ICR Simulate

Simulate an intent check for a hypothetical tool call without actually executing.

Usage

• /icr:simulate Bash "rm -rf /tmp/old-builds/*.tar.gz"
• /icr:simulate Write "/path/to/file.txt" "content here"
• /icr:simulate Edit "/path/to/file.txt" "old text" "new text"
• /icr:simulate Task "Deploy the application"

What It Shows

This shows what would happen if the tool call were made:

1. Intent Document - The structured interpretation that would be generated

   • Task description
   • Affected resources
   • Boundaries
   • Reversibility assessment

2. Confidence Score - The calculated confidence with component breakdown

   • Ambiguity analysis
   • Intent-to-action distance
   • Historical patterns
   • Uncertainty markers

3. Severity Classification - How the action would be classified

   • Classification layer (static, inferred, user rule)
   • Final severity level

4. Decision Route - What would happen

   • AUTO_APPROVE, AI_REVIEW, or HUMAN_REVIEW
   • The thresholds that triggered this route

Example Output

ICR Simulation
---------------------------------------------------------------------
Tool: Bash
Args: rm -rf /tmp/old-builds/*.tar.gz

INTENT DOCUMENT (simulated):
  Task: Delete all .tar.gz files in /tmp/old-builds/ directory

  Affected:
    - /tmp/old-builds/*.tar.gz (unknown count)

  Excluded:
    - Other file types in same directory
    - Files in subdirectories (unless -r affects them)

  Boundaries:
    - Will not delete the directory itself
    - Will not affect files outside /tmp/old-builds/

  Reversibility:
    Can undo: NO
    Method: Restore from backup
    Effort: Impossible (permanent deletion)

CLASSIFICATION:
  Severity: CRITICAL
  Classified by: userRule
  Reason: Command contains "rm -rf"

CONFIDENCE: 0.65
  Ambiguity:    0.60 (moderate - "old-builds" is descriptive)
  Distance:     0.80 (close - explicit delete command)
  Historical:   0.50 (no similar patterns)
  Uncertainty:  0.70 (no hedging, clear intent)

DECISION: HUMAN_REVIEW
  Confidence 0.65 < AI review threshold 0.90 for CRITICAL
  Would require human approval before executing

---------------------------------------------------------------------
Note: This is a simulation. No action was taken.

Use Cases

• Preview before action: See what ICR would do before you ask Claude to actually run something
• Learning: Understand how ICR classifies and routes different types of actions
• Testing rules: Verify your custom severity rules work as expected
• Documentation: Show examples of how ICR handles specific scenarios

Notes

• No action is taken during simulation
• No receipt is created
• The simulation uses current configuration and session state
• Historical pattern matching uses real history, but the simulation itself doesn't affect history