Fallback to script parsing if prompt wasn't available through hooks

[Soph]

The E2E tests are currently failing for Factory AI Droid because the tests assumes an extracted prompt but using droid exec will not trigger UserPromptSubmit which means we got no prompt. This meant so far an empty prompt was stored.

If we now have no prompt.txt at the end of the turn we fallback to session transcript parsing to extract the prompt.

[cursor]

PR Summary

Medium Risk
Changes turn-end lifecycle behavior to write prompt.txt and potentially update SessionState.LastPrompt based on transcript parsing, which can affect downstream commit message generation and metadata. Risk is moderate due to new control flow and file/state writes, but scoped to the missing-prompt fallback path.

Overview
Adds a new optional PromptExtractor agent interface plus AsPromptExtractor capability gating, enabling agents to provide prompt extraction from transcripts.

On TurnEnd, when prompt.txt is missing/empty, the CLI now attempts to backfill prompts from the transcript (joined with the existing --- separator) and, if successful, updates SessionState.LastPrompt after SaveStep to avoid being overwritten.

Factory AI Droid implements ExtractPrompts by scanning user transcript entries (stripping IDE context tags), and new unit tests cover the backfill behavior, preservation of existing prompt.txt, and session state update.

^{Written by Cursor Bugbot for commit 8fa1853. Configure here.}

[Copilot]

Pull request overview

This PR addresses missing prompt capture for agents (notably Factory AI Droid in exec mode) by backfilling prompt.txt from the session transcript when hooks didn’t provide a prompt, ensuring stored session metadata contains the user prompt.

Changes:

• Add a new agent.PromptExtractor interface plus AsPromptExtractor capability gating helper.
• Backfill prompt.txt during TurnEnd by extracting prompts from the transcript when prompt.txt is absent/empty, and update session state accordingly.
• Add lifecycle and capability tests covering prompt writing/backfill behavior.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
cmd/entire/cli/lifecycle.go	Implements prompt.txt backfill from transcript on TurnEnd and updates session state after SaveStep.
cmd/entire/cli/lifecycle_test.go	Adds tests for prompt file writing, transcript backfill, and session-state update behavior.
cmd/entire/cli/agent/agent.go	Introduces the PromptExtractor interface for transcript-based prompt extraction.
cmd/entire/cli/agent/capabilities.go	Adds AsPromptExtractor helper that gates usage for capability-declaring agents.
cmd/entire/cli/agent/capabilities_test.go	Adds unit tests verifying AsPromptExtractor behavior for different agent types/cap declarations.
cmd/entire/cli/agent/factoryaidroid/lifecycle.go	Declares Factory AI Droid agent implements PromptExtractor.

---

You can also share your feedback on Copilot code review. Take the survey.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{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