agentjido/jido-action

Jido Action

jido_action is the upstream Hex package name.

Start Here

Use this skill when the task is centered on Jido.Action as the unit of executable work.

Good triggers:

• "Create a new Jido.Action for this side effect."
• "Review this action for schema validation or output shape."
• "Turn the action docs into a runnable example."
• "Figure out whether this logic belongs in one action or in agent orchestration."

Read references/builder-notes.md before editing when the task involves action contracts, composition, retries, or error boundaries.

Primary Workflows

Scaffold an action

• Start from the action's boundary: one job, one clear input contract, one clear result shape.
• Define parameters and validation up front instead of burying checks in run/2.
• Keep side effects explicit so callers can reason about retries and compensation.

Extend or refactor an action

• Split oversized actions when they hide orchestration, branching, or unrelated IO.
• Preserve stable inputs and outputs when improving internals.
• Prefer composition through other Jido layers over building a giant action with internal state machines.

Turn docs into runnable examples

• Use realistic params and output values.
• Show how an action is invoked in isolation before showing how an agent or directive uses it.
• Include one error path if the action handles validation or external failures.

Review boundaries

• Keep low-level units of work in Jido.Action.
• Keep routing, branching, and multi-step control flow in jido or jido-behaviortree.
• Keep signal schemas in jido-signal.

Build Checklist

• Define the input schema first.
• Decide whether the action is pure, IO-bound, or adapter-like.
• Return consistent success and failure shapes.
• Add focused tests for validation, happy path, and recoverable failure.

Boundaries

• Do not use one action to represent a whole agent workflow.
• Do not hide long-lived state inside the action module.
• Do not mix unrelated side effects just because they happen in one feature.