refactor(task): switch to <feedback> wrapper to prevent focus drift after context-management event (condense/truncate)

[hannesrudolph]

This PR updates Task.startTask() to send the initial user content wrapped in instead of .

This PR switches the initial wrapper to a with the aim of preventing focus drift after context-management event (condense/truncate). The aim of this the most minimal adjustment to address this drift that is likely caused by the user messages which are structured as follows

The user has provided feedback on the results. Consider their input to continue the task, and then attempt completion again.\n<feedback>\nUSER MESSAGE GOES HERE\n</feedback>

This hints to the model to change focus to the "" which only the very first message is wrapped in. Condensing used to condense and remove the first message so this mention of task was seen as more of a general reference to the "task" and not the specific wrapped user input. Since we changed this the original block is much more likely to steal focus and this prevents the session from properly continuing from after the condensing summary as intended.

Why:

• Prevent focus drift after a successful context-management event by signaling the next user block as feedback rather than a new task.
• Aligns with existing flows that emit (attempt-completion and standard responses).
• Mentions parsing already treats as a trigger, so enrichment behavior is unchanged.

Impact:

• No change to tool execution, streaming, checkpointing, or persistence.
• Unit tests for mentions continue to pass as both and are supported.

---

Important

Refactor Task.startTask() to use <feedback> instead of <task> to prevent focus drift, updating related tests and maintaining existing functionality.

• Behavior:
  • Task.startTask() in Task.ts now wraps initial user content in <feedback> instead of <task> to prevent focus drift.
  • processUserContentMentions.ts updated to process <feedback> tags instead of <task>.
• Tests:
  • Updated processUserContentMentions.spec.ts to reflect changes from <task> to <feedback>.
  • Updated Task.spec.ts to ensure task processing aligns with new <feedback> tag usage.
• Impact:
  • No changes to tool execution, streaming, checkpointing, or persistence.
  • Unit tests confirm both <task> and <feedback> are supported.

^{This description was created by for 7089e11. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Rooviewer   See task on Roo Cloud

Review completed. Found 1 issue to address:

• Update <task> wrapper in built-in-commands.ts to align with feedback standardization

Previous reviews

• 812d52d: Review #1

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[mrubens]

I think there are a few other places we refer to <task> - can we clean those up too so other people don't get confused?