Restore BNB testnet agents and submission readiness

[jh1nresh]

Product outcome

Restore the four curated BSC Testnet Agent Studio integrations and prepare truthful main-track submission documentation. The old indexer dependency and noncanonical card URLs marked all agents unavailable; the old quote schema required a field absent from every real seller quote.

• Read pinned chain-97 ERC-8004 wallet and data-URI metadata directly; verify exact HTTP/A2A endpoints before fetching the canonical agent card.
• Verify actual seller v1 quotes: strict fields, sorted-key keccak256, EIP-191 signer, chain, identity, selector, payment targets and positive fee cap.
• Retain four real signed public fixtures and seeded property tests. Update HACKATHON.md without claiming funded execution, official adoption or mainnet readiness.
• Scope a Jayson 4.1.3 override to @solana/web3.js, which declares ^4.1.1. Remove the newly flagged stream-json dependency without changing the audit allowlist. Add actual SDK JSON-RPC success/error compatibility tests with an offline transport.

Demand proof/pricing hypothesis: N/A for this repair; existing third-party testnet fees and product scope are unchanged. Distribution target: existing AgentShack marketplace, pending release verification.

State and trust boundaries

No database migration, contract edits, key access, signer/role changes, settlement, cron/config changes, mainnet actions, token movement or external form submission. All transaction code remains chain 97 only. Unrelated AGENTS.md changes and instruction PR #5 are excluded. No UI layout/style change.

Verification

• Clean npm ci and npx prisma generate passed.
• npm run lint and npm run typecheck passed.
• npm test -- --run: 25 files, 181 tests passed, including 128-case property loops and real quote signature verification.
• npm run audit:runtime passed against the unchanged reviewed baseline: 31 affected packages (16 low, 13 moderate, 2 high, zero critical). This is not zero known vulnerabilities.
• forge build --sizes / forge test -vv: 172 passed; fuzz cases run 256 times each.
• npm run build passed. Remote CI run 34217393314 passed app and contracts for candidate d3a9a2c42675029c93a088f816724909c57c7606 against base 468e7986270fff192c530117656e532f6fbff62a. The main-only release-readiness job was correctly skipped on that PR run. PR Restore BNB testnet agents and submission readiness #6 subsequently merged as 12be25d7abfb7a068b3232be1c3e8e9342037e1f with a tree identical to the reviewed head; main CI is the merge verification record.
• Production-mode local HTTP smoke: market/setup 200, all four registry entries live, all four quote API calls 200, malformed/caller-endpoint inputs 400. After the dependency pin, repeated registry/quote checks passed on the committed candidate; local health confirmed its full SHA, bscTestnet and mainnetContractsReady=false. Not a wallet-funded end-to-end test.

Focused security receipt

Accepted defects: noncanonical endpoint/indexer availability mismatch; real signed quotes rejected by the old wallet-field schema; new runtime dependency audit regression.

The fix keeps server-owned endpoints, rejects redirects, limits card/seller streams to 100 KB, rejects unsupported metadata URIs, and fails closed on identity/RPC errors. Quote fixtures test valid signatures, changed targets/terms, wrong signer, invalid hashes, zero/oversized/over-cap fees and funding retry selection. A valid quote is not proof of job execution.

Dependency review: upstream 4.1.3...4.3.0 diff shows the stream parser replacement and browser-client callback-return refactors. The SDK imports jayson/lib/client/browser, not the TCP/TLS streaming parser. The scoped pin stays inside the SDK's declared range and is covered through Connection.getBalance using the real RPC client. The older TCP/TLS parser is not newly exposed by this application.

GHSA-528h-pc64-c93x affects stream-json path filters; a reachable application exploit was not reproduced. The dependency was removed instead of waived. Existing reviewed-baseline dependency risks remain.

Independent read-only code review PASS was completed in user-authorized task 01a080b6-e041-7bb1-81bd-6e9d45f97b2c against head d3a9a2c42675029c93a088f816724909c57c7606 and base 468e7986270fff192c530117656e532f6fbff62a. The separate checker read all 11 changed files and consumers, verified actual CI logs, ran 49 focused tests and 17 independent API handler probes, and read back all four live chain-97 identities/cards. No new blocking finding. Final live SHA recheck: 2026-09-08 11:15:16 UTC. This was not maker self-approval or a GitHub reviewer approval.

Retained pre-existing limitations: activation API validates the curated wallet without a fresh per-call registry read; retry protection is limited to already-confirmed funding within the same page session; runtime audit retains 31 affected packages / 2 high. No funded-hire, delivery, deployment or submission PASS is implied.

Delivery / human boundaries

Exact-head CI and independent code review passed, and the user authorized merge and branch cleanup. Submission and deployment readiness remain separate. Preserve existing Vercel aliases and require commit/chain-aware readback before claiming deployment. Do not change the old linked repository, production secrets or existing cron settings as part of this patch.

Production delivery readback still reports a50d2b11b308bf902f96435dde77b245d0046acc; no deployment PASS receipt was generated. The local readiness JSON records the passing CI checks only, not independent review or a release.

Remaining user/organizer inputs: buyer public address and wallet signatures for all four funded hires, registration confirmation or required personal fields/terms, and confirmation that a testnet-only entry meets main-track eligibility. No new jobs, delivery receipts or registration confirmation were fabricated.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-08T11:20:02.535354Z	d3a9a2c	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.