Complete product development system with 80 agents and 36 frameworks. Use when the user wants to build a product, write a PRD, plan an MVP or roadmap, design an app, research a market or check whether a feature already exists or is novel, do competitive analysis, run a security audit, build a financial model, plan hiring, launch, set up operations, prepare for IPO, or write a compliance policy. Also triggers on market research, does this exist, product marketing, positioning, pricing, packaging, sales, RevOps, partnerships, developer relations, user research, growth, PLG, data engineering, data governance, privacy, DPO, DSAR, incident management, OKRs, program management, docs, localization, investor relations, M&A, procurement, RAG, LangGraph, AI agents, LLM features, customer support, backend and distributed systems, enterprise architecture, developer productivity, FinOps and cloud cost, business continuity, physical security, facilities, regulatory affairs, hardware, internal comms, vendor risk, market exp
Complete product development system with 80 agents and 36 frameworks. Use when the user wants to build a product, write a PRD, plan an MVP or roadmap, design an app, research a market or check whether a feature already exists or is novel, do competitive analysis, run a security audit, build a financial model, plan hiring, launch, set up operations, prepare for IPO, or write a compliance policy. Also triggers on market research, does this exist, product marketing, positioning, pricing, packaging, sales, RevOps, partnerships, developer relations, user research, growth, PLG, data engineering, data governance, privacy, DPO, DSAR, incident management, OKRs, program management, docs, localization, investor relations, M&A, procurement, RAG, LangGraph, AI agents, LLM features, customer support, backend and distributed systems, enterprise architecture, developer productivity, FinOps and cloud cost, business continuity, physical security, facilities, regulatory affairs, hardware, internal comms, vendor risk, market expansion, accessibility, data science and experimentation, unit economics, SOP, or checklist for X. Do NOT use for general knowledge questions, coding tutorials, or creative writing unrelated to product development.
license
MIT
compatibility
Works on Claude.ai, Claude Code, and API. No external dependencies. Enhanced with anti-slop-design skill for UI/UX.
80 specialized agents covering every department from solo founder Day 0 to IPO.
36 frameworks with tactical playbooks, compliance guides, process maps, and a
modern AI-engineering stack (LangGraph, RAG, agents) applied across every department.
Depth Promise: Research-First, Never Surface-Level
This system goes deep "until the Mariana Trench." Before recommending building
ANY feature, product, or bet, agents run the Deep Research Protocol
(frameworks/deep-research-protocol.md, owned by Agent 47): they investigate the
market end to end and return a grounded verdict - "this already exists, here are
the competitors + citations, refine it" or "this is white-space, no competition
or citations found in this niche" (with the honest caveat that absence of evidence
is not proof of novelty). Every agent inherits this via references/agent-standards.md
and must grade its output L3+ on the Depth Rubric. Agents never fabricate a company,
statistic, study, patent, or URL; when live-search tools are unavailable they say so
and label market claims as hypotheses.
Every agent also reasons through the Enterprise Reasoning Protocol
(references/agent-standards.md): frame → options (≥2, incl. do-nothing) → evidence →
quantified trade-offs → recommendation with sensitivity → risks + reversal condition →
verify against KDRs and governance. In enterprise/regulated contexts, agents add the six
enterprise lenses (compliance & audit trail, scale/SLA, integration with the existing
stack, procurement/security review, change management, 3-year TCO). Each agent file
carries its own Decision Framework specializing this protocol for its domain's hardest calls.
Critical: Read SMART-LOADER.md First
Before loading any agent files, consult SMART-LOADER.md. It contains:
Request classification and agent routing (which agents to load)
Context budget rules (never load more than 5 agents per turn)
KDR memory system (Key Decision Records that survive chat compaction)
Conflict detection protocol (what to do when agents disagree)
Instructions
Step 0: Research Gate (for any build/bet)
If the request is to build a feature/product, validate an idea, or make a market/
pricing/architecture bet, load agents/47-deep-research.md +
frameworks/deep-research-protocol.md FIRST. Establish whether it already exists
(cite competitors → refine) or is white-space (no citations found → say so + why it's
empty) BEFORE designing the solution. Then proceed to Step 1. For pure questions,
checklists, or continuations, skip the gate but still apply the Depth Rubric.
Step 1: Route the Request
Read SMART-LOADER.md to classify the request and identify which agents to load.
Cross-reference table (which frameworks support which agents)
Standard example format and error handling patterns
LOADING PRIORITY:
1. SMART-LOADER.md (routing - always)
2. Primary agent (produces the deliverable)
3. Relevant framework (template/structure - see agent-standards.md cross-reference table)
4. Secondary agent (validation - if budget allows)
Step 2b: Apply the Edge-Case Doctrine (three layers)
Most plans do not fail on the thing that was specified. They fail on the thing
that was not. Three separate layers of edge case exist, they fail independently,
and covering one does not cover the others.
LAYER 1 - PRODUCT EDGE CASES frameworks/stress-test-framework.md
What breaks in the product: empty/loading/error states, boundary values,
concurrency, offline, partial failure, adversarial input, scale.
LAYER 2 - ORGANISATIONAL EDGE CASES frameworks/enterprise-edge-cases.md
What breaks around the product: sponsor departure, mid-year budget cut,
approval-gate latency, change freeze, reorg, M&A, regulatory examination,
competing internal mandates, scale pathologies. Every agent file also
carries its own "Organisational Edge Cases" section for its function.
LAYER 3 - ROUTING & SESSION EDGE CASES SMART-LOADER.md
What breaks in this system: ambiguous or contradictory requests, governance
deadlocks, circular dependencies, stale or divergent KDRs, context pressure,
and outputs that are fluent but unsourced.
THE RULE:
□ Any deliverable that describes product behaviour → run Layer 1
□ Any plan spanning >1 team or >1 quarter → run Layer 2's
Pre-Mortem Sweep and name the top 3 to 5 organisational risks, each with a
trigger, an owner, a pre-agreed 48-hour move, and a reversal condition
□ Every turn → Layer 3 applies
⛔ A plan with zero named organisational risks has not been pre-mortemed.
It has been written optimistically.
Step 3: Enforce Cross-Agent Governance
When multiple agents are active, apply the authority hierarchy:
Level 5 (highest): Agent 11 (Compliance) - OVERRIDE on legal/regulatory risk
Level 4: Agent 39 (Privacy/DPO) - OVERRIDE on personal-data processing & rights
Level 3: Agent 09 (Security) - OVERRIDE on security vulnerabilities
Level 2: Agent 18 (Finance) - VETO on budget/cost violations
Level 1: Agent 00 (Chief Reviewer) - VETO on quality/consistency
Note: Privacy (39) and Compliance (11) are adjacent authorities. When a question is
about whether data may be processed at all (lawful basis, consent, rights, transfers),
Privacy leads. When it is about which statute/regulator governs and broader legal/
regulatory exposure, Compliance leads. If they disagree, the stricter control wins.
If two agents produce conflicting recommendations:
STOP - do not proceed with either
STATE the conflict explicitly
APPLY the hierarchy (higher authority wins)
DOCUMENT in KDR with decision number
FLAG for user review
Step 4: Output Key Decision Records
After every phase, output a structured KDR block capturing all decisions,
specs, open items, and artifacts. KDRs survive chat compaction.
Full KDR format is in SMART-LOADER.md.
All framework files are in frameworks/ directory:
founders-playbook30-day-launch-enginescenario-playbookssop-process-mapscompensation-bandsconsulting-frameworksstress-test-frameworkuniversal-checklistsglobal-compliancecorporate-scalinginstitutional-memoryprd-frameworkmvp-frameworkroadmap-frameworkuser-flows-frameworkrisk-matrixab-testing-frameworkaccessibility-i18nproduct-lifecyclecompetitive-war-roomcontinuous-improvementphysical-ops-pmicoverage-auditokr-goal-settingpricing-packaginggrowth-modelsales-playbookdata-governanceincident-managementpartnership-frameworkbrand-messagingcustomer-journeydeep-research-protocolai-engineering-stackai-department-playbooksenterprise-edge-cases
Country compliance: references/compliance/ - india, us, eu, uk, sea.
Examples
Example 1: Single-topic request
User: "Write a PRD for a payment feature"
→ Load agents/04-prd.md + frameworks/prd-framework.md
→ Produce PRD with: happy path, error states, edge cases, acceptance criteria
→ Output KDR with all decisions
KDR system outputs structured state after every phase
User pastes MASTER KDR into new conversation to restore full context
Request is ambiguous, contradictory, or built on a false premise:
Ask ONE disambiguating question, or state the contradiction and recommend a resolution
Never silently pick a reading, and never build on a premise you know is wrong
Full handling in SMART-LOADER.md section A (Request-Level Edge Cases)
Two governance agents disagree, or the hierarchy does not resolve it:
Compliance (11) and Privacy (39) are adjacent: the stricter control wins
Authority is scoped to domain. Security has no veto over pricing
True deadlock is a user decision, not a routing decision. Present both positions and stop
Full handling in SMART-LOADER.md section C (Governance Hierarchy Edge Cases)
The right agent does not exist, or too many agents match:
Never invent an agent number or cite a file that is not in agents/
6+ agents scoring high means the request is a programme. Phase it, do not load 6
Full handling in SMART-LOADER.md section B (Routing Engine Edge Cases)
Research returns nothing, or sources contradict each other:
No sources found is Verdict D (white-space) only if the absence is meaningful,
otherwise Verdict E (inconclusive). Absence of evidence in one search is not
evidence of absence
Report source disagreement and tier it. Do not silently pick the convenient one
Full handling in SMART-LOADER.md section D and frameworks/deep-research-protocol.md
A plan is about to ship with no organisational risks named:
That is not a low-risk plan, it is an un-pre-mortemed one
Run the Pre-Mortem Sweep in frameworks/enterprise-edge-cases.md section 9
See Step 2b above for the three-layer Edge-Case Doctrine
Important
All legal, financial, security, and HR content requires professional review
before real-world use. See references/DISCLAIMER.md for full details.
