detail-strategy

name	detail-strategy
description	Unified detail/information/describe retrieval strategy. Handles MAS models, Job models, JobDef models, SCR models, and tables. Verify resources exist using find-resources skill before retrieving details (except SCR models, which can be queried directly).

Detail Strategy

Use this strategy when the user requests information about a resource.

Supported resource types: MAS model, Job model, JobDef model, SCR model, or table.

If the specified resource is ambiguous, ask the user for clarification (e.g. "Are you asking about a MAS model, a Job model, a JobDef model, a SCR model or a table?").

Model: If user just says "model X" without specifying type, default to MAS model (this is an explicit exception to the 'never invent' rule, but it is a predefined convention in SAS).

Verification rules

• Always verify the target resource exists using the appropriate find-* tool before attempting retrieval, except for SCR models (see Exceptions below).
• For tables, determine server (CAS vs SAS) during verification; if ambiguous, ask the user for the full lib.table or server.
• For models, strip suffixes such as .mas, .job, .jobdef or .scr before lookup to avoid lookup failures.

Defaults & exceptions

• Default model type: MAS when the user says model X without specifying a type.
• If verification fails, inform the user and ask for corrected or more specific identifiers.

---

Classification & Verification Process

Phase 1: Classify the Resource Type

Determine resource type from context/naming conventions:

Pattern	Resource Type
mas X or X.mas	MAS model X
mas model X	MAS model X
scr X or X.scr	SCR model X
scr model X	SCR model X
job model X	Job model X
jobdef X or X.jobdef	JobDef model X
jobdef model X	JobDef model X
table X in library Y	Table X in library Y
model X.mas	MAS model X
model X.job	Job model X
model X.jobdef	JobDef model X
model X.scr	SCR model X
model X (ambiguous)	Default to MAS model (explicit convention)

If resource is ambiguous, ask user for clarification.

Phase 2: Verify Resource Exists (Skip for SCR Models)

IMPORTANT: Strip the suffix if user included it, use base name for lookup (e.g. "churnRisk.mas" → "churnRisk") to avoid lookup failures.

For each resource type, use the appropriate verification tool:

Resource Type	Tool
MAS model	sas-score-find-mas
Job model	sas-score-find-job
JobDef model	sas-score-find-jobdef
Table	sas-score-find-table
SCR model	(no verification needed)

If verification fails, inform the user and ask for additional details or corrections.

---

Detail Retrieval Process

Phase 3: Get Details

Option A: MAS Model Details

Trigger phrases: "what inputs does mas model X need", "describe mas X", "show variables for model mas X, describe X.mas", "mas model X metadata", "mas model X information", "describe model X" (default to MAS), "what inputs does mas X need", "describe mas X"

Tool: sas-score-mas-info

Parameters:

sas-score-mas-info({
  model: "<model name>"
})

Returns:

• Input variables (name, type, role)
• Output variables (name, type, possible_values)
• Model type
• Description

Example:

User: "What inputs does mas model churnRisk need?"

1. Find: sas-score-find-mas({ name: "churnRisk" })
2. Get info: sas-score-mas-info({ model: "churnRisk" })
3. Return: { inputs: [...], outputs: [...], description: "..." }

---

Option B: SCR Model Details

Trigger phrases: "what does SCR model X need", "describe SCR model X", "scr model X inputs", "scr model X schema", "what inputs does scr model X need", "describe scr X"

Tool: sas-score-scr-info

Parameters:

sas-score-scr-info({
  url: "<scr endpoint>"
})

Returns:

• Input schema (variable names, types, required/optional)
• Output schema (prediction, probabilities, scores)
• Model metadata

Example:

User: "Show inputs for SCR model at https://scr-host/models/loan"

1. Get info: sas-score-scr-info({ url: "https://scr-host/models/loan" })
2. Return: { inputs: [...], outputs: [...] }

Note: SCR models typically do not require pre-verification (can call scr-info directly)

---

Option C: Table Details

Trigger phrases: "what columns in table X", "describe table X", "show schema for table X", "table X structure", "table X metadata"

Tool: sas-score-table-info

Parameters:

sas-score-table-info({
  lib: "<library>",
  table: "<table name>",
  server: "<cas|sas>"
})

Returns:

• Columns array (name, type, label, format, length)
• Table info (rowCount, fileSize, created, modified)

Example:

User: "What columns are in the customers table in Public?"

1. Find: sas-score-find-table({ lib: "Public", name: "customers", server: "cas" })
2. Get info: sas-score-table-info({ lib: "Public", table: "customers", server: "cas" })
3. Return: { columns: [...], tableInfo: {...} }

---

Option D: Job Model Details

Trigger phrases: "what inputs does job model X need", "describe job model X", "show variables for job model X", "job model X metadata", "job model X information", "what inputs does job X need", "describe job X"

Tool: sas-score-job-info

Parameters:

sas-score-job-info({
  model: "<model name>"
})

Returns:

• Input variables (name, type, role)

Example:

User: "What inputs does job model churnRisk need?"

1. Find: sas-score-find-job({ name: "churnRisk" })
2. Get info: sas-score-job-info({ model: "churnRisk" })
3. Return: { inputs: [...] }

Decision Tree

User requests information/details
  ├─ About a MAS model?
  │   → Verify: sas-score-find-mas
  │   → Call: sas-score-mas-info
  │
  ├─ About a SCR model?
  │   → Call: sas-score-scr-info (skip verification; validate URL first)
  │
  ├─ About a Job model?
  │   → Verify: sas-score-find-job
  │   → Call: sas-score-job-info
  │
  ├─ About a JobDef model?
  │   → Verify: sas-score-find-jobdef
  │   → Call: sas-score-job-info
  │
  └─ About a table?
      → Verify: sas-score-find-table (determine CAS or SAS server)
      → Call: sas-score-table-info

---

Implementation Checklist

For each detail/information request:

• Classify resource type (MAS / Job / JobDef / SCR / table)
• Verify resource exists (use find-resources skill, except SCR)
• Determine server for tables (CAS or SAS)
• Execute appropriate detail tool
• Format results (column alignment, readable structure)
• Append Strategy Summary to response

---

Response Format

Always append a Strategy Summary to responses:

---

**Strategy Summary:**
- **Classification**: [Resource type identified]
- **Verification**: [Resource found or skipped (SCR)]
- **Tool Used**: [Detail tool invoked]
- **Server**: [CAS/SAS for tables, N/A for models]

---

Error Recovery

If a request fails:

1. Resource not found → Ask user to verify name/spelling
2. Server mismatch → Re-verify server location with find-resources
3. Invalid URL (SCR) → Ask for correct SCR endpoint URL
4. Tool error → Return error message verbatim and ask for clarification

---

Examples

Example 1: Model Information

User: "Describe model creditScore"

Workflow:

1. Classify: MAS model detail request
2. Verify: Find model creditScore → Found ✓
3. Execute: sas-score-mas-info({ model: "creditScore" })
4. Return: Model inputs, outputs, and description

Example 2: SCR Model Schema

User: "What inputs does the SCR loan model need?"

Workflow:

1. Classify: SCR model detail request
2. Execute: sas-score-scr-info({ url: "https://scr-host/models/loan" })
3. Return: Input schema and output schema

Example 3: Table Columns

User: "Show columns for customers in Public"

Workflow:

1. Classify: Table detail request
2. Verify: Find table customers in Public → CAS ✓
3. Execute: sas-score-table-info({ lib: "Public", table: "customers", server: "cas" })
4. Return: Column names, types, and table metadata