Dashboard/Algolia-Search-Audit/TODO.md
Algolia-Search-Audit

TODO.md

TODO — Algolia Search Audit Refactor

Single source of truth for task tracking. Updated by Claude across sessions. Dates are absolute (never relative).

Active

  • [ ] Replace SimilarWeb MCP API with browser-login screenshot approach — added 2026-04-14
  • Context: SimilarWeb API (mcp-remote) exhausted quota (remaining_hits: 0) and even when working only returned 1/14 endpoints. Manual screenshot capture from the SimilarWeb web UI yielded dramatically richer data: referrals, outgoing traffic, paid keywords, channel mix, competitor comparison — none of which the API returned. Proposed new approach: skill launches browser, logs into similarweb.com with Arijit's credentials, navigates to each section (Traffic & Engagement, Marketing Channels, Geography, Referrals, Outgoing), takes screenshots, extracts structured data from screenshots via vision. This matches exactly what produced the verified FACT data on 2026-04-14 for Nike and OTC.

  • Action needed: Design browser-login skill variant for algolia-intel-traffic. Store SimilarWeb credentials securely (env var or keychain). Map the 6 SimilarWeb UI sections to the 11-endpoint schema the skill currently expects.

  • [ ] Fix report blind spot — referral, outgoing, paid search data not surfaced — added 2026-04-14

  • Context: The algolia-audit-report skill and all downstream deliverables (deck, AE playbook, leave-behind) currently ignore referral traffic, outgoing traffic patterns, and paid search channel intelligence captured in 03-traffic-data.md. These contain some of the most powerful Algolia pitch signals discovered so far: (1) Google outbound traffic = quantified search abandonment, (2) ChatGPT referral growth = AI discovery behavior shift, (3) Paid keyword conquest (OTC bidding on "party city") = explicit competitive signals, (4) Capital One Shopping referrals = price-sensitive audience profile. None of this appears in the report scoring, deck narrative, or AE talking points.

  • Action needed: Audit the algolia-audit-report skill. Add a "Traffic Intelligence" section to the report template that explicitly surfaces: top referral sources + MoM trend, outgoing traffic search abandonment signal (% going to google.com), paid keyword intelligence, AI chatbot referral share. Wire these signals into AE playbook talking points and the pitch narrative.

  • [ ] Write intel-company module spec — added 2026-04-08

  • Context: Gap analysis done. Awaiting Arijit's feedback on BuiltWith keywords-api (keep or drop?). Then write spec to Modules/intel-company.md using Module-Spec-Template.

  • [ ] Per-skill design review: Wave 1 remaining (1B-1L) — added 2026-04-08

  • Context: 10 modules remaining in Wave 1. For each: read current skill, gap analysis vs PRISM, propose changes, write module spec.

  • [ ] Per-skill design review: Wave 2 (intel-queries) — added 2026-04-08

  • Context: Query generation module. Needs JSON output schema, typed query arrays.

  • [ ] Per-skill design review: Layer 2 (audit-browser) — added 2026-04-08

  • Context: Playwright-based browser testing. Needs finding schema, screenshot naming convention.

  • [ ] Per-skill design review: Layer 3 synthesis (4 modules) — added 2026-04-08

  • Context: synth-business-case, synth-sales-plays, audit-report, campaign-abx.

  • [ ] Per-skill design review: Layer 4 quality (2 modules) — added 2026-04-08

  • Context: audit-factcheck (mechanical + LLM split), audit-eval.

  • [ ] Design new insights-engine module — added 2026-04-08

  • Context: Cross-audit benchmarking. Exists in PRISM, no skill equivalent. Needs spec from scratch.

  • [ ] Copy algolia-customer-evidence.json to vault Data/ folder — added 2026-04-08

  • Context: Built from spreadsheet (866 customers, 333 quotes, 81 proof points). Currently at rc2-algolia/data/. Needs canonical vault copy.

Blocked

  • [ ] Answer Decision 2: Skill files vs prompts — blocked since 2026-04-08

  • Blocked by: Needs Arijit's decision. Keep skill files for manual use + separate prompts/*.txt for harness? Or consolidate?

  • [ ] Answer Decision 3: Scope / sprint strategy — blocked since 2026-04-08

  • Blocked by: Needs Arijit's decision. One big plan or sprint-based?

  • [ ] Write comprehensive implementation plan — blocked since 2026-04-08

  • Blocked by: All module specs must be written first. Decisions 2 and 3 must be answered.

Completed

  • [x] Read Anthropic agent harness engineering blogs — done 2026-04-08
  • [x] Write AgentHarnessPatterns.md to vault — done 2026-04-08
  • [x] Read and assess all 20 skills — done 2026-04-08
  • [x] Create vault project docs (Index, Module-Catalog, Known-Issues, Refactor-Architecture) — done 2026-04-08
  • [x] Build algolia-customer-evidence.json from spreadsheet — done 2026-04-08
  • [x] Read PRISM foundation docs (Module-Spec-Template, Patterns-For-Skills, Module-Catalog) — done 2026-04-08
  • [x] intel-company gap analysis (current vs PRISM) — done 2026-04-08
  • [x] Decision 1: Static customer dataset — done 2026-04-08
  • Resolved: Built from CustomerEvidence-Algolia.xlsx instead of scraping algolia.com/customers.