feat(explain): scope transcript and prompts to checkpoint boundaries

[khaong]

Summary

• When viewing a checkpoint with entire explain -c {checkpoint}, prompts and AI-generated summaries now show only content relevant to that specific checkpoint
• Uses the existing TranscriptLinesAtStart metadata field to slice the transcript at checkpoint boundaries
• Adds SliceFromLine() utility to the transcript package for byte-level slicing

Test plan

• Unit tests for SliceFromLine() covering edge cases (zero start, out of bounds, empty content, no trailing newline)
• Integration tests for scopeTranscriptForCheckpoint() and extractPromptsFromTranscript()
• End-to-end test TestFormatCheckpointOutput_UsesScopedPrompts verifying scoped prompts appear correctly
• Manual test: run entire explain -c {checkpoint} on a checkpoint with multiple prior prompts and verify only relevant prompts are shown

🤖 Generated with Claude Code

---

Note

Medium Risk
User-visible explain output and summary generation now depend on correct transcript slicing; off-by-one or missing metadata could hide prompts or prevent summary generation. Changes are localized to CLI formatting/utilities but touch checkpoint listing behavior and parsing utilities.

Overview
entire explain --checkpoint now scopes prompts and generated summaries to the checkpoint’s portion of the transcript by slicing from TranscriptLinesAtStart, instead of using the full session transcript.

Adds transcript.SliceFromLine() for byte-level JSONL slicing, updates branch checkpoint listing to derive the displayed session prompt from the scoped transcript, and extends tests to cover slicing edge cases plus scoped-prompt behavior and backwards-compatible fallback to stored prompts when no transcript exists.

^{Written by Cursor Bugbot for commit 0e7a05f. This will update automatically on new commits. Configure here.}

[Copilot]

Pull request overview

This PR updates the entire explain -c checkpoint view so that both AI summaries and displayed prompts are scoped to the transcript region corresponding to that specific checkpoint, using TranscriptLinesAtStart to slice the cumulative transcript.

Changes:

• Added SliceFromLine() in the transcript package to perform byte-level slicing of JSONL transcript content from a given 0-based line offset, with comprehensive unit tests covering edge cases.
• Updated checkpoint explanation and summary generation in explain.go to scope transcripts to TranscriptLinesAtStart, then derive prompts from the scoped transcript via extractPromptsFromTranscript.
• Extended explain tests to validate transcript scoping, prompt extraction, and that verbose checkpoint output only shows prompts that fall within the checkpoint’s scoped transcript region.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
cmd/entire/cli/transcript/parse.go	Adds SliceFromLine helper to slice JSONL transcript bytes starting at a given 0-based line index, used to implement checkpoint-aware transcript scoping.
cmd/entire/cli/transcript/parse_test.go	Introduces unit tests for SliceFromLine, validating correct skipping behavior, zero-offset behavior, out-of-bounds handling, empty content, and no-trailing-newline scenarios.
cmd/entire/cli/explain.go	Adjusts summary generation and checkpoint formatting to use a scoped transcript (scopeTranscriptForCheckpoint) and derive user prompts from it (extractPromptsFromTranscript), ensuring intent and prompts reflect only the current checkpoint.
cmd/entire/cli/explain_test.go	Updates and adds tests to cover verbose output with transcripts, transcript scoping behavior, prompt extraction from scoped transcripts, and that verbose explain output only includes prompts within the checkpoint’s scoped region.

[khaong]

Response to PR Review

Re: Full mode transcript not scoped to checkpoint boundaries

This is intentional behavior. The --full flag is meant to show the complete session transcript for context, while prompts are scoped to the checkpoint. This allows users to see exactly what happened during a checkpoint (scoped prompts) while still being able to reference the full conversation history if needed.

Added TestFormatCheckpointOutput_FullShowsEntireTranscript to document this as expected behavior.

---

Re: Intent fallback ignores existing prompts when transcript missing

Fixed in 265cf2d. Added backwards compatibility fallback: when a checkpoint has stored prompts but no transcript (older checkpoints), we now fall back to using result.Prompts instead of showing "(not generated)".

Added TestFormatCheckpointOutput_FallsBackToStoredPrompts to verify this behavior.

[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.}

[cursor]

Missing backwards compatibility fallback for checkpoint prompts

Medium Severity

In getBranchCheckpoints, when extracting prompts for the branch list view, the code only uses scopedPrompts from the transcript but lacks a fallback to result.Prompts when the scoped prompts are empty. This is inconsistent with formatCheckpointOutput which explicitly falls back to result.Prompts for backwards compatibility with older checkpoints (lines 525-530 and 569-571). Older checkpoints without transcripts will show "(no prompt)" in the branch list view, but display correctly in the detailed checkpoint view.