Research Hub Pack
Structured deep research with source validation.
# Install the Research Hub pack into .claude/agents/ + .claude/skills/ npx @frankx/acos install research # Or via the manual fallback (git + cp) git clone https://github.com/frankxai/agentic-creator-os.git tmp-acos cp -r tmp-acos/packs/research/agents/* .claude/agents/ cp -r tmp-acos/packs/research/skills/* .claude/skills/ rm -rf tmp-acos
9 specialists in this pack
Status badges reflect what’s shipped in the current catalog. In-progress slots ship as Frank writes them. Buyers get all updates as the pack evolves.
3-phase deep research (scoping/parallel/synthesis) with cross-reference validation (≥2 sources for high confidence) + AEO-optimized output (TL;DR ≤50 words, validated-claims table, FAQ-style H2s) + quality gates. Wraps /deepresearch. Pass 2 (2026-04-28).
3-mode router (scan/deep-dive/publish) across 3 domains (Generative AI, Systems & Performance, Personal Development). Cross-domain synthesis + content-opportunity surfacing + GEO-structured publication (pillar 2500-4000w, brief 800-1200w, thread 10-15, linkedin 1300ch, newsletter-section 400-600w). Wraps /research. Pass 2 (2026-04-28).
Weekly 5-section newsletter generator (Headline / Tools & Releases / Research Frontier / Frank's Take / The Build) with word-target ±15% per section + ≥3 citations/section + voice consistency check + carry-forward thread detection across past 4 weeks. Slack approval ready. Wraps /ai-architect-newsletter. Pass 2 (2026-04-28).
Frontier model launch tracker: ingest announcement + verify benchmarks (≥2 sources) + map capabilities + compute deltas-vs-prior (% lift, pricing change, capability additions) + Frank's recommendation. Updates data/models/registry.json. Wraps /new-model. Pass 2 (2026-04-28).
Progressive context refinement loop: broad retrieval → relevance scoring (1-5 scale) → query refinement (add high-relevance terms, exclude noise) → narrower retrieval → iterate until precision ≥0.8 OR plateau (3 consecutive no-improvement) OR max-iter (default 5). Wraps iterative-retrieval skill. Pass 2 (2026-04-28).
Build-vs-reuse gate (Haiku): 4-criteria evaluation (maintenance < 12mo / popularity ≥1k stars or ≥10k weekly DLs / license-permissive / fit ≥70%) → 3-recommendation classification (clear-reuse / partial-fit + gap analysis / build-recommended). Wraps search-first skill. Pass 2 (2026-04-28).
Multi-perspective reasoner (Opus): framing structure (core question + constraints + success + out-of-scope) + 3-perspective hard cap + trade-off matrix + framing-conflict detection (>50% divergence → no clean recommendation) + needs-scoping for under-specified problems. Wraps opus-extended-thinking skill. Pass 2 (2026-04-28).
7-flow-type Opus composer (deep-research/daily-scan/topic-dive/new-model/newsletter/source-validation/superintelligence) dispatching 7 P6 specialists + builtin Research Librarian with abort-on-validation-gate-fail + durable flow record at data/research/flows/. P6 top-level composer. Repurposed from /superintelligence slot. Pass 2 (2026-04-28).
Fact-check + source verification. DUPLICATE with P4 — ADR-002 proposes consolidation to Citation Auditor.
What you install
Compatible runtimes
Plain text artifacts. Drop the files in, your CLI does the rest.
Pre-order the pack
Paid packs launch with Stripe checkout in the next sprint. Join the waitlist and pay only when the pack ships at L83.