angry-diagnose

/angry-diagnose Command

Adversarial bug investigation pipeline. Generates competing hypotheses for a bug's root cause using external LLMs, then systematically proves/disproves each hypothesis via failing tests before implementing the fix.

1. Argument Parsing

Parse the invocation arguments:

• @file references (optional, multiple) -- Error logs, stack traces, suspect source files as context.
• Prompt text (required) -- Description of the problem/symptom.
• --auto (optional) -- Skip clarifying questions, auto-accept findings, test all hypotheses.
• --max-hypotheses N (optional, default: 5) -- Maximum number of hypotheses to generate.
• --help (optional) -- Print command reference and exit.

--help Output

If --help is passed, print the following and stop:

angry-diagnose — Adversarial bug diagnosis with differential diagnosis and hypothesis-driven TDD.

Usage:
  /angry-diagnose [@context-files...] "problem description" [--auto] [--max-hypotheses N]

Examples:
  /angry-diagnose "the API returns 500 on large payloads"
  /angry-diagnose @error.log "users can't login after password reset"
  /angry-diagnose @src/auth.ts @error.log "this middleware leaks sessions"

Options:
  --auto                Skip questions, auto-accept findings, test all hypotheses.
  --max-hypotheses N    Max competing hypotheses to generate (default: 5).

Phases:
  1. INVESTIGATE    Gather context, build case file, interview user.
  2. DIAGNOSE       Generate competing hypotheses via external LLMs (differential diagnosis).
  3. FIX            Prove/disprove hypotheses via diagnostic tests, fix confirmed root cause.
  4. VERIFY         Mechanical gates + adversarial verification of the fix.

State: .ralph-state/ (gitignored)
Artifacts: .planning/diagnosis/ (gitignored)

If no prompt text is provided and --help is not passed, report the error and display the help text.

2. Environment Validation

Run via the Bash tool:

bash ${CLAUDE_PLUGIN_ROOT}/scripts/checks/validate-env.sh