Dashboard/Algolia-Central (Second-Brain)/spectrum/tasks.md
Algolia-Central (Second-Brain)

spectrum/tasks.md

Plan (the target state — % complete is measured against THIS)

Track A (the real offer-caching-race fix) is SHIPPED — done, deployed, verified live. This session did doc cleanup + brainstorming only, no code. Remaining work: get the 3 pending decisions from Arijit (mandate-guard, AC2 roster/topology/dead-code), then build the approved empty-state spec, then the eval/calibration pass.

Tasks

  • [x] Build real (non-faked) E2E test suite for the pure-orchestrator design · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Run T1 (baseline full text), T5 (offer precision, 10 real Qs), T3 (real chaining), T4 (specialist real answer) · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Fix test script's null === false scoring bug in e2e-orchestrator-validation.mjs · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Full development-loop build (A1-A9): dedicated classifier agent replacing racy config.suggestions offer signal · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Kill pure-orchestrator for ACS (provably redundant once classifier exists); keep shelved spec for AC2 · owner:claude/arijit · added:2026-07-10 · done:2026-07-10
  • [x] Live acceptance-gate probe (7-pair) — found + fixed classifier over-triggering on design questions before flip · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Live flip + deploy: ACS-classifier-neural created, Generic/Technical IDs unchanged · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Fix: classifier offering deep-dive on an empty Generic answer · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Fix: numbered-list rendering (real fix — line-run grouping, not whole-block classification) · owner:claude · added:2026-07-10 · done:2026-07-10
  • [x] Scope the sample-questions UI suggestion via the brainstorming skill — root-caused (chips look inert, no label) and wrote approved spec · owner:claude · added:2026-07-10 · done:2026-07-12
  • [x] Archive stale docs/plans/2026-07-09-revert-to-tool-call-and-port-to-ac2.md · owner:claude · added:2026-07-10 · done:2026-07-12
  • [x] Reconcile docs/plans/2026-07-10-reconciled-handoff-architecture-build.md — deleted+rewrote as as-shipped summary + scoped AC2-port task list · owner:claude · added:2026-07-10 · done:2026-07-12
  • [ ] Build the empty-state sample-question nudge per approved spec (docs/superpowers/specs/2026-07-10-empty-state-sample-question-nudge-design.md) — copy label + promoted resting-style chips · owner:claude · added:2026-07-12
  • [ ] Run T6 — real browser, network tab, live click-through of the (now-shelved) 3-agent orchestrator chain · owner:claude · added:2026-07-10 · blocked-on:only relevant if AC2 revives the orchestrator design
  • [ ] Phase 1 (port to AC2) — 3 decisions pending Arijit before code: agent roster collapse (Gen3's 3 personas vs ACS's 2+classifier pattern), browser-direct vs backend-mediated topology, dead-code disposition (archive via git tag vs delete) · owner:arijit(decide)+claude(build) · added:2026-07-09 · re-scoped:2026-07-12
  • [ ] Resolve mandate-guard.sh vs. Claude Code auto-mode classifier conflict — Arijit's decision 2026-07-12: redesign the hook (real permission-prompt, not a settings.json allow-rule) · owner:claude · added:2026-07-09 · decided:2026-07-12
  • [ ] react-instantsearch <Chat> widget swap — not started · owner:claude · added:2026-07-08
  • [ ] Full 100-question eval — never run to completion; explicitly deferred again 2026-07-12 · owner:claude · added:2026-07-08
  • [ ] Judge calibration (P2b) — never run, all eval scores directional only; explicitly deferred again 2026-07-12 · owner:claude · added:2026-07-03