Workflows via Ask Gina Skill

What It Does

Provides a practical workflow-authoring and operations standard for Gina sandbox automation.

• Creates and validates workflow definitions.
• Runs workflows and inspects artifacts/logs.
• Applies a repeatable eval -> optimize -> compare loop.
• Uses safe TypeScript/SQL/KV patterns for step logic.

When To Use

• You are creating or maintaining multi-step workflow orchestration.
• You need reproducible debugging from run artifacts.
• You want measurable improvements using baseline comparisons.

When Not To Use

• The task is a single action with no orchestration requirement.
• You only need high-level strategy language without runnable steps.
• You cannot provide explicit permissions or side effects.

Inputs

• Workflow intent and success criteria.
• Trigger definition and input schema.
• Required tools/data sources and permission scope.
• Optional baseline run ID for optimization.

Outputs

• Validated workflow definition (.ts).
• Runnable execution with traceable artifacts.
• Evaluation record with baseline comparison.
• Clear rollback path for regressions.

Core Commands

workflow create <id>
workflow validate <id>
workflow run <id> [--input JSON]
workflow status <run-id>
workflow logs <run-id> [--step <step-id>]
workflow eval <run-id>
workflow optimize <id> --baseline <run-id>
workflow rollback <id> <opt-run-id>

Setup

• Confirm workflow tooling is available (workflow list should succeed).
• Scaffold or open the target workflow in /workspace/.harness/workflows/.
• Keep active versions on @latest.ts naming when versioned variants exist.
• Validate before every run: workflow validate .
• For risky changes, capture a baseline run and eval before editing.

Capability Contract Checklist

For each workflow entry, explicitly define:

• Trigger.
• Inputs.
• Outputs.
• Side effects.
• Failure modes.
• Permission scope.

Failure Modes

• Validation failure from malformed step definitions.
• Runtime errors in TS/SQL/Bash steps.
• Missing tool permissions or tool availability.
• Data shape changes causing parse/cast failures.
• Timeout/retry exhaustion in external calls.

Security And Permissions

• Use least privilege by step using allow and block.
• Declare permissions in the submission contract (no wildcard permissions).
• Never include raw secrets in skill text, logs, or examples.
• Treat writes (files, KV, external posts, trading actions) as explicit side effects.

Evidence Expectations

• Setup path that a reviewer can execute in under 10 minutes.
• One reproducible run artifact or run log example.
• Clear statement of expected outputs and acceptable failure behavior.

Optional Directories

workflows/
  SKILL.md
  references/   # implementation and API details
  scripts/      # optional helpers for repeatable checks
  assets/       # optional diagrams/screenshots

Reference Material

Deep technical references are intentionally split out:

• references/cli-and-definition.md
• references/eval-optimize-and-artifacts.md
• references/polymarket-patterns.md