Add option to manually attach a session

[computermode]

Summary

Adds a new attach command accessible via entire attach <session-id>. Enables users to retroactively capture an agent session that wasn't tracked (ie if the agent was created outside of the project's root directory).

The attach command allows for an optional -a or --agent flag to set the agent. It can auto-detect the agent type for sessions from Gemini and Claude. It also allows for a -f or --force flag that will append the agent session to the user's last commit message. Otherwise, the user has the option to either append or copy the checkpoint ID themselves via the dialog prompts.

Notes/known issues

• session_turn_count defaults to 0 if we can't find it in the metadata. I don't believe we display the session turns in the UI (we display steps) so I don't believe this is presented to the user. That data is not always captured - not sure how big of a deal this is for the first pass through/version for this command.
• Opencode sessions are already tracked by Entire when they're started from sub-folders. The attach command works for Opencode sessions that weren't tracked in the project folder (say, if Entire was enabled after Opencode was installed), but it won't detect Opencode sessions from folders outside the project. (attach doesn't care about folder location for other agents: this is isolated to Opencode atm)

Test plan

• Unit tests for missing session ID, transcript not found, success path, duplicate session detection, and checkpoint ID output
• Manual test: entire attach <real-session-id> with Claude sessions
• Manual test: entire attach --force <session-id> amends the latest commit without prompting
• Manual test: entire attach --agent gemini <session-id> with Gemini sessions
• Manual test: entire attach --agent factoryai-droid <session-id> with Droid sessions
• Manual test: entire attach --agent cursor <session-id> with Cursor sessions
• Manual test: entire attach <gemini-session-id> with Gemini sessions
• Checkpoints for Claude sessions are viewable in entire.io w/ all the proper metadata
• Checkpoints for Gemini sessions are viewable in entire.io w/ all the proper metadata
• Checkpoints for Factoryai-droid sessions are viewable in entire.io w/ all the proper metadata
• Checkpoints for cursor sessions are viewable in entire.io w/ all the proper metadata
• Verify entire rewind works on attach-created checkpoints

[Copilot]

Pull request overview

Adds a new entire attach <session-id> command to retroactively capture an agent session (from an existing transcript) and turn it into an Entire checkpoint + tracked session.

Changes:

• Registers a new attach Cobra command and implements session attachment + checkpoint creation/condensation flow.
• Adds transcript-path fallback logic to search Claude project directories when sessions were started from subdirectories.
• Adds unit tests covering several attach scenarios (missing args, transcript missing, success path, duplicate session, output includes trailer).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 10 comments.

File	Description
cmd/entire/cli/transcript.go	Adds helper functions to locate session transcripts (including subdirectory search).
cmd/entire/cli/root.go	Registers the new attach command on the root CLI.
cmd/entire/cli/attach.go	Implements attach end-to-end: resolve transcript, extract files/prompt, save checkpoint, condense, optionally amend commit.
cmd/entire/cli/attach_test.go	Introduces unit tests for attach behaviors and outputs.

Comments suppressed due to low confidence (1)

cmd/entire/cli/transcript.go:96

• sanitizePathForSearch does not match Claude Code's SanitizePathForClaude behavior for non-ASCII paths (this implementation replaces each UTF-8 byte with '-', while SanitizePathForClaude replaces each non-alphanumeric rune once via regex). This can cause searchTranscriptInSubdirs to compute the wrong repoPrefix and fail to find transcripts for repos under paths containing non-ASCII characters. Prefer reusing the existing sanitizer (e.g., paths.SanitizePathForClaude / claudecode.SanitizePathForClaude) instead of duplicating the logic here.

}

// FindCheckpointUUID finds the UUID of the message containing the tool_result
// for the given tool_use_id. This is used to find the checkpoint point for
// transcript truncation when rewinding to a task.
// Returns the UUID and true if found, empty string and false otherwise.
func FindCheckpointUUID(transcript []transcriptLine, toolUseID string) (string, bool) {
	for _, line := range transcript {
		if line.Type != transcriptTypeUser {
			continue
		}

		var msg userMessageWithToolResults

[Copilot]

Pull request overview

Copilot reviewed 4 out of 5 changed files in this pull request and generated 5 comments.