config-analysis

name: config-analysis description: >- Configuration analysis and advisory for F5 XC platform resources. Analyzes customer JSON configurations to answer questions about security posture, feature enablement, mode settings, exclusions, and best practices. Supports multi-turn Q&A within agent dispatch. Use when user provides a JSON config and asks questions about it, or asks to analyze, review, or audit a configuration. user-invocable: false

Config Analysis — Configuration Q&A

Analyzes customer JSON configurations against the api-operations resource profiles to answer questions about security posture, feature enablement, and best practices. Delegates analysis to the config-analyzer agent to keep large JSON configs out of the main session.

When This Skill Applies

Route here when the user:

• Provides a JSON configuration and asks a question about it
• Asks "is [feature] enabled/disabled" with a config present
• Asks about security posture, mode settings, or configuration review
• Asks "how to change/enable/disable [feature]" in context of a provided config
• Asks to analyze, review, audit, or explain a configuration
• Asks follow-up questions about a previously analyzed config

Critical: Subagent Delegation

Never analyze large JSON configs in the main session. Customer configurations can be hundreds of lines. All analysis must be delegated to the config-analyzer agent:

Agent(
  subagent_type="platform:config-analyzer",
  description="Analyze {resource_type} configuration",
  prompt="<task details with JSON config and reference file paths>"
)

The agent reads the resource profile reference files itself and returns a structured analysis report.

Resource Type Detection

Identify the resource type from the JSON config before dispatching to the agent. Use these structural indicators:

All profile paths are relative to: skills/api-operations/references/

If the resource type is ambiguous, include multiple candidate profile paths in the agent prompt — the agent will determine the correct one.

Dispatch Patterns

First Question (new config)

Agent(
  subagent_type="platform:config-analyzer",
  description="Analyze {resource_type} configuration",
  prompt="Read these reference files first:
  1. skills/api-operations/references/resources/{domain}/{resource}.md
  [2. additional profiles if cross-resource analysis needed]

  Then analyze this JSON configuration:
  ```json
  {paste the customer's JSON config here}
  ```

  Question: {user's question}

  Use the resource profile to interpret mutually exclusive groups,
  constrained fields, dependencies, and relationships. Cite specific
  sections from the reference files in your findings."
)

Follow-up Question (same config)

For follow-up questions about a previously analyzed configuration, re-dispatch with accumulated context:

Agent(
  subagent_type="platform:config-analyzer",
  description="Follow-up analysis on {resource_type}",
  prompt="Read these reference files first:
  1. skills/api-operations/references/resources/{domain}/{resource}.md

  Original JSON configuration:
  ```json
  {same config — paste again}
  ```

  Previous findings:
  - {compressed key findings from the prior analysis report}

  New question: {follow-up question}

  Build on the previous findings. Do not repeat analysis already
  covered unless the new question contradicts or extends it."
)

Cross-Resource Analysis

When a config references other resource types (e.g., an HTTP LB config that includes enable_waf referencing an app_firewall), include both resource profiles:

Agent(
  subagent_type="platform:config-analyzer",
  description="Analyze {primary_type} with {related_type} context",
  prompt="Read these reference files first:
  1. skills/api-operations/references/resources/{domain}/{primary}.md
  2. skills/api-operations/references/resources/{domain}/{related}.md

  Then analyze this JSON configuration:
  ...
  "
)

Multi-Turn Q&A Protocol

The config-analyzer agent is ephemeral — it has no memory between dispatches. The main session must maintain conversational state:

1. First question: Dispatch agent with JSON config + question. Store the agent's Findings and Security Posture sections.
2. Follow-up: Re-dispatch with same JSON config + compressed previous findings + new question.
3. Context compression: Extract only the key findings from the previous analysis — do not accumulate full agent outputs. Include them as bullet points under "Previous findings" in the new dispatch prompt.

This keeps follow-up dispatches focused while preserving conversational continuity.

Antipatterns — Do NOT Use

• SendMessage to a completed agent: The config-analyzer agent is ephemeral. Once it returns, it is gone. Do not attempt SendMessage or team-based continuation. Always spawn a fresh Agent invocation with accumulated context.
• Team-based coordination: Do not create a team or use persistent agent patterns for config analysis. The dispatch-and-forget model is intentional — it prevents context accumulation in long-lived agents.
• Raw JSON accumulation: Do not pass full prior agent outputs into follow-up prompts. Compress findings to bullet points to keep dispatch token cost low.

Common Question Patterns