml Forward fabro-store/test-support through dependent test-support features 2026-08-24 14:02:35 -04:00
README.md fix(workflow): make publish failures terminal 2026-07-27 11:25:18 -04:00

fabro-workflow

A DOT-based pipeline runner for multi-stage AI workflows. Define workflows as Graphviz digraph files and execute them with pluggable handlers, conditional routing, human-in-the-loop gates, parallel branching, retry policies, and checkpoint-based recovery.

Key Concepts

  • Graph -- A directed graph parsed from DOT syntax containing nodes, edges, and attributes. The graph carries a goal describing the pipeline's purpose.
  • Node -- A workflow step. Graphviz shapes map to handler types (e.g., Mdiamond = start, Msquare = exit, box = agent, tab = prompt, diamond = conditional, hexagon = human gate, component = parallel).
  • Edge -- A connection between nodes with optional condition, label, weight, and fidelity attributes that control routing.
  • Handler -- An async trait implementation that executes a node and returns an Outcome. Built-in handlers include StartHandler, ExitHandler, AgentHandler, PromptHandler, ConditionalHandler, HumanHandler, ParallelHandler, FanInHandler, CommandHandler, and SubWorkflowHandler.
  • Outcome -- The result of executing a handler, carrying a StageOutcome (Success, Fail, PartialSuccess, Retry, Skipped), optional routing hints (preferred_label, suggested_next_ids), and context updates.
  • Context -- A thread-safe key-value store shared across pipeline stages, supporting snapshots and isolated cloning for parallel branches.
  • Interviewer -- A trait for human-in-the-loop interactions. Implementations include AutoApproveInterviewer, QueueInterviewer, CallbackInterviewer, ConsoleInterviewer, and RecordingInterviewer.
  • Checkpoint -- A serializable snapshot of execution state (completed nodes, context values) for crash recovery and resume.

Pipeline Definition

Pipelines are defined using Graphviz DOT syntax:

digraph MyPipeline {
    graph [goal="Implement and validate a feature"]
    rankdir=LR
    node [shape=box, timeout="900s"]

    start     [shape=Mdiamond, label="Start"]
    exit      [shape=Msquare, label="Exit"]
    plan      [label="Plan", prompt="Plan the implementation"]
    implement [label="Implement", prompt="Implement the plan"]
    validate  [label="Validate", prompt="Run tests"]
    gate      [shape=diamond, label="Tests passing?"]

    start -> plan -> implement -> validate -> gate
    gate -> exit      [label="Yes", condition="outcome=succeeded"]
    gate -> implement [label="No", condition="outcome!=succeeded"]
}

Usage

Parsing and Validating a Pipeline

use fabro_workflow::operations::{create, CreateOptions};

let dot_source = r#"digraph Simple {
    graph [goal="Run tests"]
    start [shape=Mdiamond]
    exit  [shape=Msquare]
    work  [shape=box, prompt="Run the test suite"]
    start -> work -> exit
}"#;

let validated = create(dot_source, CreateOptions::default())
    .expect("pipeline should parse");
validated.raise_on_errors().expect("pipeline should validate");
let (graph, _, _) = validated.into_parts();
assert_eq!(graph.name, "Simple");
assert_eq!(graph.goal(), "Run tests");

operations::create parses the DOT source, applies built-in transforms (variable expansion, stylesheet application, preamble injection), and returns diagnostics through Validated.

Running a Pipeline

use fabro_workflow::operations::start;
use fabro_workflow::pipeline;

// Use `operations::start(...)` for the full
// initialize -> execute -> conclude -> publish -> finalize flow.
// Use `pipeline::initialize(...)` + `pipeline::execute(...)` when you need partial lifecycle control.

Custom Handlers

Implement the Handler trait to add custom node behavior:

use arc_workflows::handler::Handler;
use arc_workflows::context::Context;
use arc_workflows::graph::{Graph, Node};
use arc_workflows::outcome::Outcome;
use arc_workflows::error::ArcError;
use async_trait::async_trait;
use std::path::Path;

struct MyHandler;

#[async_trait]
impl Handler for MyHandler {
    async fn execute(
        &self,
        node: &Node,
        context: &Context,
        graph: &Graph,
        run_dir: &Path,
    ) -> Result<Outcome, ArcError> {
        // Custom logic here
        Ok(Outcome::success())
    }
}

Model Stylesheets

CSS-like stylesheets control LLM model assignment with specificity-based cascading:

digraph Styled {
    graph [
        goal="Build feature",
        model_stylesheet="
            * { model: claude-sonnet-4-5;}
            .code { model: claude-opus-4-6; }
            #critical_review { model: gpt-5.2;}
        "
    ]
    // ...
}

Selectors by specificity: * (universal, 0) < shape (1) < .class (2) < #id (3). Explicit node attributes are never overridden.

Condition Expressions

Edge conditions use a simple expression syntax for routing:

outcome=succeeded
outcome!=failed
outcome=succeeded && context.tests_passed=true
my_flag

Clauses support =, !=, and bare key truthiness checks, joined with &&.

Human-in-the-Loop Gates

Nodes with shape=hexagon or type="human" pause execution for human input. Outgoing edge labels become selectable options, with accelerator key parsing for patterns like [A] Approve and F) Fix.

Parallel Execution

Nodes with shape=component fan out to branches concurrently. Branches receive isolated context forks, share the same sandbox checkout, and always finish before the workflow continues. Use max_parallel to limit concurrency; concurrent workspace writes are user-managed.

Checkpoints and Resume

The engine saves a checkpoint after each node. Resume from a checkpoint with engine.run_from_checkpoint(&graph, &config, &checkpoint).

Architecture

parser (DOT -> AST -> Graph)
  -> transform (variable expansion, stylesheet, preamble)
    -> validation (14 lint rules)
      -> engine (execution loop with retry, edge selection, goal gates)
        -> handler (pluggable node executors)
          -> interviewer (human-in-the-loop I/O)