Agent Workflow Notes

After a requested change is verified, commit it without waiting for a separate prompt unless the user explicitly asks not to commit.
Commit independent fixes separately. Before pushing, review the changed files and group commits by concern instead of bundling unrelated fixes together.
When the user asks to publish finished work, commit the intended fixes and push the current branch after verification passes.

Governance rotation notes

Rotating ContentRegistry.setVotingEngine alone is insufficient: escrows pin the engine at initialization and reject new work with "Stale engine" until a full replacement stack is deployed and rewired (QuestionRewardPoolEscrow, FeedbackBonusEscrow, FeedbackRegistry, fee creditor on FrontendRegistry, and X402 submitter escrow pointers as applicable). Those contracts expose no governed setVotingEngine; plan a coordinated redeploy and rewire, not a registry-only timelock action.
Treat engine migration as a coordinated governance runbook, not a single-timelock action.

RateLoop handoff images support JPG, PNG, and WEBP files up to 10 MB per image. Prefer 16:9 for newly generated public images; other ratios are allowed when useful. Do not downscale or recompress a readable mockup just because base64 output would be too large for the terminal or chat transcript.
When creating a browser handoff from local/generated images, prefer the file-backed CLI path: yarn workspace @rateloop/agents handoff --file <ask.json> --image <image.png>. It reads bytes from disk, computes sha256/sizeBytes, and stages large files through the handoff blob-upload route instead of squeezing base64 through a single JSON request.
If using MCP directly, pass image bytes from file-backed tooling or an SDK process. Avoid cat, print, or shelling base64 into the conversation as the transport layer.

Do not assume RateLoop is trying to make every oracle-adjacent path fully economically secured per snapshot on-chain. ClusterPayoutOracle payout roots are intentionally optimistic: globally bonded frontend operators publish public deterministic artifacts, challengers/auditors can recompute them during the challenge window, governance arbitrates challenged roots, and bad proposers can lose reputation, future fee income, and their FrontendRegistry LREP bond.
Treat ClusterPayoutOracle challenge bonds as anti-spam bonds, not payout-value coverage bonds. The default challenge bond is 5 USDC in atomic units (5_000_000).
Treat the 60-minute revealGracePeriod as an accepted product/security parameter, even for long human-duration blind phases. Do not raise it as an audit finding merely because it is shorter than a question's custom epoch duration.