All skills
simota avatar

/nexus

@c805268
by shingo imotasimota/agent-skills85 stars
15

Orchestrating multi-specialist task chains and scope-adaptive product delivery: classifies intent, selects and executes the minimum viable chain, aggregates results, and verifies acceptance criteria. For multi-domain tasks, build-first delivery, and product lifecycle execution.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/nexus

This session only. Nothing lands on disk.

referenceventure-recipe.md

≈6.3k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Nexus Venture Recipe Reference

Purpose: Generate a comprehensive, cross-functional business documentation package from a single business idea — research → product spine → ~11 parallel documentation tracks → overview synthesis → traceability/validation → multi-format file tree + zip. The package is consumable by founders, PMs, designers, engineers, marketers, sales, investors, QA, and legal without further translation. Read when: User invokes /nexus venture, or asks for a complete business/product documentation set ("business plan package", "MVP dossier", "startup documentation set", "investor material bundle") that spans planning, branding, product, UX, LP, marketing, tech, AI policy, legal, testing, PM, mock data, and assets together.

venture is the startup preset of the generalized package recipe. /nexus venture ≡ /nexus package domain=startup. The shared engine (domain-agnostic Phase 0-6) and the full domain-preset registry live in reference/package-recipe.md; this file is the startup preset's detailed 14-directory blueprint and per-file → agent mapping. Read this file for startup detail, package-recipe.md for the engine and other domains.

Generated-content language: the documents written into the package follow the CLI output-language config (settings.json language). This reference file and all recipe instructions are English; the produced package is in the user's configured output language. File names, IDs, schema keys, and code stay English.

Contents

  • Overview
  • Invocation Modes
  • When to Use Venture
  • Depth Tiers and Mode Overlays
  • Topology
  • Phase 0: Framing
  • Phase Contracts (1 → 6)
  • Traceability Anchor (the core design constraint)
  • Directory → File → Agent Mapping (14 directories)
  • Output and Packaging
  • Conditional Inclusion
  • AUTORUN Chain Template
  • Failure Escalation

Overview

Venture is a document-generation fan-out, not an implementation loop. It differs from apex (which ships working code) by producing a structured multi-format file tree — Markdown, CSV, JSON, YAML, SQL, HTML/CSS, and Mermaid — packaged as a downloadable zip. The single hard design constraint is traceability: a canonical feature_id set (F-001, F-002, …) is fixed at Phase 2 and propagated to every downstream track so that user stories, acceptance criteria, test cases (TC-001…), and backlog items (BL-001…) all reference the same features without drift.

The chain is depth-scaled and mode-overlaid. A lite run produces a lean prototype dossier with ~6 agents; a full run fans out to all 14 directories with ~24-28 agents. Mode overlays (mvp-dev / fundraising / b2b-saas / b2c-growth / ai-product) bias which tracks get depth, not which tracks exist.

Venture is not a default recipe. It is opt-in for the "I have an idea, give me everything a team needs to act on it" request. Confirm before launching at full depth (24+ agents).

Invocation Modes

Form Behavior
/nexus venture <business idea + optional conditions> Standard. Parse idea + optional condition fields, apply default mode (mvp-dev + business-prep) and inferred depth, run Phase 0-6.
/nexus venture (no idea) Escalate: a business idea is mandatory. Ask for the idea (clarify carve-out) — do not auto-invent a venture.
/nexus venture depth=<lite|mvp|raise|full> mode=<...> idea="<X>" Explicit depth + mode overlay. mode accepts comma-separated overlays.

Optional condition fields are parsed when present and otherwise filled with documented reasonable assumptions written into 00_overview/assumptions.md: target / region / platform / phase / revenue_model / team / tech_pref / budget / launch_timing / tone / depth.

When to Use Venture

Use Venture when the request matches at least 3 of:

  • A business/product idea exists but documentation does not — greenfield or pre-MVP.
  • Output must serve multiple roles at once (Biz + Dev + Design + Marketing + Legal + QA).
  • Multiple file formats are required (not just Markdown — CSV/JSON/YAML/SQL/HTML/CSS).
  • A single source of truth with cross-document consistency is needed (feature_id traceability).
  • The deliverable is a package to hand off, not a single artifact.
  • Investor / decision-maker summary material is part of the ask.

Route elsewhere when the task is:

  • A single PRD or spec → scribe[unified] / scribe direct
  • A single landing page → funnel (or funnel[premium] for premium LP)
  • Market/competitor research only → field / compete direct
  • Brand system only → vision / muse direct
  • Working feature implementation → apex or feature recipe
  • Must-have / killer feature decision → essential / killer recipe
  • Deep multi-source research report → deep-research

Depth Tiers and Mode Overlays

Depth controls fan-out breadth. Inferred from the depth condition field, or explicit depth=. Tier semantics and cost bands are the engine's (reference/package-recipe.md § Cost and Latency Profile); the table below is the startup delta — it renames the default tier (mvp, not standard), binds each tier to startup directories, and narrows the agent counts, which are canonical for venture where they differ from the engine's.

Depth Directories produced Approx agents Use
lite (lightweight prototype) 00, 01(lite), 03, 04(lite), 05 6-8 Fast concept validation, hackathon, idea triage
mvp (default — MVP build + business prep) 00-13, lean 14-18 The standard full package
raise (fundraising) 00-13, research/overview/marketing/KPI deepened 16-20 Fundraising-grade
full (full commercialization) 00-13, all tracks deep + void/canon[regulatory]/crypt 24-28 Production business build — confirm before launch

Venture is not free. Budget guardrails (Nexus chain confirmation for 5+ agent chains, full-depth confirmation, no-secrets package scrub) are enforced. For repeated ventures with a stable house style, propose a Sigil-generated project skill to amortize the chain design cost.

Mode overlays bias depth allocation. Multiple may combine; they reweight tracks, never remove the core 14.

Overlay Tracks deepened
mvp-dev (default) 03 product, 04 ux, 05 lp, 07 tech, 10 testing
fundraising 00 overview, 01 research, 06 marketing (GTM/pricing), 03 KPI tree, one_page_pitch
b2b-saas 07 security/auth, 09 legal/canon[regulatory], 11 PM (SLA/sales), admin-panel screens in 04
b2c-growth 02 brand, 05 lp, 06 SEO/social/onboarding/retention, 04 onboarding_flow
ai-product 08 ai_policy (full), 07 AI/LLM stack, 10 ai_evaluation_cases, human-review workflow

Topology

Phase 0          Phase 1                Phase 2 [BARRIER]      Phase 3 [11 tracks, waves]         Phase 4          Phase 5              Phase 6
[Framing]        [Research]             [Product Spine]        (feature_id-bound; see Phase        [Overview]       [Integrate+Validate] [Package]
┌────────────┐   ┌──────────────────┐   ┌──────────────────┐    Contracts § Phase 3 table below)  ┌───────────┐    ┌────────────────┐   ┌──────────┐
│ parse idea │──▶│ field+compete    │──▶│ scribe[unified]+spark     │───▶┌────────────────────────────┐───▶│ spark     │───▶│ attest/judge   │──▶│ write    │
│ +mode/depth│   │ ‖ echo[demand]+cast      │   │ +rank+pulse      │    │ 02,04-13 parallel doc tracks│    │ +scribe   │    │ traceability   │   │ tree+zip │
│ +clarify≤3 │   │ (web-grounded or │   │ ═══ F-001… +     │    └────────────────────────────┘    │ +magi     │    │ +manifest+lint │   │ +report  │
└────────────┘   │  research_todo)  │   │  MoSCoW FIXED ═══│                                      └───────────┘    └────────────────┘   └──────────┘
                 └──────────────────┘   └──────────────────┘

Hub-and-spoke is preserved: Nexus is the only top-level orchestrator. Phase 2 is a hard barrier — no Phase 3 track may start until the canonical feature_id + MoSCoW table exists, because every track (testing, PM, KPI, LP) references it. Phase 3 tracks are mutually independent (separate output files, no shared mutable state) and reconverge only at Phase 5 validation. With 11 tracks, Phase 3 exceeds the ≤7-per-hub cap and is run as waves (or a native Dynamic Workflow); group dependent file-writers so no two agents write the same file.

Phase 0: Framing

As the engine's Phase 0 (reference/package-recipe.md § Shared Engine) — parse + resolve depth/mode + WebSearch availability check (01_research/references.md if grounded, else 01_research/research_todo.md) + ≤3 clarify gate, each with a stated fallback assumption so work proceeds without a reply. Clarify only on: business domain unidentifiable, target extremely unclear, B2C vs B2B fork, high-risk domain (legal/medical/finance/security), content may be impermissible.

Startup-specific contract fields (extend package_contract, entity_anchor: F-):

venture_contract:
  idea: <one-paragraph normalized idea>
  audience_tone: <startup | investor | internal | enterprise | dev>
  business_model: B2C | B2B | API | marketplace | hybrid
  output_dir: project_document_package
  zip_name: project_document_package_<slug>.zip   # [A-Za-z0-9_-] only

Phase Contracts

Phase 1: Research (01_research)

Agent Role Required
field Market background, trends, JTBD synthesis, interview/survey design; WebSearch-grounded with sources → references.md (or research_todo.md if ungrounded) Yes
compete Direct + indirect competitor analysis, differentiation gap, positioning input Yes (skip at lite)
echo[demand] Synthetic user demands / pain points / unmet needs across personas Yes
cast Persona generation → personas.md Conditional: depth ≥ mvp

Outputs: market_research.md, user_research_plan.md, interview_script.md, survey_questions.md, personas.md, jobs_to_be_done.md, competitor_analysis.md, trend_analysis.md, references.md, research_todo.md. Exit gate: Uncertain claims are flagged as hypotheses; every external claim has a source ref or a research_todo entry.

Phase 2: Product Spine (03_product) — BARRIER

Agent Role Required
scribe[unified] PRD (L0→L3), user stories, acceptance criteria, IA, non-functional requirements, release plan Yes
spark Feature catalog with feature_id (F-001…), name, description, user_value, mvp_or_later, related KPI/test/backlog stubs Yes
rank MoSCoW classification (Must / Should / Could / Won't for MVP) Yes
pulse KPI tree + product metrics, each KPI linked to feature_id(s) Yes (skip at lite)
void YAGNI scope cut (keep MVP lean) Conditional: depth = full or scope bloat

Outputs: prd.md, feature_catalog.md, mvp_scope.md, roadmap.md, user_stories.md, acceptance_criteria.md, kpi_tree.md, information_architecture.md, product_metrics.md, release_plan.md. Exit gate (HARD): The canonical feature_id table (F-001… with MoSCoW tier + mvp_or_later) is finalized and emitted in _AGENT_CONTEXT to every Phase 3 track. No Phase 3 agent starts before this. MVP scope must not contain Won't-have items; roadmap and mvp_scope must agree.

Phase 3: Parallel Documentation Tracks

Each track receives the framing contract + canonical feature_id table. Tracks write to disjoint files. Run in waves to respect the per-hub cap.

Track Dir Agents Key outputs
Brand 02 vision (direction) → muse (design_tokens.json) ‖ prose (copy/voice) brand_strategy, naming_candidates (≥20), positioning, brand_voice, messaging_framework, visual_direction, design_tokens.json, copy_examples, brand_checklist
UX/UI 04 palette (usability/states) ‖ canvas (Mermaid wireframes) ‖ echo (walkthrough) ‖ prose (empty/error/loading copy) ux_flows, screen_specifications, wireframes_mermaid, component_inventory, state_design, onboarding_flow, accessibility_guidelines, responsive_design_policy
LP 05 funnel (index.html + styles.css + lp_copy + conversion) ‖ prose (microcopy) lp_copy, index.html, styles.css, faq, conversion_strategy, seo_metadata, analytics_plan
Marketing 06 funnel/funnel[premium] (GTM/channels) ‖ pulse (metrics) ‖ experiment (growth_experiments.md) go_to_market_strategy, channel_strategy, pricing_strategy, content_marketing_plan, launch_plan, social_posts (30-day), email_sequences, pr_plan, sales_material_outline, growth_experiments
Tech 07 atlas (architecture+Mermaid) → schema (database_schema.sql) ‖ gateway (api_design_openapi.yaml) ‖ beacon (monitoring) ‖ gear (CI/CD) ‖ crypt? (auth/crypto) system_architecture, tech_stack, data_model, database_schema.sql, api_design_openapi.yaml, data_pipeline, auth_and_permissions, security_privacy, monitoring_observability, infrastructure_plan, ci_cd_plan, technical_risks
AI Policy 08 oracle (AI usage, prompts, eval, guardrails, human review, logging) ai_usage_policy, prompt_design, evaluation_policy, hallucination_risk_controls, human_review_workflow, model_selection, ai_logging_policy, ai_disclaimer_templates
Legal/Risk 09 canon[legal] (ToS/Privacy/Cookie drafts) ‖ cloak (privacy/PII) ‖ canon[regulatory]? (compliance_checklist) ‖ omen+ripple (risk_register.md) legal_considerations, data_rights_policy, privacy_policy_draft, terms_of_service_draft, cookie_policy_draft, risk_register, compliance_checklist
Testing 10 matrix qa-scenario (test_cases.csv TC-001…, mapped to feature_id) ‖ radar (strategy + ai_evaluation_cases.csv) test_strategy, test_cases.csv, qa_checklist, ai_evaluation_cases.csv, performance_test_plan, security_test_plan, accessibility_test_plan, release_checklist
PM 11 sherpa (backlog.csv BL-001…, mapped to feature_id) ‖ rank (priority) ‖ scribe (RACI/milestones) backlog.csv, milestones, team_structure, budget_estimate, raci_matrix, decision_log, meeting_cadence, outsourcing_plan
Mock Data 12 radar (fictional sample data only) sample_users.json, sample_events.json, sample_content.csv, sample_notifications.json, sample_settings.json, sample_logs.json
Assets 13 builder[image] (image_generation_prompts) ‖ canvas (diagram_index) README, icon_direction, image_generation_prompts, diagram_index

Per-file writing convention: every Markdown file includes these sections — Purpose / Intended readers / Assumptions / Body / MVP treatment / Future expansion / Next steps / Related files. CSV/JSON/YAML/SQL/HTML/CSS must be real, loadable, syntactically valid structures.

Legal disclaimer rule: all 09 legal content states that professional/legal counsel review is required — never asserted as definitive legal advice.

Phase 4: Overview Synthesis (00_overview)

Run after Phase 3 so the overview reflects (not predicts) the package.

Agent Role
spark product_concept, one_page_pitch (for investors / collaborators)
scribe executive_summary, decision_summary, 90_day_action_plan
magi (depth ≥ raise) sanity-check success conditions, risks, business model coherence

Outputs: executive_summary.md, product_concept.md, one_page_pitch.md, assumptions.md, decision_summary.md, 90_day_action_plan.md.

Phase 5: Integration and Validation

Agent Role
attest / judge Build the traceability matrix and adversarially check consistency across artifacts
Nexus (internal) Generate document_manifest.csv, validation_report.md, README.md; run format syntax checks

Traceability matrix verifies: feature_id↔user_story, feature_id↔acceptance_criteria, feature_id↔test_cases.csv, feature_id↔backlog.csv, KPI↔feature, LP value-prop↔brand message, risk↔mitigation, AI usage↔evaluation case. Consistency checks: idea↔all docs; MVP not over-scoped; MVP vs future not conflated; tech stack ↔ DB/API/infra; target↔LP↔channel↔pricing; tests exist for major features; AI eval cases exist if AI is used; legal/privacy/security noted. validation_report.md also records: directory/file completeness vs the depth tier's required set; CSV/JSON/YAML/SQL/HTML/CSS syntax validity (lint command results); post-unzip usability (file count + structure from unzip -l).

Phase 6: Package

As the engine's Phase 6 (reference/package-recipe.md § Shared Engine, incl. the format syntax lint commands) — write tree (UTF-8) → syntax lint → zip → expansion test → secrets/PII scrub. Startup-specific naming: tree root project_document_package/, archive project_document_package_<slug>.zip.

Traceability Anchor (core design constraint)

As the engine's Generalized Traceability Anchor (reference/package-recipe.md § Generalized Traceability Anchor) — the F-001 entity-id barrier at Phase 2, passed verbatim into every Phase 3 _AGENT_CONTEXT handoff; tracks reference existing IDs only, never mint new ones. Startup-specific ID formats: features F-001, test cases TC-001, backlog BL-001. Phase 5 fails the package if any TC-/BL- item references a non-existent F-, or any Must-have feature lacks a user story / AC / test case.

Output and Packaging

  • Deliverable is a filesystem tree + zip, not a chat-embedded "download link" (CLI has no download concept). The final report gives the absolute path to the zip.
  • Directory structure matches the canonical layout exactly (project_document_package/ with 00_overview … 13_assets + top-level README.md, document_manifest.csv, validation_report.md).
  • document_manifest.csv columns: path,title,purpose,target_reader,status,related_files,priority.
  • risk_register.md columns: risk_id,category,description,likelihood,impact,mitigation,owner,status.
  • test_cases.csv columns: id,feature_id,category,feature,scenario,precondition,steps,expected_result,priority.
  • ai_evaluation_cases.csv columns: id,input,expected_behavior,risk_type,evaluation_criteria,pass_fail.
  • backlog.csv columns: id,epic,feature_id,task,description,priority,owner,estimate,dependency,status.
  • growth_experiments.md items: experiment_id,hypothesis,target_segment,channel,action,success_metric,duration,required_assets,priority.

Conditional Inclusion

Condition Add Skip
depth = lite — 02, 06, 07-deep, 08, 09, 10, 11, 12, 13 (keep 00/01-lite/03/04-lite/05)
mode includes ai-product full 08 + ai_evaluation_cases + human_review — (08 never skipped in this mode)
business_model = B2B / b2b-saas canon[regulatory], security_privacy deepened, sales_material, SLA in 11 —
business_model = B2C / b2c-growth 02 brand deep, 05 LP, SEO/social/onboarding in 06 —
UI surface absent (API/infra product) — 04 UX, 05 LP (replace with API docs emphasis in 07)
Figma in workflow frame (extract tokens) —

(depth = full and web_grounding = unavailable rows are as the engine's Conditional Inclusion — see reference/package-recipe.md.)

AUTORUN Chain Template

Nexus AUTORUN venture idea="<X>" depth=<...> mode=<...>
  ── Phase 0 Framing ──────────────────────────────────
  → as package AUTORUN Phase 0 (`reference/package-recipe.md`), fixed to preset=startup (no
    auto-detect); clarify gate: domain/target/B2X-fork/high-risk/impermissible (else assume)
  → emit venture_contract
  ── Phase 1 Research ─────────────────────────────────
  → field(market+trend+JTBD, web-grounded|research_todo)
  ‖ compete(direct+indirect+diff)?        # skip at lite
  ‖ echo[demand](user demands) → cast(personas)?  # cast if depth≥mvp
  ── Phase 2 Product Spine [BARRIER] ──────────────────
  → scribe[unified](PRD+stories+AC+IA+NFR+release)
  → spark(feature_catalog: F-001… + mvp_or_later)
  → rank(MoSCoW)  → pulse(KPI tree ↔ feature_id)?
  → void(YAGNI)?                          # depth=full or bloat
  → ═══ EMIT canonical feature_id table → bind to all Phase 3 ═══
  ── Phase 3 Parallel Doc Tracks (waves, feature_id-bound) ─
  → [Brand]  vision → muse(design_tokens.json) ‖ prose
  ‖ [UX]     palette ‖ canvas(mermaid) ‖ echo ‖ prose(states)
  ‖ [LP]     funnel(index.html+styles.css+copy) ‖ prose
  ‖ [Mktg]   funnel/funnel[premium] ‖ pulse ‖ experiment(growth_experiments)
  ‖ [Tech]   atlas(arch+mermaid) → schema(schema.sql) ‖ gateway(openapi.yaml)
                                  ‖ beacon ‖ gear ‖ crypt?
  ‖ [AI]     oracle(policy+prompts+eval+guardrails+human_review)
  ‖ [Legal]  canon[legal](ToS/Privacy/Cookie) ‖ cloak ‖ canon[regulatory]? ‖ omen+ripple(risk_register)
  ‖ [Test]   matrix qa-scenario(test_cases.csv TC-001 ↔ F-id) ‖ radar(strategy + ai_eval_cases.csv)
  ‖ [PM]     sherpa(backlog.csv BL-001 ↔ F-id) ‖ rank ‖ scribe(raci/milestones)
  ‖ [Mock]   radar(sample_*.json/csv — fictional only)
  ‖ [Assets] builder[image](image_prompts) ‖ canvas(diagram_index)
  ── Phase 4 Overview Synthesis (post-tracks) ─────────
  → spark(concept+one_page_pitch) ‖ scribe(exec_summary+decision+90day)
  → magi(coherence)?                      # depth≥raise
  ── Phase 5 Integrate + Validate ─────────────────────
  → as package AUTORUN Phase 5 (`reference/package-recipe.md`) — traceability matrix + cross-doc
    consistency + universal grounding gate + document_manifest.csv/validation_report.md/README.md
    + syntax lint (no legal/ai-adoption risk gate; startup carries none)
  ── Phase 6 Package ──────────────────────────────────
  → as package AUTORUN Phase 6 (`reference/package-recipe.md`); startup naming:
    project_document_package/ tree → project_document_package_<slug>.zip

Output Report

venture emits the engine's Package Manifest Report (reference/package-recipe.md § Output Report) with the startup preset's inventory — no separate report name. The startup-specific traceability anchor (feature_id) and distinctive files are listed in § Output and Packaging above.

Termination Bound / Resume

Both inherited from the package engine (reference/package-recipe.md): Termination bound N/A (non-loop, single forward pass) and checkpoint-resume with the Phase 2 feature_id list frozen across resumes. This blueprint adds neither.

Shared-Protocol References

venture cites the engine's protocol set rather than re-deriving it — reference/doc-quality-protocol.md (W1-W12, discharged by the engine's Phase 5 validation), reference/autonomy-quality-protocol.md (intent contract, Acceptance Provenance), and _common/TRACEABILITY.md (the feature_id anchor). See reference/package-recipe.md for how each is applied.

Failure Modes Prevented

venture is the startup preset of the package engine and prevents failures through the engine's consolidated controls — entity-id barrier, Universal Grounding Gate, per-preset Risk Gates, cross-doc consistency, and format lint. See reference/package-recipe.md § Failure Modes Prevented; this blueprint adds no preset-specific failure modes beyond the startup risk-gate row tabulated there.

Failure Escalation

Failure Detected by Escalation
Business idea missing Phase 0 Ask for the idea — do not invent a venture
MVP over-scoped (Won't-have leaked) Phase 5 / void Re-run rank or void, downgrade scope
tech_stack ↔ schema/API mismatch Phase 5 Return Tech track
LP copy ↔ brand mismatch Phase 5 Return LP/Brand track

5 further rows (domain-unidentifiable clarify gate, Phase 2 entity-table barrier, dangling entity-id reference, format-lint failure, secrets/PII scrub) inherit the engine's Failure Escalation verbatim — see reference/package-recipe.md § Failure Escalation.

Source: SKILL.md on GitHub

3 warnings13d5 checks · Risk MEDIUM
  • Gen Agent Trust Hub13d

    The 'nexus' skill is a comprehensive multi-agent orchestration framework that manages complex task chains. While it incorporates extensive internal guardrails and verification protocols, it explicitly mandates the use of high-risk flags that bypass security permissions to achieve autonomy. It also provides instructions for establishing persistent tasks via cron and GitHub Actions, and utilizes external research tools to fetch content from the web.

  • Socket13d

    1 alert: gptSecurity

  • Snyk13d

    Risk: LOW · No issues

  • Runlayer6mo

    6/22 files flagged

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at c805268. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated 2 weeks ago

README badge

README badge for simota/agent-skills/nexus