agent-adoption-triage

An Agent Skill holding a five-question triage for routing AI work to the right architecture — script, algorithmic search, LLM workflow, or autonomous agentic loop — before any agent is built. The core inversion: the question is not whether an agent can do the work (it usually can), but whether accountability can still be distributed after you deploy it. Route the work first; pick the architecture second.

Install

Claude Code

# Copy into your global skills directory
cp -r skills/agent-adoption-triage ~/.claude/skills/agent-adoption-triage

SkillsMP

/skills add shimo4228/agent-adoption-triage

The five questions

#	Question	Routes to
Q0	Design phase or operation phase?	Surfaces the Phase-crossing decision when it is load-bearing
Q1	Can the work be expressed as deterministic rules?	No → an LLM is in the loop somewhere
Q2	Is the search space static and finite?	Yes → Script Quadrant / No → Algorithmic Search Quadrant
Q3	Can you draw the workflow on a whiteboard before running it?	Yes → LLM Workflow Quadrant
Q4	Can a pre-named gap-bearer be identified (and, if operation-phase, is the Phase-crossing decision recorded)?	Yes → Autonomous Agentic Loop Quadrant / No → do not deploy

The four quadrants (Business AI Quadrants) classify the work, not the agent: deterministic × semantic-judgment on one axis, pre-defined × exploratory control flow on the other. Choosing the Autonomous Agentic Loop Quadrant commits the organization to a non-removable attribution gap — which is why Q4 demands a named party who absorbs it at deployment time, not post-incident.

What Else Is Inside

• Per-quadrant governance sets: required controls and the redirect path when something fails (component-level redirect works in the LLM Workflow Quadrant; in the Autonomous Agentic Loop Quadrant it is foreclosed and terminates at the gap-bearer)
• Hybrid architecture guidance (Plan-and-Execute, router agents, tiered handoff) — run the tree per portion
• A 10-item anti-patterns checklist (e.g. "bounded work on an autonomous loop", "approval gate as runtime checkbox", "audit log as sample") with recovery paths
• A one-page triage record template to keep with the deployment decision

When It Triggers

• Someone says "let's build an agent for X"
• Deciding whether a use case actually needs an autonomous agent
• Reviewing an agent deployment proposal, or auditing an existing agent for accountability risk
• Choosing between a pre-defined LLM workflow and a runtime-directed loop

Curation model

This repository is manually curated, not script-synced: it is the installable publication of the adoption navigator maintained in the Agent Attribution Practice (AAP) research repository (docs/quadrants/). Improvements flow here editorially, the same way the paired ADRs were extracted.

About this skill

This skill is a design-pattern skill from the Agent Attribution Practice (AAP) research line — harness-neutral ADRs on accountability distribution for agent systems (DOI 10.5281/zenodo.19652013). It is the "how" counterpart to AAP ADR-0009 (Triage Before Autonomy) and ADR-0010 (Phase Separation) and the docs/quadrants/ navigator. Sibling skill from the same research line: llm-agent-security-principles (the "how" counterpart to AAP ADR-0001..0004). AAP is one of three research lines by @shimo4228, alongside the Agent Knowledge Cycle (AKC) (DOI 10.5281/zenodo.19200726) — a six-phase bidirectional growth loop for agent-operator intent alignment — and Contemplative Agent (DOI 10.5281/zenodo.19212118) — autonomous agents grounded in four contemplative axioms.

License

MIT