Execution Boundary
The agent may draft assignments, score contribution, summarize evidence, and stage reward batches. It should not execute irreversible transfers or public claims without explicit operator approval.
Risk model
Wallet actions stay behind explicit approval.
GuildForge avoids guaranteed financial language and treats settlement as a staged, reviewable workflow before any irreversible action.
Control Policy
Minimum authorization scopes for every wallet action
Human approval before settlement or public announcement
Revocable permissions with clear operator ownership
Editable AI output with linked source data
Dispute windows before contribution scores become final
No guaranteed profit, yield, or financial outcome claims
Operator Disclosure
Chain actions, reward programs, auctions, and token distributions must include risk language, editable summaries, and accessible records for disputes and treasury review.