Task Report

---
name: task-report
description: Record the results of a workhub task - store the raw report in the vault AI zone, polished notes in the human zone, update the task's result section, and set status to review. Use when finishing work started via task-start, or when the user asks to report/close out a task.
argument-hint: "<task-id>"
---

# task-report — Record task results into the workhub vault

## Steps

1. **Resolve the vault** (same order as the other task skills:
   `WORKHUB_VAULT` env var → the current directory if it is a vault →
   `vault_path` in `%APPDATA%\workhub\config.json`) and locate the task
   file by `id`.
2. **Write the raw report** to `<vault>/_ai/logs/<task-id>-<yyyymmdd>.md`:
   what was done, key decisions, files changed (as `path:line` references),
   verification results, and anything a future agent needs to resume.
3. **Write human-readable deliverables** where they belong:
   - Knowledge gained (research, gotchas, how-tos) → a note under
     `<vault>/knowledge/`, added to `knowledge/_index.md`.
   - Project-specific outcomes → a note under `<vault>/projects/`.
   - Keep these polished and short; link to the raw log with a wikilink
     only if the detail matters.
4. **Append to the task's `## Results` section** (Edit tool): a 2-4 line
   summary and wikilinks to the notes created in step 3. Change nothing
   else in the body.
5. **Feed the owner's judgement calls back into the policy.** If the owner
   settled anything during this task — a question you put to them, a
   correction to your approach, a preference they stated in passing — append
   the rule it establishes to `<vault>/profile/decision-policy.md`:
   `## Past decisions` for a one-off call
   (`- <date> <task-id> <the rule> (from: <the question>)`), `## Preferences`
   for a standing leaning. Say in the report which lines you added. This is
   what stops the next task asking the same question.
6. **Close out the status** with the bundled CLI (preferred — it sets
   `status: review` + `updated`, clears the active-task marker, and
   refreshes the index in one step):

   ```bash
   node "${CLAUDE_PLUGIN_ROOT}/scripts/task-cli.mjs" report <task-id>
   ```

   *Fallback (no node, or script missing):* set `status: review` and
   `updated: <today>` in the frontmatter by hand (preserve the rest), and
   delete `<vault>/_ai/memory/active-task.json` if it refers to this task.
7. **Offer to clean up the worktree — only for worktree-mode tasks**
   (`worktree: true`). Once the work is committed/pushed and no longer needed,
   **propose** removing the task's worktree (do not delete it automatically —
   the user may still want to inspect it):

   ```bash
   git -C <repo> worktree remove "<repo>/../.worktrees/<repo-name>/<task-id>"
   ```

   Mention the branch `task/<task-id>` is left in place for the PR/merge.

## Rules

- AI never sets `status: done` — a human does that in the workhub app after
  reviewing.
- If the work is incomplete or blocked, still report: describe the blocker in
  `## Results`, keep `status: doing`, and leave the active-task marker in place.
  When the blocker is a question filed for the owner (`_ai/comms/`), name the
  question id and mark the task blocked so the board shows the wait:
  `task-cli.mjs update <id> --blocked true --blocked-note "waiting on Q-0001"`.
- Do not overwrite existing human notes; create new ones and link them.
- Never modify the task's `## Plan` section — it is the approved plan and the
  user's approval record, not a place for results.
- Write `## Results` in the language given by the workhub **Task language**
  setting, stated in the app's launch prompt (default English). That setting
  governs the task file's `## Plan` and `## Results` only — notes under
  `knowledge/` and `projects/`, and everything in target repositories, follow
  their own conventions.