phase-10i-transcript-recovery-review — Claude Code prompt
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.
Apache-2.0 ★ 1,497 updated 2026-06-16
.claude/commands/phase-10i-transcript-recovery-review.mdWhat phase-10i-transcript-recovery-review does
The prompt assigns Claude Code a combined code review, architecture, product, and security brief. It points the agent to three production files, their focused tests, and two Phase 10I research documents. The review must rely on the repository contents rather than summaries.
It asks whether recovery remains safe after Murph compaction, whether the current session id reaches the provider without unnecessary context growth, and whether the reflective search tool has suitable bounds and citations. Query, role, sequence-window, and limit handling are checked alongside possible exposure of magic links, API keys, base64 data, attachments, and hidden runtime records.
The agent must inspect or run the listed bun test, bun run lint, and bun run typecheck commands, then write a report to research/chat-experience/phase-10i-transcript-recovery-review.md. Production files must remain unchanged and no commit should be created. The review is tied to the repository path /Users/truffle/work/phantom-murph-hardening, so a different checkout changes what can be verified.
What it helps with
- audit transcript recovery after Murph compaction
- check session-id injection against provider-context growth
- detect exposed links, keys, attachments, or runtime records
- verify query, role, sequence-window, and result-limit rules
- classify findings from P0 through P3
Example prompts for the agent
- Run
/phase-10i-transcript-recovery-reviewagainst the Phase 10I transcript recovery slice. - Read the listed source files, tests, and research, then run the required
bun test,bun run lint, andbun run typecheckchecks. - Write the review to
research/chat-experience/phase-10i-transcript-recovery-review.md; do not edit production code or commit changes.
How to install phase-10i-transcript-recovery-review
Manually: put the files in place
.claude/commands/phase-10i-transcript-recovery-review.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about phase-10i-transcript-recovery-review
How do I install phase-10i-transcript-recovery-review in Claude Code?
Place the command file at .claude/commands/phase-10i-transcript-recovery-review.md in the Claude Code project.
What does phase-10i-transcript-recovery-review do?
It reviews the Phase 10I transcript recovery slice across implementation, tests, and research. The checks cover compaction recovery, provider context, search bounds, citations, and sensitive-data exposure.
Which verification commands does the prompt require?
It requires bun test src/chat/__tests__/transcript-search.test.ts src/chat/__tests__/continuity-context.test.ts src/agent/__tests__/reflective-transcript-tools.test.ts, followed by bun run lint and bun run typecheck.
Source text
phase-10i-transcript-recovery-review.md · Apache-2.0
ultrathink. ultrathink. ultrathink. You are a principal engineer, principal architect, principal product manager, and security reviewer at Anthropic, all at once. Review the Phase 10I transcript recovery slice in `/Users/truffle/work/phantom-murph-hardening`. No context anxiety. No time pressure. Be exact. Read actual files and tests. Do not rely on summaries. ## Scope Review these changed production files: 1. `src/chat/transcript-search.ts` 2. `src/chat/continuity-context.ts` 3. `src/agent/in-process-reflective-tools.ts` Review these tests: 1. `src/chat/__tests__/transcript-search.test.ts` 2. `src/chat/__tests__/continuity-context.test.ts` 3. `src/agent/__tests__/reflective-transcript-tools.test.ts` Also read relevant current research: 1. `research/chat-experience/phase-10i-memory-architecture-research.md` 2. `research/chat-experience/phase-10i-transcript-search-research.md` ## Questions 1. Does this implement a production-safe transcript recovery path after Murph compaction? 2. Is the current session id injected without bloating provider context? 3. Is the agent tool bounded, cited, and redacted enough? 4. Could it leak magic links, API keys, raw base64, attachment payloads, or hidden runtime records? 5. Are query, role, seq window, and limit behavior correct? 6. Does the implementation misuse Murph or Pi instead of building Phantom-specific glue? 7. Are there P0, P1, or P2 issues that must be fixed before commit? 8. Are there lower-priority follow-ups that should be documented but not block this slice? ## Required Verification Run or inspect results for: - `bun test src/chat/__tests__/transcript-search.test.ts src/chat/__tests__/continuity-context.test.ts src/agent/__tests__/reflective-transcript-tools.test.ts` - `bun run lint` - `bun run typecheck` ## Output Write a concise review report to: `/Users/truffle/work/phantom-murph-hardening/research/chat-experience/phase-10i-transcript-recovery-review.md` Use severity labels P0, P1, P2, P3. If there are no P0/P1/P2 findings, say that clearly. Do not edit production code. Do not commit. ultrathink. This is a trust boundary. Be skeptical.
© ghostwright · Apache-2.0 · ghostwright/phantom