snowflake-labs ICR Reference
ICR = Total Required Operations (Ops) / Intent Expressions
1 intent : N ops
| ICR Range | Classification | Agentic Readiness |
|---|---|---|
| 1–3 | Command relay | Low — agent is just a wrapper |
| 4–8 | Automation wrapper | Medium — meaningful compression |
| 9–15 | Architectural partner | High — strong agentic readiness |
| 15+ | Intent engine | Very high — system owns the DAG |
Glass Box principle: ICR alone is not the goal. Target: high ICR + full observability + safe retry = Glass Box Compression.
Stopping Points
Ask the user to provide the workflow:
ask_user_question:
header: "Workflow"
question: "Paste your workflow, script, Taskfile, CI config, or describe the steps you want to measure."
type: text
defaultValue: ""
Optionally ask for context:
ask_user_question:
header: "Intent count"
question: "How many natural language intent expressions does the user write to trigger this? (e.g. 1 for a single $scaffold invocation)"
type: text
defaultValue: "1"
Analysis
Step 1 — Count Operations (Ops)
Walk through the workflow and count each distinct state-changing step:
- Every API call / CLI command that creates, modifies, or deletes state
- Every validation / verification step
- Every credential or secret operation
- Every git operation
Do NOT count: read-only checks, display output, echo/print statements.
Present the count as a table:
Category Count Examples
──────────────────── ────── ────────────────────────────────────
Resource creation N gh repo create, snow sql setup.sql
Auth / secret ops N gh secret set, glab variable set
Verification N DESC USER, snow connection test
CI/CD operations N gh workflow run, glab pipeline run
──────────────────────────────────────────────────────────────
Total Ops N
Step 2 — Count Intent Expressions
Count the number of distinct natural language goals the user has to express to trigger the entire workflow (typically 1 for a single skill invocation).
Step 3 — Calculate and Classify
ICR = Total Ops / Intent Expressions = N
Classification: <relay | wrapper | partner | engine>
Step 4 — Glass Box Assessment
Evaluate the workflow on three Glass Box dimensions:
| Dimension | Present? | Notes |
|---|---|---|
| Observability | yes/no | Does it report what it created/changed? |
| Safe retry | yes/no | Can individual steps be retried without side effects? |
| Partial override | yes/no | Can the user skip or override specific steps? |
Step 5 — Recommendations
Identify the top 2–3 operations that could be compressed further:
- Which repeated patterns could become a single intent expression?
- Which verification steps could be folded into the preceding action?
- What Glass Box reporting is missing?
Presenting Results
ICR Analysis: <workflow name>
─────────────────────────────────────
Total Ops : N
Intent Expressions : N
ICR : N (<classification>)
─────────────────────────────────────
Glass Box score : X/3 (observability + retry + override)
─────────────────────────────────────
Top compression opportunities:
1. <opportunity>
2. <opportunity>
Then ask:
ask_user_question:
header: "Next step"
question: "What would you like to do?"
options:
- label: "Apply IDD to improve this workflow"
- label: "Rewrite a specific prompt in this workflow"
description: "Hand off to rewrite-prompt"
- label: "Done"