agent-readiness - SKILL.md Agent Skill

name: agent-readiness description: Score a task or workflow on its coordination-vs-judgment ratio to tell whether it's a good agent candidate (EXO 3.0 task-decomposition matrix). metadata: user_invocable: "true"

/agent-readiness

Salim Ismail's EXO 3.0 diagnostic, made runnable: score a piece of work on its coordination-vs-judgment ratio. Coordination work — routing information, approvals, scheduling, status tracking, prescriptive/standardized steps — is what AI agents do best, so it's agent-ready. Judgment work — resolving ambiguity, handling exceptions, navigating relationships, making a call with no playbook — should stay with (or escalate to) humans.

When to use

Before delegating a task/workflow to an agent — is it actually a good candidate?
When deciding whether a process should be fully automated, agent-with-oversight, hybrid, or kept human-led.
When mapping which of your workflows a small team + agents could rebuild first (Salim: "every task that scores high on coordination has agent readiness").

How

Score a task:

curl -X POST -H "Authorization: Bearer $AUTH" -H 'Content-Type: application/json' \
  -d '{"task":{"name":"Invoice intake","description":"Route invoices, schedule approvals, track status, compile a weekly report, notify owners."}}' \
  http://localhost:${INSTAR_PORT:-4042}/agent-readiness/score

Score a workflow (by its steps):

curl -X POST -H "Authorization: Bearer $AUTH" -H 'Content-Type: application/json' \
  -d '{"workflow":{"steps":["Fetch the record","Assign accounts","Schedule orientation","Update the tracker"]}}' \
  http://localhost:${INSTAR_PORT:-4042}/agent-readiness/score

Returns:

{
  "coordinationSignals": 5, "judgmentSignals": 0,
  "coordinationRatio": 1.0, "overallReadiness": 100,
  "recommendation": "deploy-agent",
  "reason": "...",
  "matched": { "coordination": ["route","schedule",...], "judgment": [] }
}

recommendation is one of: deploy-agent (75+), agent-with-oversight (55–74), hybrid (40–54), human-led (<40). Deterministic + advisory — it answers a question; it never blocks. Pair it with the MTP Protocol (/intent/org/test-action) to check both "is this agent-ready?" and "does our purpose endorse it?"