fix(review): include checkpoint context and review flow fixes

[peyton-alt]

https://entire.io/gh/entireio/cli/trails/312

Summary

• include compact checkpoint context in generated entire review prompts, preferring checkpoint summaries and falling back to stored prompts without inlining raw transcripts
• repair the review configuration flow so built-in commands/plugin skills stay visible, multi-agent selections persist per agent, and additional instructions augment selected skills
• restore live review output: single-agent runs show progress, Codex /review expands into a scoped prompt sent through codex exec -, and Codex assistant/tool events stream while the process is running
• harden final review rendering so Markdown code blocks keep styling without Chroma panics; renderer failures fall back to raw markdown instead of crashing after agents finish

Why the diff is larger than the fixes look

• The user-facing issue crosses picker config, agent adapters, and final output rendering. Most added lines are regression tests for those paths.
• Codex needed parser coverage for live assistant text, tool calls, duplicate summary filtering, and streaming before EOF. That accounts for much of the added test surface.
• This is not a broad review refactor. The changes stay within the review command, Codex/Claude reviewer tests, and the shared markdown renderer used by review dumps.

Tests

• go test ./cmd/entire/cli/mdrender ./cmd/entire/cli/agent/codex ./cmd/entire/cli/review -count=1
• mise run lint
• mise run build
• manual smoke with rebuilt binary: entire review --edit and entire review verified the configured Claude Code + Codex flow, final Codex output, and no renderer panic after agents completed

---

Note

Medium Risk
Adds dynamic, git- and checkpoint-derived context into all generated review prompts, which could affect agent behavior and relies on best-effort parsing of repo history/metadata. Risk is moderated by guardrails (scope required, truncation, no raw transcript inlining) and added tests.

Overview
Generated entire review prompts now include a best-effort “checkpoint context” section derived from checkpoints found in commits within the detected branch scope.

This threads a new RunConfig.CheckpointContext through both single- and multi-agent paths, appends it in ComposeReviewPrompt (while keeping PromptOverride verbatim), and wires the implementation via review_bridge.go to avoid import cycles.

A new review_context.go builds the compact context by scanning scoped commits for Entire-Checkpoint trailers, preferring stored checkpoint summaries and falling back to prompts, truncating detail, and pointing users to entire explain <id> for full context; tests cover prompt composition and an end-to-end smoke capture of the emitted prompt.

^{Reviewed by Cursor Bugbot for commit 0c630e2. Configure here.}

[Copilot]

Pull request overview

This PR enhances entire review prompt generation by appending compact, best-effort checkpoint-derived context for commits within the detected review scope, so reviewers can reference relevant checkpoint IDs and summaries without inlining raw transcripts.

Changes:

• Add CheckpointContext to review run configuration and append it to composed prompts (while keeping PromptOverride verbatim).
• Introduce checkpoint-context extraction for the review scope, preferring checkpoint summaries and falling back to stored prompts, with guidance to use entire explain for full context.
• Wire the new context provider through dependency injection and add unit/smoke tests covering prompt composition and end-to-end prompt capture for a stubbed agent.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
cmd/entire/cli/review/types/reviewer.go	Extends review run configuration with checkpoint context.
cmd/entire/cli/review/tui_model.go	Updates nolint rationale comments for Bubble Tea interface returns.
cmd/entire/cli/review/prompt.go	Appends checkpoint context to generated prompts (non-override path).
cmd/entire/cli/review/prompt_test.go	Adds coverage ensuring checkpoint context is included and override remains verbatim.
cmd/entire/cli/review/cmd.go	Injects and threads checkpoint context into single- and multi-agent run configs.
cmd/entire/cli/review/cmd_test.go	Adjusts stub reviewer Start signature to match interface usage style.
cmd/entire/cli/review_context.go	Implements best-effort checkpoint context extraction for a scoped range of commits.
cmd/entire/cli/review_context_test.go	Adds tests for summary/prompt fallback behavior and a smoke test verifying prompt inclusion.
cmd/entire/cli/review_bridge.go	Wires the checkpoint-context provider into review deps from the CLI package.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 0c630e2. Configure here.}

[alishakawaguchi]

🥇