@staff0rd/assist 0.798.0 → 0.800.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -65,7 +65,7 @@ After installation, the `assist` command will be available globally. You can als
65
65
  - `/restructure` - Analyze and restructure tightly-coupled files
66
66
  - `/review-config` - Configure this repo's high-level review checklist keys
67
67
  - `/review-ci` - Install a GitHub workflow that reviews new PRs with Claude and Codex through LiteLLM or Azure AI Foundry, without depending on assist
68
- - `/review-high-level [number]` - Open the high-level review checklist for the current branch's PR, or PR `<number>`
68
+ - `/review-high-level [number]` - Open the high-level review checklist for the current branch's PR, or PR `<number>`, as a background task, answering diff questions while it is open
69
69
  - `/review-pr-comments` - Process PR review comments one by one
70
70
  - `/jira [action] [KEY] [args]` - Jira actions: `view`, `associate`, `update`, `started`, `done`, `help`. `[KEY]` defaults to the session's backlog item's
71
71
  - `/github [action] [ref] [args]` - GitHub issue actions: `view`, `edit`, `associate`, `update`, `started`, `done`, `help`. `[ref]` defaults to the session's backlog item's; a bare `/github <ref>` runs `edit`
@@ -134,7 +134,7 @@ Every command supports `--help` for full detail on its flags and behaviour.
134
134
  - `--apply` - Skip posting; walk through each finding asking apply/skip
135
135
  - `--backlog` - Skip posting; file all findings as a single bug backlog item with one phase per finding
136
136
  - `--checkout-only` - Skip the review; check the PR out and leave an idle Claude session in it
137
- - `--high-level` - Skip the LLM review; step through the high-level review checklist in the web preview pane. See [docs/high-level-review.md](docs/high-level-review.md)
137
+ - `--high-level` - Skip the LLM review; step through the high-level review checklist in the web preview pane. With `[number]` outside a Claude session, check the PR out and start a Claude session running `/review-high-level <number>`, which opens the checklist as a background task so the agent can answer questions meanwhile. See [docs/high-level-review.md](docs/high-level-review.md)
138
138
  - `--configure` - With `--high-level`: configure the checklist's `review.highLevel.*` keys instead of reviewing
139
139
  - `--scope <repo|project>` - With `--configure`: write to this repo's override in the shared assist database or the project `assist.yml` instead of asking
140
140
  - `--answer <key=value>` - With `--configure`: answer one key without prompting, repeatable; an empty value leaves the key unset
@@ -1,6 +1,6 @@
1
1
  ---
2
2
  description: Open the high-level review checklist for a PR
3
- allowed_args: "[PR number]"
3
+ allowed_args: "[PR number] [--force]"
4
4
  ---
5
5
 
6
6
  Run the high-level review (see `docs/high-level-review.md`) for the PR given in `$ARGUMENTS`, or for the current branch's PR when none is given:
@@ -9,6 +9,8 @@ Run the high-level review (see `docs/high-level-review.md`) for the PR given in
9
9
  assist review --high-level $ARGUMENTS
10
10
  ```
11
11
 
12
- Run it in the foreground with the maximum timeout and wait for it to finish — it opens the checklist in the web UI preview pane and returns once the user approves or requests changes. Do not review the PR yourself or post anything to GitHub.
12
+ Run it as a background task — it opens the checklist in the web UI preview pane and returns once the user approves or requests changes. Do not review the PR yourself or post anything to GitHub.
13
13
 
14
- When it returns, report the verdict and the path it saved the review to, quoting the command's own output. If the user left comments on any item, list them.
14
+ While it runs, the user may send diff comments from the pane: a file path and line range, the quoted code, and a note. Answer each one in the terminal by reading the code and the PR's diff — explain, or say whether the concern holds. Do not edit files or post to GitHub in reply.
15
+
16
+ When the task returns, report the verdict and the path it saved the review to, quoting the command's own output. If the user left comments on any item, list them.