Fix Prompt Generation

# Trail LLM Fix Prompt Generation

**Purpose:** Trail-specific action verbs, suppression cases, template fields, and worked example for the `## LLM Fix Prompt` block at the end of every Trail investigation report.
**Read when:** You are writing the `## LLM Fix Prompt` block for a Trail report, choosing an action verb, or deciding whether to suppress.

> Universal authoring rules and prompt structure: `_common/LLM_PROMPT_GENERATION.md`.
> This file documents only Trail-specific verbs, suppression cases, template fields, and an example.

## Contents

- Trail action verbs
- Verb selection heuristic
- Trail-specific suppression cases
- Per-regression fix prompt template (Trail-specific fields)
- Worked example

---

## Trail Action Verbs

Each fix prompt declares one verb at the top of `# Your task`.

| Verb | When to use | Receiving agent / LLM |
|------|-------------|----------------------|
| `FIX-REGRESSION` | HIGH confidence breaking commit identified, forward fix is straightforward | Builder, Claude, Codex |
| `REVERT` | Breaking commit is isolated, dependent changes are minimal, safe to revert wholesale (use `git revert`, never `reset --hard` on shared history) | Builder + Guardian |
| `REVERT-WITH-FORWARD-FIX` | Breaking commit must be reverted to stop the bleeding, but the original intent was valid — forward fix needed afterward | Builder (revert + re-implement) |
| `INVESTIGATE-FURTHER` | Bisect inconclusive, multiple suspect commits, or non-deterministic reproduction — receiving LLM must verify before changing code | Claude / Codex (investigation mode), or Trail (re-entry) |
| `REFACTOR-FIX` | Regression reflects a structural design issue; spot fix would re-introduce the bug elsewhere | Atlas → Builder |

---

## Verb Selection Heuristic