home/writing/fix-prompt-generation-4

Fix Prompt Generation

GPTClaudeGemini··164 copies·updated 2026-07-14
fix-prompt-generation-4.prompt
# Attest LLM Fix Prompt Generation

**Purpose:** Attest-specific action verbs, suppression cases, template fields, and worked example for the `## LLM Fix Prompt` block when Attest confirms an AC gap and hands remediation to Builder (or Scribe/Accord for spec rewrites).
**Read when:** Attest has issued a per-criterion verdict of `FAIL` or `PARTIAL` and remediation must be paired with a paste-ready prompt for the receiving agent.

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

## Contents

- When Attest emits a Fix Prompt vs withholds
- Attest action verbs
- Verb selection heuristic
- Attest-specific suppression cases
- Per-finding fix prompt template (Attest-specific fields)
- Worked example

---

## When Attest Emits a Fix Prompt vs Withholds

Attest never modifies code. Every confirmed AC gap with sufficient evidence pairs with a Fix Prompt addressed to the agent best positioned to remediate. The prompt is withheld only in the suppression cases below.

| Situation | Action |
|-----------|--------|
| Per-criterion verdict is `FAIL` or `PARTIAL` and the implementation must change | Emit `CLOSE-GAP` (or `BREAKING-CLOSE` if scope is wider) — Builder |
| Implementation behavior is correct but the spec is wrong/outdated/missing the criterion | Emit `RECONCILE-SPEC` — Scribe / Accord |
| AC interpretation is ambiguous and a code change without clarification would miss intent | Emit `INVESTIGATE-FURTHER` — spec author or Attest re-entry |
| AC is not applicable in the current context (deprecated, out-of-scope feature flag) | Emit `WAIVE` — Builder + Scribe (waiver doc) |
| Verification-only mode requested (compliance verdict only, no remediation scope) | **Suppress prompt.** Note "Fix prompt withheld per scope: verification only." |
| Multiple ACs require coordinated spec rewrite | **Suppress per-finding prompts.** Hand the whole bundle to Scribe/Accord. |
| Implementation passes all ACs (no gaps) | **Suppress prompt.** Note "Fix prompt N/A — full conformance verified." |
| Verdict is `NOT_TESTED` (runtime-only) and no static gap is established | **Suppress prompt.** Hand off to Voyager/Radar with runtime plan instead. |

The `ATTEST_TO_BUILDER_HANDOFF` and `ATTEST_TO_SCRIBE_HANDOFF` payloads carry a `fix_prompt` field; populate it whenever the verdict is `FAIL` or `PARTIAL` and a suppression case does not apply.

---

## Attest Action Verbs

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

| Verb | When to use | Receiving agent |
|------|-------------|----------------|
| `CLOSE-GAP` | Implementation is missing an AC; scoped fix to satisfy the AC | Builder |
| `RECONCILE-SPEC` | Implementation behavior is correct but the spec is wrong/outdated; update spec instead of code | Scribe / Accord |
| `BREAKING-CLOSE` | Closing the gap requires a breaking change (API contract, behavior change visible to clients) | Builder + Guardian + Launch |
| `INVESTIGATE-FURTHER` | AC interpretation ambiguous; need to clarify with spec author / stakeholder before changing code | Spec author OR Attest re-entry with clarified spec |
| `WAIVE` | AC not applicable in the current context; document waiver with rationale | Builder + Scribe (waiver doc) |

---

## Verb Selection Heuristic

when to use it

Community prompt sourced from the open-source GitHub repo simota/agent-skills (MIT). A "Fix Prompt Generation" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.

tags

writingcommunitygeneral

source

simota/agent-skills · MIT