AI-SHIPR — Intelligence Layer

Agent & Skill Map

Which agents orchestrate which skills — and which agents operate on their own. The full intelligence layer of AI-SHIPR, mapped.

Agent
Persistent intelligence
An agent is a role with memory and judgment. It reads context files before acting, applies a consistent mindset, and builds on what it already knows about you. It is who is thinking.
Skill
Single-task specialist
A skill is a precise procedure. It takes structured input, applies one framework, and returns structured output. It has no memory. It is how the thinking gets done.
Orchestrator Agents — call skills to do their work
Product Strategist
Iris — Strategy layer, one level above initiatives
Orchestrator

Builds complete product strategy using JTBD, Opportunity Solution Tree, and OKR — in sequence. Each mode invokes the right skill for that step. When the job is unclear, Iris starts with research and persona. When options are contested, she hands off to Decision Architect.

Research-Synthesizer
Convert raw research into structured signals before JTBD work begins
When: research files exist in I-Information/Resources/
Persona-Builder
Ground the job executor in a real persona before writing the JTBD statement
When: always, in Mode 1 before JTBD-Mapper
JTBD-Mapper
Produce job statement, dimensions, and unmet needs → writes S-Strategy/JTBD.md
When: always, Mode 1 — job definition
Opportunity-Tree
Map desired outcome to ranked opportunities → writes S-Strategy/Opportunity-Tree.md
When: always, Mode 2 — opportunity mapping
Priority-Stack
Score and rank opportunities against each other with explicit criteria
When: 4 or more opportunities surface from the tree
OKR-Partner
Set measurable goals per bet — linked to opportunity and strategic period
When: always, Mode 3 — once per selected opportunity
Lean Product Canvas
Leo — Initiative layer, one level below strategy
Orchestrator

Runs the Lean Product Canvas end-to-end — one dedicated skill per canvas box. Problem before solution, user outcome before feature. Each skill produces structured output that fills the next box. When the canvas is complete, it exports directly into AI-SHIPR artifact files.

Persona-Builder
Box 3 — Users. Ground the canvas in a named persona before writing any outcome
When: always, before Box 4
Research-Synthesizer
Box 4 — User Outcomes. Convert raw research into user job signals
When: raw research exists in I-Information/Resources/
OKR-Partner
Box 2 — Business Outcomes. Sharpen metrics against existing KPIs
When: always, after drafting Box 2
Tradeoff-Mapper
Box 5 — Solutions. Compare solution options structurally
When: 2 or more solution options exist
Assumption-Extractor
Box 6 — Hypotheses (Step 1). Surface hidden assumptions in the leading solution
When: always, before Hypothesis-Builder
Hypothesis-Builder
Box 6 — Hypotheses (Step 2). Convert each assumption into a falsifiable hypothesis
When: always, after Assumption-Extractor
Edge-Case-Finder
Box 7 — Riskiest Assumption. Stress-test the hypothesis and surface failure modes
When: always, before ranking risks
Experiment-Designer
Box 8 — Minimum Experiment. Design the smallest test that proves or disproves the riskiest assumption
When: always, produces structured experiment card → P-Proof/
Standalone Agents — operate directly on input, no skill calls
🔍
Problem Framer
Clara
Forces clarity on problem statements before any solution work begins. Surfaces vague framing, missing outcomes, and broken strategic links.
Operates through dialogue — asks the right questions, not through procedural skill invocations. Uses no sub-skills.
⚖️
Decision Architect
Finn
Structures complex decisions so they can be made explicitly. Forces options, criteria, tradeoffs, and thresholds into the open.
Called by Product Strategist when opportunity ranking is contested. Works on the input directly — no skill chain.
🗣️
Stakeholder Translator
Lena
Prepares alignment before meetings. Reframes initiatives in stakeholder language, surfaces objections in advance, and identifies tension points before they derail the room.
Context and judgment only — role-specific translation requires no procedural skill invocations.
🔬
Structural Integrity Auditor
Rex
Runs a rigorous structural audit on any AI-SHIPR artifact — initiative, hypothesis, strategy, or experiment — against a fixed checklist.
Audit is deterministic — reads the artifact, checks against checklists, reports gaps. No sub-skills needed.
📚
Learner
Sage
Scans new resources in I-Information/Resources/ and surfaces what is relevant to current strategy and active initiatives. A filter, not a summarizer.
Reads files directly. Filtering against current strategy context is its core job — no procedural skills needed.
🧠
PM Coach
Nora
A thinking partner for interpersonal challenges. Helps PMs handle difficult situations, generate alternatives when stuck, and extract patterns from past experience.
May route to Decision Architect (agent) when a people problem is actually a product decision in disguise.
📊
Portfolio Strategist
Reid
Cross-product strategy and resource allocation. Surfaces gaps, conflicts, and imbalances that are invisible when looking at one PM's work at a time.
Lead Mode only. Reads Roster, Strategic Bets, and all initiative files directly — no skill invocations.
👥
Team Manager
Grant
Thinking partner for managing a product team. Helps with 1:1 prep, performance conversations, coaching PMs, and surfacing team health signals.
Lead Mode only. Reads relationship and 1:1 log files directly — no procedural skill invocations.
The Full Strategy-to-Execution Chain
Intended Sequence
From raw research to a filed experiment card — how the pieces connect
Research-Synthesizer + Persona-Builder
JTBD-Mapper S-Strategy/JTBD.md
Opportunity-Tree OKR-Partner S-Strategy/Strategic-Bets.md
Lean Product Canvas Assumption-Extractor Hypothesis-Builder H-Hypotheses/
Experiment-Designer P-Proof/
Performance-Tracker validate / invalidate / iterate
Key
Agent Persistent intelligence — reads context, applies judgment, builds on memory
Skill Always invoked — runs every time this step is reached
Skill Conditionally invoked — only when the trigger condition is true