Cache ERC-8004 agent data in users table

[realproject7]

Summary

Fixes #589

• Migration (00029): Adds 8 agent columns to users table — agent_id, agent_name, agent_description, agent_genre, agent_llm_model, agent_wallet, agent_owner, agent_registered_at with partial indexes
• API endpoints: POST /api/user/agent-register (upsert after registration) and POST /api/user/agent-update (update after management actions)
• DB persistence: AgentRegister persists after registration + wallet binding; AgentManage persists after URI update, wallet change, wallet unset
• DB-first detection: Agents page checks dbUser.agent_id before falling back to RPC; eliminates 3-6 RPC calls for known agents
• detectWriterType(): Queries DB first (agent_wallet or primary_address), RPC only for unknown wallets
• fetchAgentMetadata(): Returns cached DB data when available; auto-caches externally registered agents on first RPC lookup
• Type updates: lib/supabase.ts Row/Insert/Update types include new agent columns

Note: Migration has operator gate per issue spec — included in PR but rollout requires operator approval.

Test plan

• Build passes (tsc --noEmit clean)
• After registering an agent, users row has agent columns populated
• After updating URI in management, agent_name/agent_description etc. updated in DB
• After wallet change/unset, agent_wallet column updated
• Returning to /agents page loads from DB without RPC calls
• Externally registered agent (no DB record) detected via RPC fallback and cached
• Profile page shows agent metadata from DB cache
• Non-agent users unaffected

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
plotlink	Ignored		Mar 27, 2026 3:15pm

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The migration and DB cache plumbing are substantial, but two core cache paths still miss the owner-vs-agent-wallet cases that issue #589 requires.

Findings

• [high] The dashboard still uses the old RPC-only detection flow, so one of the main pages called out in #589 continues to make registry reads on every load instead of using the new cache.
  • File: src/components/AgentDashboard.tsx:13
  • Suggestion: mirror the DB-first lookup logic here as well, so known agents do not need agentIdByWallet / balanceOf / tokenOfOwnerByIndex / getAgentWallet on every dashboard visit.
• [high] The new DB-first lookup/update paths still key off verified_addresses / primary_address, not the new agent_wallet / agent_owner fields, so owner wallets for externally registered agents will miss the cache and management updates from the owner wallet can 404 silently.
  • File: src/app/agents/page.tsx:25
  • File: src/app/api/user/agent-update/route.ts:46
  • Suggestion: add a dedicated agent cache lookup by agent_wallet or agent_owner (or agent_id) and use it in both the /agents read path and the agent-update route.

Decision

Request changes. The PR is close, but these gaps mean the cache still does not reliably cover the owner/agent-wallet split or eliminate dashboard RPCs for known agents.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The dashboard is now DB-first, but the shared lookup/update matching still misses a common owner-wallet case, so the cache is not actually reliable yet for externally registered agents.

Findings

• [high] Owner wallets that already have a normal users row still bypass the cached agent row, so DB-first detection remains broken for that case and management writes can keep updating the wrong row.
  • File: lib/actions.ts:97
  • File: src/app/api/user/agent-update/route.ts:46
  • Suggestion: do not stop at the first verified_addresses / primary_address match. For agent-specific reads/updates, prefer a row with non-null agent_id, or add a dedicated helper keyed by agent_wallet / agent_owner / agent_id.

Decision

Request changes. Example failure: an externally registered agent is auto-cached under the bound agent-wallet row, then the owner later connects with an existing non-agent PlotLink row. getUserFromDB(owner) returns the non-agent row first, dbDetected stays false, RPC fallback is still required, and agent-update writes back to that non-agent row instead of the cached agent row.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The owner-row matching is improved, but the PR still does not implement the required auto-cache behavior for externally registered agents on first /agents/dashboard visit.

Findings

• [high] RPC fallback detects external agents on /agents and dashboard, but never persists that result to the new cache, so the explicit "first visit then cached" requirement in #589 is still unmet.
  • File: src/app/agents/page.tsx:72
  • File: src/components/AgentDashboard.tsx:79
  • Suggestion: when the DB lookup misses and RPC fallback identifies an owner or agent wallet, fetch the metadata once and write the agent row immediately so subsequent visits are truly DB-first.

Decision

Request changes. Right now an externally registered agent owner can connect to /agents, be detected via RPC, and still have no cached row afterward. That means the next visit repeats the RPC path instead of using the DB cache, which misses the core acceptance criterion.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The new fallback effect fixes caching for direct agent-wallet logins, but owner-wallet logins for externally registered agents still do not persist the DB cache on first visit.

Findings

• [high] Owner-wallet fallback still calls the wallet-only metadata path, so the cache write never happens when the connected wallet owns an agent NFT but is not the bound agent wallet.
  • File: src/app/agents/page.tsx:86
  • File: src/components/AgentDashboard.tsx:95
  • File: lib/actions.ts:55
  • File: lib/contracts/erc8004.ts:245
  • Suggestion: when RPC fallback identifies an owner via tokenOfOwnerByIndex, persist using that detected agentId and its bound wallet/owner data instead of calling fetchAgentMetadata(address) with the owner wallet address.

Decision

Issue #589 requires external agents to be detected via RPC on first visit and then cached. That still fails for owner-wallet sessions where agentIdByWallet(owner) returns 0, so I’m keeping this PR in request-changes state.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The new cacheAgentById() path fixes owner-wallet caching by agent ID, but it still persists unbound agents incorrectly when getAgentWallet() returns the zero address.

Findings

• [high] Externally registered agents with no bound wallet are cached with agent_wallet = 0x000...000, so subsequent DB-backed owner dashboard visits query storylines for the zero address instead of leaving the writer address on the owner wallet.
  • File: lib/contracts/erc8004.ts:320
  • File: lib/actions.ts:119
  • File: src/components/AgentDashboard.tsx:77
  • Suggestion: normalize the zero address to null when caching by agentId, or guard it before using cached agent_wallet in the dashboard.

Decision

Issue #589 requires the first RPC-detected visit to cache data so later visits work from DB. For externally registered agents whose bound wallet is unset, the first visit works but the cached path regresses the dashboard on later visits, so I’m keeping this PR in request-changes state.

[project7-interns]

Verdict: APPROVE

Summary

The owner-wallet fallback is now cached by agentId, and the zero-address bound-wallet case is normalized correctly so later DB-backed visits do not regress the dashboard.

Findings

• None.

Decision

This now satisfies issue #589’s first-visit-then-cached requirement for both bound agent wallets and owner-wallet sessions, including unbound external agents.