fix(registry): canonicalize agent_url on the registered path (#3573)

[brandonling27]

Summary

Closes #3573.

The crawler/discovered path already canonicalized agent_url via canonicalizeAgentUrl (write site in crawler.ts:reconcileLegacyAdagentsAgents, read-side SQL safety net in federated-index-db.ts). The member-profile registered path did not — POST/PATCH/DELETE on /api/me/agents, the bulk PUT /api/me/member-profile, and Addie's save_agent/remove_saved_agent all wrote raw URLs into member_profiles.agents JSONB and agent_registry_metadata. Two writes for the same logical agent differing only in case or trailing slash landed as separate rows, and the JS map joins in FederatedIndexService.lookupDomain / listAllAgents enriched on raw string equality — so the member badge silently dropped off any discovered authorization whose URL differed from the registered one only in case or trailing slash.

This change is forward-only, application-layer only (no DB migration, per #3573's "out of scope"). Canonicalize at every member-side write boundary; canonicalize both the map key and the lookup key in every JS read site so legacy non-canonical rows continue to match canonical inputs until the member next saves.

Write sites (canonicalize, reject ?/#, 400 on null)

• POST/PATCH/DELETE in server/src/routes/member-agents.ts (handler boundary, mirroring the precedent at routes/registry-api.ts:7370-7377)
• Bulk PUT /api/me/member-profile in server/src/routes/member-profiles.ts
• Addie save_agent and remove_saved_agent in server/src/addie/mcp/member-tools.ts
• Defense-in-depth in applyMemberAgentMutation's agent_registry_metadata seed loop so any future writer that forgets is still caught

Read sites (canonical key with ?? raw fallback)

• FederatedIndexService.listAllAgents map key
• FederatedIndexService.lookupDomain map key plus the two lookup sites (auth.agent_url and claim.discovered_by_agent)
• FederatedIndexService.getAllAgentDomainPairs (bulk snapshot reader)

Tests

• server/tests/integration/member-agents-api.test.ts — POST collapses mixed-case/trailing-slash to canonical; idempotency across forms yields one JSONB row + one agent_registry_metadata row; PATCH/DELETE match canonical rows from non-canonical url-encoded paths; PATCH accepts case/slash-only body.url differences; 400 on query/fragment/wildcard.
• server/tests/integration/registry-crawler-cache.test.ts — Pin 1: registered https://agent.example/ + discovered https://agent.example collapse with member-badge enrichment. Pin 2: scheme mismatch (http:// vs https://) intentionally does NOT collapse — different security posture per fix(registry): canonicalize agent_url before registered/discovered collapse (trailing slash + scheme) #3573.

Trade-offs

• Path case is lowercased (existing canonicalizeAgentUrl behavior) rather than scheme+host-only as fix(registry): canonicalize agent_url before registered/discovered collapse (trailing slash + scheme) #3573 phrased it. Chosen for consistency with the discovered side; switching both sides to a stricter canonicalizer would have a much larger blast radius (every existing caller in registry-api.ts, the SQL safety net in federated-index-db.ts, etc.).
• No backfill. Existing non-canonical rows in member_profiles.agents and agent_registry_metadata persist until the member next saves. Read-side canonicalization in federated-index.ts papers over this for the registry surface; direct readers of agent_registry_metadata (compliance heartbeat, lifecycle dashboards) will continue to see un-collapsed legacy rows until they're rewritten. The issue explicitly defers cleanup ("write a one-off cleanup if a sweep finds collisions").
• brand.json will publish the canonical form for any public agent whose stored URL gets rewritten on next save. Believed harmless (same URL, normalized) but worth flagging for downstream consumers.

Test plan

• npm run typecheck — passes locally
• DATABASE_URL=… npx vitest run server/tests/integration/member-agents-api.test.ts — new canonicalization cases
• DATABASE_URL=… npx vitest run server/tests/integration/registry-crawler-cache.test.ts — Pin 1 and Pin 2
• Manual: POST https://Example.com/Agent/ to /api/me/agents; confirm response and DB row both show https://example.com/agent. Crawl a fixture authorizing https://EXAMPLE.com/agent; confirm lookupDomain returns the member badge on the discovered authorization.

🤖 Generated with Claude Code

[aao-ipr-bot]

IPR Policy Agreement Required

@brandonling27 — thanks for the contribution. Before this PR can be merged, the AgenticAdvertising.Org IPR Policy requires your agreement.

To agree, post a new comment on this PR with the exact phrase:

I have read the IPR Policy

Your signature is recorded once and covers all contributions to AAO repositories. See signatures/README.md for what gets recorded and why.

[brandonling27]

I have read the IPR Policy

[aao-ipr-bot]

IPR Policy — signed

Thanks, @brandonling27. Your agreement to the IPR Policy is recorded at signatures/ipr-signatures.json and applies to all AAO repositories.