> ## Documentation Index
> Fetch the complete documentation index at: https://docs.forge.shieldtp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Frequently asked questions

> Quick answers about Forge access, data, AI behavior, integrations, and source-of-truth boundaries.

<AccordionGroup>
  <Accordion title="Does Forge replace our PSA?">
    No. Forge is the AI work surface for context, reasoning, controlled action, and review. The PSA remains authoritative for provider-backed ticket, project, time, billing, assignment, and administrative records.
  </Accordion>

  <Accordion title="Why don't I see a section another user sees?">
    Navigation depends on your organization rollout and role. Ask an administrator to check the feature permission and whether that product area is enabled.
  </Accordion>

  <Accordion title="Why is a connected system missing from Chat?">
    The system and tool must be valid for the current scope, connected, enabled for Chat, and allowed by your role. Account-scoped tools may also require an account mapping.
  </Accordion>

  <Accordion title="What is the difference between Ask mode and Full access?">
    Ask mode is read-oriented. Full access can expose allowed write tools. Neither mode bypasses role permissions, system access, tool access, Availability, provider support, or configured interactive approvals.
  </Accordion>

  <Accordion title="Do approval policies protect scheduled agents?">
    No. Interactive approval prompts require a person to respond. Configure Custom Agent, Helpdesk, or other unattended tool access with Availability and the underlying role and tool controls.
  </Accordion>

  <Accordion title="What happens when I deny an approval?">
    That tool call does not proceed. The Chat remains available, so you can explain a safer alternative, change the request, or continue with read-only work.
  </Accordion>

  <Accordion title="Does a successful agent run prove the answer is correct?">
    No. It proves the configured workflow reached a successful terminal state. Review the source evidence, output, judge results, and any external writeback before relying on a material conclusion.
  </Accordion>

  <Accordion title="What is the difference between a runbook and a skill?">
    A runbook is a helpdesk procedure and can participate in AI Engineer eligibility. A skill is reusable guidance for a defined task across supported Forge surfaces.
  </Accordion>

  <Accordion title="Does approving an AI BDR proposal send the message or update the provider?">
    No. Approval changes the proposal's Forge review state. Complete the actual external action through the appropriate supported workflow, then record completion.
  </Accordion>

  <Accordion title="Why does a provider record look stale?">
    Confirm organization and account scope, check connection and sync health, compare the source provider, and refresh the supported sync. Do not make a material decision from a visibly stale record.
  </Accordion>

  <Accordion title="Where can I see who changed something?">
    Use the Forge Audit Log for supported terminal write outcomes, then verify the result and provider audit history in the authoritative connected system.
  </Accordion>

  <Accordion title="Can I publish a Custom Agent after one good test?">
    Use several representative dry runs, including missing data, conflicting sources, provider failure, and a condition that should produce no action. Activate only after the workflow owner accepts the behavior and an administrator confirms Availability.
  </Accordion>
</AccordionGroup>