Skip to content

CHANGELOG.md

Latest commit

 

History

History
123 lines (102 loc) · 21.6 KB

CHANGELOG.md

File metadata and controls

123 lines (102 loc) · 21.6 KB

AFI Collaboration Log (Hackathon Journal)

This is the official human↔agent collaboration log for Synthesis. It is intentionally high-level and safe to publish.

Rules:

  • Every agent conversation must be logged the same day.
  • Every decision or action that changes the build direction must be logged.
  • Do not include secrets, API keys, private hashes, or internal transaction IDs.

2026-03-16

  • Reset the repo to start a fresh AFI MVP after deciding the previous project direction was not aligned.
  • Defined AFI as a behavior-first observability layer focused on x402 + Locus + Base signals.
  • Implemented a minimal backend for ingestion, normalization, and evidence packet export.
  • Implemented a minimal UI for listing interactions and downloading evidence packets.
  • Added a SQLite store for normalized events.
  • Wired Locus status/transactions + checkout endpoints for demo use.
  • Added x402 header capture helper and fixture-based integration tests.
  • Started dev servers and verified UI renders with an empty state.
  • Registered the AFI agent on Synthesis and stored credentials locally (secrets kept out of this log).
  • Implemented full AFI adapters (Locus, Base, EAS, PEAC), expanded evidence schema, added metrics endpoints, and rebuilt the UI with agent/counterparty/flow views.
  • Refactored server runtime to separate testable API logic from the network listener (server/index.ts + server/cli.ts) and added end-to-end handler tests.
  • Fixed POST /api/locus/ingest/transactions foreign-key failures by ensuring an interaction exists before inserting wallet snapshots.
  • Tightened typing + runtime validation (React JSX TS config, Zod-validated Locus balance responses, and safer header iteration for x402 client).
  • Added repo-wide quality gates: ESLint + TypeScript typecheck + Vitest coverage thresholds enforcing 100% lines/statements/functions/branches for both server + UI.
  • Expanded regression coverage across store normalization, adapters, metrics, and UI flows (interaction list, packet viewer, metrics loaders, and main entry mount).
  • Validated locally with npm run validate (lint + typecheck + coverage) passing cleanly.
  • Improved x402 ingestion to infer settlement tx hashes from PAYMENT-RESPONSE when txHash is omitted, enabling Base enrichment + evidence packets without manual duplication.
  • Added regression tests covering tx hash inference and settlement status derivation while keeping 100% coverage thresholds.
  • Revalidated with npm run validate (lint + typecheck + server/UI coverage) passing cleanly.
  • Correlated stored EAS attestations onto interaction packets (case-insensitive wallet/tx matching) and included receipts/attestations in agent evidence density.
  • Fixed local validation reliability under constrained/offline environments by (1) running Vitest with a repo-local temp dir to avoid macOS /var/folders ENOSPC failures and (2) swapping the SQLite backend to Node’s built-in node:sqlite to avoid native module ABI rebuild requirements.
  • Added regression coverage for SQLite transaction rollback and for nullish Locus transaction upserts; revalidated npm run validate with 100% lines/statements/functions/branches for server + UI.
  • Implemented spend-control compliance + fulfillment latency metrics (approval rate, allowance/max-tx compliance, receipt availability, and settlement latency) for both agent and counterparty profiles; exposed derived control facts on interaction packets and rendered them in the UI.
  • Hardened validation ergonomics by isolating server/UI coverage output directories (avoids shared coverage/.tmp races) and removing the unused native better-sqlite3 dependency; revalidated npm run validate with 100% lines/statements/functions/branches.
  • Tightened the “100% coverage + typechecked boundary” claim by removing the stale server/types.ts coverage exclusion, converting the type-only server schema module to server/types.d.ts, extending tsc --noEmit to check scripts/**/*.mjs, and updating the temp-dir helper imports to pass JS typechecking.
  • Simplified service inference normalization branches so the runtime model matches actual behavior, then revalidated with npm run validate passing cleanly (lint, typecheck, build, server coverage, UI coverage all at 100%).
  • Fixed AFI’s service-model blind spot by normalizing service alongside counterparty, preserving wrapped API provider/endpoint and x402 slug hints from Locus payloads, and upgrading the Flow Explorer to render wallet -> counterparty -> service paths instead of collapsing service sectors into one opaque label.
  • Added migration + regression coverage for the new interaction service column, URL/service inference during x402 ingestion, Locus passthrough parsing, and three-node flow rendering; revalidated npm run validate with build + lint + typecheck + 100% lines/statements/functions/branches for server + UI.

2026-03-17

  • Shipped the AFI portable-packet slice end to end: added a typed canonical packet builder + backend export route contract, normalized GET /api/interactions/:id and /api/interactions/:id/packet onto the same packet envelope, and persisted x402 transcript chronology into packet protocol evidence.

  • Strengthened evidence correlation and presentation: receipt lookup now includes settlement-hash correlation, packets expose nested protocol/evidence/correlations/provenance sections, and the UI renders structured transcript, settlement, receipt, attestation, and evidence-timeline panels instead of treating the packet as a raw dump.

  • Made explorer/profile workflows navigable from the packet surface: flow helpers and UI interactions now drive packet selection plus agent/counterparty profile loading from explorer nodes, and packet downloads now use the backend export route rather than client-side data: serialization.

  • Files/modules affected: server/packet.ts, server/index.ts, server/normalize.ts, server/store.ts, server/types.d.ts, src/app.tsx, package.json, tests/api.test.ts, tests/packet.test.ts, tests/ui/app.test.tsx.

  • Validation status: npm run validate passed cleanly on March 17, 2026 with lint green, tsc --noEmit green, build green, server coverage 1357/1357 statements / 345/345 functions / 1047/1047 branches, and UI coverage 115/115 statements / 42/42 functions / 177/177 branches.

  • Promoted AFI packet export from an internal-detail dump to a first-class canonical packet flow: the API now serves /api/interactions/:id/packet, the UI renders packet metadata + verifier references directly from that contract, and packet downloads come from the canonical export route.

  • Root cause fixed at source: the repo had been partially migrated to the packet contract without the shared server/packet.ts source-of-truth module, so typecheck failed and packet/API/UI consumers drifted. Added the missing builder, merged transcript-derived headers during normalization, correlated tx-hash-only receipts into packets, normalized the UI flow helpers so filtering/profile-loading paths stay testable, and fixed the flaky rate-limit regression test by starting from a known empty token bucket instead of timing a burn-down loop.

  • Files/modules affected: server/packet.ts, server/index.ts, server/normalize.ts, server/store.ts, server/types.d.ts, src/app.tsx, src/styles.css, tests/packet.test.ts, tests/api.test.ts, tests/http-client.test.ts, tests/normalize.test.ts, tests/store.test.ts, tests/ui/app.test.tsx, package.json, README.md.

  • Validation status: npm run validate passed cleanly on March 17, 2026 with lint green, tsc --noEmit green, the UI build green, and both server + UI coverage at 100% lines / 100% statements / 100% functions / 100% branches.

  • Closed the remaining AFI enrichment-slice validation regressions by fixing the coverage runner’s shared-temp-dir collision and by covering the last protocol-aware metrics + UI flow-label branches.

  • Root cause fixed at source: counterparty-share math now divides by the real interaction count, scripts/run-with-tmpdir.mjs allocates a unique temp workspace per run, and coverage scripts emit into dedicated JSON report directories so server/UI V8 artifacts do not collide.

  • Added regression coverage for protocol-category rollups, inbound/outbound USD transfer totals, numeric stored price values, missing tx targets during protocol labeling, protocol-name/service flow labels, and packet amount rendering when amountUSD is present.

  • Validation status: npm run validate passed cleanly on March 17, 2026 with lint green, tsc --noEmit green, build green, and server + UI coverage enforced at 100% lines / 100% statements / 100% functions / 100% branches.

  • Files/modules affected: server/metrics.ts, tests/metrics.test.ts, tests/ui/app.test.tsx, scripts/run-with-tmpdir.mjs, vitest.config.ts, vitest.ui.config.ts, package.json.

  • Restored repo-wide validation to green by closing coverage gaps introduced by the new x402 transcript + onchain metrics slice.

  • Added targeted regression tests for:

    • Base block timestamp parsing edge cases (hex + non-hex formats, zero timestamps).
    • x402 capture handshake branches (no signature, signature via retry headers).
    • GET /api/interactions/:id behavior for Locus interactions (x402 transcript omitted when protocol is not x402).
    • UI handshake/settlement render branches (authorized, recorded, failed, settled, and fallback states).

  • Revalidated locally with npm run validate passing cleanly with 100% lines/statements/functions/branches for server + UI.

  • Replaced AFI’s raw x402-header-only packet view with a typed x402 transcript: added challenge/authorization/settlement decoders, captured two-step handshake state in the x402 client, exposed canonical x402 + correlated Base transaction sections on GET /api/interactions/:id, and rendered those protocol states directly in the packet panel.

  • Fixed a repo-level metrics contract regression by implementing Store.listWalletsByCounterparty, aligning async metrics endpoint tests with the current API, and removing dead branch logic in packet/onchain summaries that was blocking the repo-wide 100% coverage gate.

  • Added regression coverage for typed x402 decoding, handshake capture, canonical packet normalization, locus-vs-x402 packet API responses, Base timestamp parsing, metrics/onchain edge cases, and all UI handshake render states (complete, authorized, challenge-only, settled, not-captured, plus settlement success/failed/recorded).

  • Revalidated locally with npm run validate; lint passed, tsc --noEmit passed, the UI build passed, and both server + UI finished at 100% lines/statements/functions/branches.

  • Extended Base enrichment to attach a confirmedAt timestamp to confirmed transactions (via block timestamp lookup) and expanded agent metrics to include onchain transaction + token-transfer behavior (counterparty concentration, token diversity, inbound/outbound counts), rendered in the UI.

  • Upgraded the x402 client to capture a full paid-call transcript (402 challenge → signature → settlement) via an optional onPaymentRequired callback and added typed x402 packet decoding helpers; revalidated npm run validate with 100% coverage.

  • Closed AFI’s live-Locus observability gap by persisting wrapped API calls, Locus-routed x402 calls, checkout payments, and direct send actions into the interaction/evidence store immediately instead of waiting for the later transaction-sync endpoint.

  • Root cause fixed at source: the live Locus routes were proxy-only, so first-party paid actions disappeared from the AFI behavior graph unless /api/locus/ingest/transactions ran afterward; thin upstream { ok: true } responses also risked collapsing multiple actions onto the same synthetic interaction id. Added a shared live-action capture path, enriched best-effort wallet snapshots with explicit warning logs on snapshot failure, and salted normalization ids so repeated thin responses remain distinct interactions.

  • Added regression coverage for immediate live-action persistence, AFI response metadata, repeated thin-response collision avoidance, snapshot-failure observability, and live wallet snapshot fallbacks when Locus omits address/balance fields.

  • Files/modules affected: server/index.ts, server/normalize.ts, tests/api.test.ts.

  • Validation status: npm run validate passed cleanly on March 17, 2026 with lint green, tsc --noEmit green, build green, server coverage 1399/1399 statements / 351/351 functions / 1122/1122 branches, and UI coverage 115/115 statements / 42/42 functions / 177/177 branches.

2026-03-18

  • Added a wallet-scoped AFI behavior-model slice end to end: the backend now serves GET /api/models/wallet/:wallet through the shared cache layer, server/models.ts produces deterministic anomaly/cluster/flag outputs from live normalized AFI facts, and the cockpit renders a compact Behavior Model panel that loads alongside wallet drills from packet selection and flow exploration.

  • Root cause fixed at source: the repo already had normalized behavioral facts in SQLite plus metrics/read-model helpers, but server/models.ts still exposed an offline Python-training contract instead of a runtime wallet behavior read model. Consolidated onto one typed runtime contract, removed the stale duplicate scorer path, fixed flag severity thresholds that previously made every flag render as low, and corrected the sparse-activity burst calculation so a single recent interaction no longer forces wallets into bursty cohorts.

  • Files/modules affected: README.md, CHANGELOG.md, server/index.ts, server/models.ts, server/query-cache.ts, server/types.d.ts, src/app.tsx, src/styles.css, tests/api.test.ts, tests/models.test.ts, tests/query-cache.test.ts, tests/ui/app.test.tsx.

  • Added regression coverage for zero-signal wallets, elevated/bursty/high-value wallet cohorts, wallet-model cache reuse + invalidation, API route wiring, UI behavior-panel success/failure states, and the wallet drill path from flow explorer + packet selection.

  • Validation status: npm run validate passed cleanly on March 18, 2026 with lint green, tsc --noEmit green, build green, and server + UI coverage at 100% lines / 100% statements / 100% functions / 100% branches.

  • Added an AFI cockpit dashboard analytics slice end to end: the backend now exposes a cache-backed /api/metrics/dashboard contract, the DuckDB query layer serves overview totals/series/concentration/reliability/recent-activity aggregates, and the UI overview cards + charts consume that shared contract instead of deriving a partially stale dashboard client-side.

  • Root cause fixed at source: the repo had packet/profile/enrichment surfaces, but no single dashboard source of truth. Dashboard charts were being assembled ad hoc in the browser, the new analytics route bypassed QueryCache, and fallback states for missing analytics labels / failed analytics loads were inconsistent. Centralized dashboard aggregation in server/duckdb-queries.ts, routed it through server/query-cache.ts, wired the runtime API to that cache-backed path, simplified dead UI-only dashboard helper branches, and added deterministic fallback handling for blank labels and analytics fetch failures.

  • Files/modules affected: server/duckdb-queries.ts, server/index.ts, server/query-cache.ts, server/store.ts, src/app.tsx, src/styles.css, tests/api.test.ts, tests/duckdb-queries.test.ts, tests/query-cache.test.ts, tests/ui/app.test.tsx, tests/ui/main.test.tsx.

  • Added regression coverage for dashboard route query parsing, cache reuse + invalidation for dashboard analytics, shared-connection DuckDB overview queries, dashboard fallback rendering when analytics fail, blank-label analytics normalization to unknown, and UI entrypoint mounting.

  • Validation status: npm run validate passed cleanly on March 18, 2026 with lint green, tsc --noEmit green, build green, and server + UI coverage at 100.00% lines / 100.00% statements / 100.00% functions / 100.00% branches.

  • Integrated cache-aware AFI serving into the real server runtime, added admin/export/query-cache endpoints on server/index.ts, and removed the obsolete server/index-with-cache.ts sidecar so the repo no longer carries a dead integration path alongside the active API.

  • Root cause fixed at source: repo-wide validation had drifted because committed/generated .js artifacts were shadowing TypeScript sources, the QueryCache/parquet-export slice was only partially wired into the main runtime, and the portable packet builder was serializing evidence as an array with attached properties, which dropped timeline/receipt/attestation data when converted to JSON. Ignored generated JS, deleted the obsolete cache guide file, wired cache invalidation + flow aggregate/parquet admin routes through the main API, and corrected server/packet.ts to emit a plain object evidence contract.

  • Strengthened the AFI serving/runtime boundary by vendoring the Parquet runtime behind server/parquet-lib.ts, tightening SQLite/parquet query ordering + export typing, replacing Python-model stubs with structured runtime errors, removing redundant impossible export branches, and fixing counterparty feature extraction so null-wallet cohorts return 0 concentration instead of -Infinity.

  • Files/modules affected: .gitignore, eslint.config.js, package.json, server/index.ts, server/index-with-cache.ts (deleted), server/packet.ts, server/parquet-lib.ts, server/parquet-export.ts, server/query-cache.ts, server/store.ts, server/duckdb-queries.ts, server/models.ts, tests/api.test.ts, tests/config.test.ts, tests/duckdb-queries.test.ts, tests/jobs.test.ts, tests/models.test.ts, tests/packet.test.ts, tests/parquet-export.test.ts, tests/performance-benchmark.test.ts, tests/query-cache.test.ts, tests/store.test.ts, vendor/parquetjs-lite/**.

  • Added regression coverage for packet JSON serialization, cache invalidation + flow aggregate route wiring, parquet export success/failure placeholders, nullable export normalization, wallet-scoped parquet artifacts, non-Error query failures, null-wallet/null-counterparty feature extraction, Python runtime error contracts, disabled/enabled parquet store helpers, and cache aggregation edge cases.

  • Validation status: npm run validate passed cleanly on March 18, 2026 with lint green, tsc --noEmit green, build green, server coverage 100% lines / 100% statements / 100% functions / 100% branches, and UI coverage 100% lines / 100% statements / 100% functions / 100% branches.

  • Hardened the API request-validation boundary for AFI ingestion by validating raw unknown request bodies in server/index.ts, simplifying x402 header validation to the actual extractor contract, and normalizing legacy walletSnapshot.approvals_required inputs (true/false and 1/0) onto the domain boolean model instead of letting request typing drift from storage/UI semantics.

  • Root cause fixed at source: the ingest boundary still assumed object-shaped request bodies and relied on a record-union plus cast for x402 headers, while the new validation slice had also narrowed approvals_required away from the persisted/read-model contract. That combination made the HTTP edge less truthful than the AFI domain model and left the coverage gate red on an untested PEAC failure path.

  • Added regression coverage for legacy numeric approval flags and the invalid_receipt PEAC rejection branch so the request boundary now proves both backward-compat normalization and the explicit failure policy under the repo-wide 100% coverage gate.

  • Files/modules affected: server/index.ts, tests/api.test.ts.

  • Validation status: npm run validate passed cleanly on March 18, 2026 with lint green, tsc --noEmit green, build green, and server + UI coverage at 100% lines / 100% statements / 100% functions / 100% branches.

  • Added protocol-attribution provenance as a first-class AFI evidence surface: protocol label enrichment now resolves through a shared backend path, exposes source/metadata/timestamp on enriched interactions and packet exports, and supports deterministic on-demand refresh for a selected interaction.

  • Root cause fixed at source: protocol labels were already persisted with source, metadata, and created_at, but that provenance was dropped by the read model and packet export path, while refresh was only available through the background Dune job. Centralized contract/query-address resolution in server/protocol-labels.ts, reused it across metrics/jobs/packet export/API, and tightened the UI refresh handler so it operates on an explicit interaction id instead of nullable component state.

  • Files/modules affected: server/protocol-labels.ts, server/index.ts, server/jobs.ts, server/metrics.ts, server/packet.ts, server/types.d.ts, src/app.tsx, tests/api.test.ts, tests/jobs.test.ts, tests/packet.test.ts, tests/store.test.ts, tests/protocol-labels.test.ts, tests/ui/app.test.tsx.

  • Added regression coverage for protocol-label provenance round-trips, interaction-scoped enrichment success/failure cases, contract-address query fallback, deterministic Dune activity matching, packet/export attribution wiring, and UI refresh success/error/default-message branches.

  • Validation status: npm run validate passed cleanly on March 18, 2026 with lint green, tsc --noEmit green, build green, server coverage 1463/1463 statements / 365/365 functions / 1164/1164 branches, and UI coverage 129/129 statements / 47/47 functions / 195/195 branches.

2026-03-15

  • Confirmed Synthesis registration flow and project submission status.
  • Decided on AFI as the new build direction (behavior-first observability).

2026-03-14

  • Documented Synthesis constraints and possible integrations for paid-agent observability.

2026-03-13

  • Initial hackathon planning and exploration of fit across Synthesis themes.