Job pack · chain
Named workflow
The path is knowable. Chain, route, parallel, eval-opt. Code owns the edges. An agent island only at ambiguity.
When: You can name the stages and check the intermediates.
Do not build an agent. Build a workflow. Put code gates between hops. Add a loop only where the eval fails without one.
GET /api/canon/jobs/workflow-pipeline?format=md
Playbooks
Doctrine to load
- Workflows versus AgentsAnthropic draws a bright line: workflows orchestrate LLMs through predefined code paths; agents let the model direct its own process. Most production systems that work are workflows with a small agentic island.
- Prompt ChainingA sequence of LLM calls where each step's output is the next step's input, with programmatic gates between them. The first workflow to reach for.
- RoutingClassify the input, then send it to a specialist prompt, model, or workflow. Separation of concerns for language.
- ParallelizationFan-out independent LLM work, then reduce. Sectioning for speed; voting for diversity. Aggregation is code, not another creative writer.
- Evaluator–OptimizerOne model generates, another critiques against explicit criteria, loop until the critic accepts or the budget dies. The missing unit test of generation.
- Own Your Control FlowThe model may propose the next edge. Your code decides whether that edge is legal, budgeted, and reversible. Do not outsource the graph.
Refuse
- Unbounded loop. No max-turns, token, wall-clock, or money budget. The agent rambles until the bill or the context dies.
- Eval by demo. Architecture first, golden set never. Success is a recorded GIF.
- Star-driven selection. Picking the viral harness of the week. Contribution-free clones with disabled PRs.
Checklists
Ready to build
- Named termination: done schema, max turns, token/money cap.
- Sandbox / allowlist for every side effect. Secrets in the executor.
- An explicit context builder: pinned, working, recalled, compacted.
- A golden set of 20–50 real tasks, run in CI on prompt change.
- Launch / pause / resume APIs. HITL is a tool that checkpoints.
Eval quality
- 20–50 real tasks, not toys. Include the failures you have already seen.
- CI on prompt, tool schema, and skill changes.
- Written rubric. If you cannot score it, you cannot loop it.
- Generator and evaluator do not share prompt or incentive.
Recipes
Default corpus
n8nlanggraphdifydspy