governance-setup

Governance Setup — PromptSpeak

You are helping the user configure PromptSpeak governance for their project.

Workflow

1. Assess the project context — Ask the user:

   • What kind of agents will be governed? (coding assistants, data pipelines, customer-facing bots)
   • What's the risk tolerance? (strict/regulated vs permissive/experimental)
   • Are there specific tools or operations that must always require human approval?

2. Map answers to a governance profile, then let the user override specifics:

   Profile	When to use	Confidence	Holds	Drift	Post-Audit
   Lockdown	Regulated domains (legal, financial, healthcare), production with PII	0.95	All tool calls held for approval	ON, halt at 0.3	ON
   Standard	Production coding assistants, internal tools, CI pipelines	0.7	Destructive ops only (deletes, deployments, external API writes)	ON, halt at 0.6	ON
   Exploratory	Sandboxes, prototyping, hackathons, trusted single-user agents	0.3	None (log only)	OFF	OFF

   Present all three profiles with the table above. Recommend Standard unless the user's answers clearly indicate otherwise. After the user picks a profile, ask: "Any overrides? For example, specific tools that should always require approval, or a different drift threshold."

   Apply overrides on top of the chosen profile before proceeding to step 3.

3. Apply configuration using PromptSpeak tools:

   • Set confidence thresholds with ps_confidence_bulk_set
   • Configure hold behavior with ps_hold_config
   • Set feature flags with ps_feature_set
   • Register policy overlays with ps_config_set and ps_config_activate

4. Verify — Run ps_execute_dry_run with a sample frame to confirm the pipeline behaves as expected.

5. Export — Call ps_config_export so the user can save their configuration.