Workflows And Agents

PromptPress Documentation

Back

PromptPress workflow operations connect planning, generation, review, and publish through repeatable execution paths.

Core Building Blocks

  • Campaign: planning container for recurring initiatives.
  • Workflow: reusable process definition.
  • Step: ordered unit of execution.
  • Run: one execution instance of a workflow.
  • Agent: task executor used by workflow steps.

Where Teams Manage This

  • Campaigns: create and track initiatives.
  • Studio: run workflows and monitor step output.
  • ReleaseDeck: review generated artifacts before publishing.
  • AgentBase: execute agents and inspect history.
  • PromptSpace: manage reusable prompts and model mappings through Prompt Center and ModelSource.

Agent creation belongs in AgentBase. The current Add Agent button opens an Agent Forge coming-soon dialog.

Lifecycle Operations

Typical run lifecycle operations include:

  • Start run.
  • Cancel active run.
  • Force complete in exceptional cases.
  • Restart from a valid checkpoint step.
  • Discard/archive runs no longer needed.

Step-by-Step: Campaign Run to Publish

  1. Create a campaign with a clear objective and cadence.
  2. Link the campaign to an approved workflow.
  3. Start a run from Studio.
  4. Monitor step progress and capture failures early.
  5. Review artifacts in ReleaseDeck.
  6. Move approved output to ContentVault.
  7. Publish and record run outcomes for the next cycle.

Step-by-Step: Safe Prompt and Agent Updates

  1. Open PromptSpace.
  2. Update Prompt Center or ModelSource, one controlled change at a time.
  3. Test using a known benchmark topic.
  4. Validate output quality in Studio or PromptPress.
  5. Roll out to wider usage only after validation.

Step-by-Step: AgentBase Operational Check

  1. Open AgentBase and select the target agent.
  2. Run a task with a clear, short instruction.
  3. Review execution history and status.
  4. Capture failures with run time and task details.
  5. Escalate config changes back to PromptSpace owners.

Reliability Practices

  • Keep steps small and deterministic.
  • Avoid major graph edits during active campaign windows.
  • Version significant prompt/agent changes.
  • Use restart only after root-cause review.
  • Treat a restarted failed run as superseded once its replacement child run exists. Superseded failed parents are not counted in campaign totals or failed-job lists.
  • Review failed or discarded runs on a regular cadence.

PromptSpace and AgentBase Responsibilities

Use clear ownership:

  • PromptSpace: define and govern prompts, agent behavior inputs, and model choices.
  • AgentBase: execute and monitor operational runs.

Model defaults use a layered fallback strategy:

  1. Explicit user/run selection.
  2. Workspace default from accounts.public_data.ai_defaults.
  3. Cloud-provider default.
  4. Same-provider safe fallback.
  5. Secondary-provider fallback when configured.

Treat PromptSpace changes as production configuration, not ad hoc edits.

Governance Model

A practical split:

  • Editors own quality standards and publish readiness.
  • Operators own workflow topology and run health.
  • Admins own prompt/model governance and access policies.

Safe Decommissioning

Before retiring a workflow:

  1. Resolve or close active runs.
  2. Unlink dependent campaigns.
  3. Archive historical runs per retention policy.
  4. Remove or disable workflow after rollback window.

Related Docs