Context commands (pwd/ls) dominate error counts but aren't actionable

[evansenter]

Problem

Commands like pwd and ls account for ~3,800 of ~8,800 errors in failure analysis, but these aren't actionable insights - they're context-gathering commands that fail for expected reasons (e.g., directory doesn't exist during exploration).

Impact

Signal-to-noise ratio in analyze_failures() and get_error_details() is poor. Users have to mentally filter out context commands.

Possible Solutions

1. Command classification - Tag commands as "context" vs "user action" based on a configurable list
2. Exclude by default - Add exclude_context_commands=True parameter to failure analysis
3. Separate reporting - Show context command errors separately from actionable errors

Suggested Context Commands

• pwd, ls, cat, head, tail (exploration)
• which, type, command -v (tool detection)
• echo $VAR, env (environment inspection)

Context

Discovered during workflow analysis on genai-rs - see event bus event #1632.