fix(session): render requested review patches

[benvinegar]

Summary

• render resolved unified diff bodies in human-readable hunk session review --include-patch output
• omit the dangling Notes: heading when --include-notes finds no notes
• add formatter regression coverage and a patch changeset

Verification

• bun run typecheck
• bun run lint
• bun test src/session/agent/cliClient.test.ts src/session/agent/commands.test.ts src/session/broker/projections.test.ts src/session/broker/reviewResources.integration.test.ts
• bunx oxfmt --check src/session/agent/cliClient.ts src/session/agent/cliClient.test.ts .changeset/quiet-patches-speak.md

This PR description was generated by Pi using gpt-5.6-sol

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
hunk-web	Ignored	Preview	Aug 31, 2026 7:01pm

[greptile-apps]

Greptile Summary

This PR renders resolved unified patches in human-readable session-review output, suppresses empty note headings, and adds formatter regression coverage plus a patch changeset.

• Adds optional patch bodies beneath each review file.
• Emits the Notes: section only when notes are present.
• Covers patch and note formatting behavior in colocated tests.

Confidence Score: 3/5

This PR is not safe to merge until human-readable patch output is sanitized against terminal control-sequence injection.

Resolved patches contain repository-derived raw text and now flow unchanged into stdout, bypassing the terminal sanitization used by analogous diff-rendering paths.

Files Needing Attention: src/session/agent/cliClient.ts

Security Review

Human-readable patch output is emitted without terminal sanitization, allowing terminal control sequences from reviewed content to alter or spoof the displayed review. How this was verified: The raw patch was traced from resource decoding through the formatter’s unchanged text branch to the direct stdout write, while the analogous diff renderer sanitizes the same class of content.

Important Files Changed

Filename	Overview
src/session/agent/cliClient.ts	Adds patch rendering and empty-note suppression, but sends untrusted patch bodies to terminal output without control-sequence sanitization.
src/session/agent/cliClient.test.ts	Adds useful regression coverage for patch preservation and conditional note headings, but does not cover terminal-control input.
.changeset/quiet-patches-speak.md	Correctly records the formatter behavior changes as a patch release.

Prompt To Fix All With AI

### Issue 1
src/session/agent/cliClient.ts:512
**Patch output bypasses sanitization**

When a reviewed text patch contains terminal control sequences, `formatReviewOutput` writes them unchanged to stdout, allowing the patch to clear, reposition, or overwrite terminal content and spoof the displayed review. **How this was verified:** The raw patch was traced from resource decoding through the unchanged text-output branch to the direct stdout write, while the analogous diff renderer sanitizes the same class of content.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

_{Reviews (1): Last reviewed commit: "fix(session): render requested review pa..." | Re-trigger Greptile}

[greptile-apps]

Patch output bypasses sanitization

When a reviewed text patch contains terminal control sequences, formatReviewOutput writes them unchanged to stdout, allowing the patch to clear, reposition, or overwrite terminal content and spoof the displayed review. How this was verified: The raw patch was traced from resource decoding through the unchanged text-output branch to the direct stdout write, while the analogous diff renderer sanitizes the same class of content.

Knowledge Base Used: Agent command orchestration

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/session/agent/cliClient.ts
Line: 512

Comment:
**Patch output bypasses sanitization**

When a reviewed text patch contains terminal control sequences, `formatReviewOutput` writes them unchanged to stdout, allowing the patch to clear, reposition, or overwrite terminal content and spoof the displayed review. **How this was verified:** The raw patch was traced from resource decoding through the unchanged text-output branch to the direct stdout write, while the analogous diff renderer sanitizes the same class of content.

**Knowledge Base Used:** [Agent command orchestration](https://app.greptile.com/modem/-/custom-context/knowledge-base/modem-dev/hunk/-/docs/agent-command-orchestration.md)

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

[benvinegar]

Fixed in 462bbc8. Human-readable patch output now passes through sanitizeTerminalText, which strips terminal control sequences while preserving patch newlines and tabs. Added a regression test covering a terminal-clear sequence and diff layout preservation.

_{Responded by Pi using openai/gpt-5.6-sol.}

This comment was generated by Pi using gpt-5.6-sol