phase-10a-activity-correctness-reviewer — Claude Code prompt
The phase-10a-activity-correctness-reviewer prompt asks Claude Code to review Phantom’s Phase 10A patch and report concrete findings with file and line references. It targets changes in the named project.
Apache-2.0 ★ 1,497 updated 2026-06-16
.claude/commands/phase-10a-activity-correctness-reviewer.mdWhat phase-10a-activity-correctness-reviewer does
The reviewer reads the specified project documents, git diff, changed source and test files, then uses rg to inspect related behavior where needed. Its checks cover activity state across turns, tool-result translation, UI behavior, SSE handling and session resume. It must leave files untouched and make no commits.
The report puts findings in severity order and gives each one an exact path and line reference. It also includes a short account of tests inspected or run, and checks for exposed internals and risky gaps in test coverage. Project and prerequisite file paths are fixed in the prompt, so it is meant for a workspace with that Phantom repository available.
What it helps with
- spot correctness bugs in the Phase 10A patch
- check activity state across turns in one session
- catch stale UI after a run ends
- verify tool-result translation
- assess test coverage for risky edge cases
Example prompts for the agent
- Review the Phase 10A patch in Phantom with
/phase-10a-activity-correctness-reviewer. - Use
/phase-10a-activity-correctness-reviewerto check activity state across session turns. - Run
/phase-10a-activity-correctness-reviewerand inspect how tool results are handled.
How to install phase-10a-activity-correctness-reviewer
Manually: put the files in place
.claude/commands/phase-10a-activity-correctness-reviewer.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about phase-10a-activity-correctness-reviewer
How do I install phase-10a-activity-correctness-reviewer in Claude Code?
Add the command file at .claude/commands/phase-10a-activity-correctness-reviewer.md.
What does phase-10a-activity-correctness-reviewer do?
It reviews the Phantom Phase 10A patch against specified files and behavior. The report orders findings by severity and includes file and line references.
Can it fix the issues it finds?
No. The prompt restricts the agent to review: it must not edit files, commit changes or perform broad cleanup.
Source text
phase-10a-activity-correctness-reviewer.md · Apache-2.0
ultrathink. ultrathink. ultrathink. You are a principal engineer, principal architect, and principal product manager at Anthropic, all three at once. Your mission is to review the Phase 10A Activity Correctness patch in `/Users/truffle/work/phantom-chat-experience`. No context anxiety. No token limits. No time pressure. No cost anxiety. No "v2" thinking. There is no v2. Build it right. Take as long as you need. ## Context This patch is the first Phantom chat-experience slice after the Murph runtime integration. It should make long-running Murph turns visibly active in the web chat without rewriting the chat product or inventing a separate telemetry system. You are a reviewer only. Do not edit files. Do not commit. Do not run a broad cleanup. Findings should be concrete, reproducible, and grounded in file and line references. ## Required Reading Read these files before reviewing: 1. `/Users/truffle/.claude/AGENTS.md` 2. `/Users/truffle/.claude/CLAUDE.md` 3. `/Users/truffle/work/phantom-chat-experience/research/chat-experience/phase-10a-synthesis.md` 4. `/Users/truffle/work/phantom-chat-experience/prompts/phase-10a-activity-correctness-builder.md` 5. `git diff` in `/Users/truffle/work/phantom-chat-experience` Then inspect every changed source and test file: - `src/chat/types-tool.ts` - `src/chat/sdk-to-wire.ts` - `src/chat/__tests__/sdk-to-wire.test.ts` - `chat-ui/src/lib/chat-types.ts` - `chat-ui/src/lib/chat-activity.ts` - `chat-ui/src/lib/chat-store.ts` - `chat-ui/src/lib/chat-dispatch-tools.ts` - `chat-ui/src/lib/__tests__/chat-store.test.ts` - `chat-ui/src/hooks/use-chat.ts` - `chat-ui/src/routes/session-route.tsx` - `chat-ui/src/components/message-list.tsx` - `chat-ui/src/components/run-activity-row.tsx` - `chat-ui/src/components/thinking-block.tsx` - `chat-ui/src/components/tool-call-card.tsx` Use `rg` for adjacent behavior as needed, especially `resumeSession`, `message.assistant_start`, `message.tool_call_*`, `session.done`, and the stored message/event model. ## Review Questions 1. Does the patch introduce any P0/P1/P2 correctness bug? 2. Does activity state behave correctly across multiple turns in the same session? 3. Does preserving tool state after terminal events create stale UI or stale active-run placeholder tools? 4. Does `message.tool_call_result` translation correctly map likely SDK and Murph tool-result shapes without breaking normal user messages? 5. Does the UI reveal raw chain-of-thought, secrets, large logs, or unsafe internals by default? 6. Does the implementation respect existing SSE, resume, abort, assistant lifecycle, text reconciliation, and cost rendering contracts? 7. Are the tests meaningful enough, or are there missing focused tests for important edge cases? 8. Are any files now too close to the 300-line standard or too tangled to safely maintain? ## Severity Rubric P0: data loss, security/privacy leak, broken main chat flow, impossible to run. P1: real user-visible regression in common flow, broken long-running activity, lost stream/replay behavior, incorrect tool-result protocol. P2: edge-case regression, stale UI, incomplete test coverage for risky behavior, maintainability issue likely to bite soon. P3: polish or follow-up. ## Acceptance Criteria Final report format: 1. Findings first, ordered by severity. 2. Each finding must include exact file path and line reference. 3. Include a short "tests inspected or run" section. 4. If there are no P0/P1/P2 findings, say that clearly and name residual risks. Do not praise the patch. Be useful, precise, and skeptical. ultrathink. Review the product behavior, not just TypeScript correctness.
© ghostwright · Apache-2.0 · ghostwright/phantom