Skip to main content
prompt GP ChatGPT

phase-10f-codex-review-fix — Claude Code prompt: what it does

phase-10f-codex-review-fix — a Claude slash command that keeps the agent on three Claude review findings from PR 106, with the fix scoped to Phase 10F.

Apache-2.0 ★ 1,497 updated 2026-06-16

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/commands/phase-10f-codex-review-fix.md
Open on GitHub

What phase-10f-codex-review-fix does

The file is a slash command for Claude. It hands the agent a triple role — principal engineer, architect, product manager — and points it at three Claude review findings on PR 106. Two of them share a shape: src/agent/runtime.ts and src/agent/chat-query.ts pass one model to the SDK while reading thinking mode from another. The third sits in src/chat/sdk-to-wire-handlers.ts, where streamed tool JSON must not leak across a reused index when a new tool-use block starts.

Acceptance criteria spell out the finish line. Every Agent SDK callsite has to take its thinking config from the same resolved model it puts in options.model. Tool input buffers get reset whenever a content block opens at an index already in use. Regression tests cover both callsites and the reuse case, and the existing suite, typecheck, lint and build stay green.

Paths and the PR number belong to the phantom repo. Point this at your own files before running it, or the agent will hunt for src/chat/sdk-to-wire-handlers.ts in a tree that never had one.

What it helps with

  • Close three Claude review findings on a specific PR without widening the phase scope
  • Make every Agent SDK callsite read thinking mode from the model it sends
  • Reset streamed tool input buffers when an index gets reused
  • Cover both callsites and the index reuse case with regression tests
  • Keep tests, typecheck, lint and build passing after the fix

Example prompts for the agent

  • Run /phase-10f-codex-review-fix and fix the three review findings without touching anything outside Phase 10F.
  • Use /phase-10f-codex-review-fix, start from src/agent/runtime.ts and show where the thinking mode comes from.
  • /phase-10f-codex-review-fix — add regression coverage for the reused tool index in the stream.

How to install phase-10f-codex-review-fix

Manually: put the files in place

.claude/commands/phase-10f-codex-review-fix.md

Questions about phase-10f-codex-review-fix

How do I install phase-10f-codex-review-fix in Claude Code?

Drop the command file into your project at .claude/commands/phase-10f-codex-review-fix.md. It then runs as /phase-10f-codex-review-fix.

What does phase-10f-codex-review-fix do?

It walks the agent through three Claude review findings on PR 106 and sets acceptance criteria: one model as the source of thinking config, buffer resets on tool input, regression tests.

Which files does the prompt touch?

src/agent/runtime.ts, src/agent/chat-query.ts and src/chat/sdk-to-wire-handlers.ts, the three named in the review findings.

Source text

phase-10f-codex-review-fix.md · Apache-2.0
ultrathink. ultrathink. ultrathink.

# Phase 10F: Codex Review Fix

You are a principal engineer, principal architect, and principal product
manager at Anthropic, all three at once.

## Mission

Address the actionable Codex review findings on PR 106 without widening the
Phase 10F scope.

## Findings

1. `src/agent/runtime.ts` sends `queryModel` to the SDK but derives thinking
   mode from `config.model`.
2. `src/agent/chat-query.ts` sends `queryModel` to the SDK but derives
   thinking mode from `deps.config.model`.
3. `src/chat/sdk-to-wire-handlers.ts` needs explicit protection against stale
   streamed tool JSON when a new tool-use block reuses an index.

## Acceptance Criteria

1. Every Agent SDK callsite derives thinking config from the same resolved
   model it sends as `options.model`.
2. Streamed tool input buffers are reset when a new content block starts at an
   already-used index.
3. Regression tests cover the runtime callsite, chat callsite, and streamed
   tool index reuse.
4. Existing tests, typecheck, lint, and build still pass.

© ghostwright · Apache-2.0 · ghostwright/phantom

Similar prompts

All prompts for Code review and refactoring →

From the same set38

The whole set →
phase-10i-transcript-recovery-final-re-review phase-10i-transcript-recovery-final-re-review is a Claude Code prompt for re-reviewing transcript recovery and verifying attachment metadata redaction fixes. GP ChatGPT ★ 1,497repo stars phase-10i-transcript-recovery-review The phase-10i-transcript-recovery-review prompt audits transcript recovery after Murph compaction, context handling, and secret exposure. Use it to review the Phase 10I slice before commit. GP ChatGPT ★ 1,497repo stars phase-10f-run-console-ui phase-10f-run-console-ui — a prompt for building a run console UI in Phantom on Murph: displays live tool activity, counts, and session state with durable rehydration after refresh. Keeps the chat responsive during long operations. GP ChatGPT ★ 1,497repo stars phase-10h-phantom-chat-review The phase-10h-phantom-chat-review prompt runs an in-depth audit of Phantom's chat UI and event pipeline after PR 109. It flags UI gaps and builds an implementation slice without modifying code. GP ChatGPT ★ 1,497repo stars list-plugins list-plugins reads settings.json and reports active Claude Code plugins with their marketplace IDs. Use it to check which plugins are currently enabled. CX Codex +1 ★ 1,497repo stars phase-10b-durable-timeline-reviewer phase-10b-durable-timeline-reviewer — a specialized prompt for reviewing Phase 10B of Phantom's chat system: validates timeline correctness, durability, idempotency, and data privacy. Reads diffs, identifies security issues and regressions. GP ChatGPT ★ 1,497repo stars phase-10h-chat-integrity-review phase-10h-chat-integrity-review reviews uncommitted git diffs for Phantom chat transcript integrity in Claude Code. It inspects attachment persistence and message rendering without modifying project code. GP ChatGPT ★ 1,497repo stars phase-10h-pi-thinking-research The phase-10h-pi-thinking-research prompt asks Claude Code to examine how Pi presents model thinking, tool activity, and long-running work, then write chat UI guidance for Phantom. GP ChatGPT ★ 1,497repo stars

More in the catalog

Project sponsors

Companies that keep this analytics open