Add confirmation UX and immutable audit logging for agent actions

[michaelmwu]

Parent: #257

Goal

Add confirmation UX and immutable audit logging for agent-driven actions from Discord.

Scope

• Show exact planned actions before non-trivial writes.
• Require explicit confirmation for medium/high risk actions.
• Require a second authorized human for critical actions.
• Store audit records for every attempted tool call, allowed or denied.
• Keep audit writes best effort where appropriate so command flows do not fail solely because audit logging failed.

Audit fields

• actor user ID
• Discord user ID
• Discord guild/channel/message IDs
• organization/workspace/project context
• operation ID
• requested natural-language input
• proposed intent
• tool name
• sanitized arguments
• policy decision
• confirmation or approval actor
• execution result
• timestamps

Discord UX example

I’m going to update TASK-123, assign it to Sarah, and set the due date to May 15, 2026.
Reply `confirm` to proceed.

Acceptance criteria

• Confirmed execution uses the previously frozen plan.
• Denied and failed actions are audited.
• Secrets and sensitive values are redacted in audit payloads.
• Audit behavior follows the repo guidance for best-effort human-triggered service actions.

[michaelmwu]

Implemented in PR #264. Discord agent writes now return frozen confirmation plans, confirmation/cancel buttons execute the stored plan, backend confirmation is single-claim/re-authorized, and request/confirmation/execution outcomes are written through the audit path with sensitive payload handling.