@staff0rd/assist 0.623.1 → 0.624.0

package/README.md

@@ -288,6 +288,7 @@ The Config tab of the sessions web dashboard never receives secret values: `GET
 - `assist jira view <issue-key>` - Print the title and description of a Jira issue
   - Note: Claude fetches Jira context via the MCP Atlassian server, so `/jira` and Jira-key mentions go through MCP. These CLI commands remain for direct human use.
 - `assist miro extract [name] [--items <file>] [--top-left <id|link> --bottom-right <id|link>] [--ignore <file>] [--out <file>] [--board <id>] [--frame <id>] [--save <name>] [-g] [-r [repo]]` - Print the text of every box inside a rectangle on a Miro board as a YAML list, leftmost box edge first (topmost edge breaks ties). `--items` is a file of raw `board_list_items` response pages; anchors accept a bare widget id or a `?moveToWidget=<id>` link. Omit both anchors to pick them by clicking the top-left then the bottom-right box in the assist web UI preview pane, which echoes the pair back as flags for later runs. `--ignore <file>` is a YAML list of box texts to drop, warning about entries that matched nothing; repeated identical text is listed once at its highest-priority position. `--out <file>` writes the YAML to a file with a header recording the board, frame, anchors and computed rectangle instead of printing to stdout; `--board`/`--frame` override the ids read from the items for that header. After a pick the selection can be saved as a named extract under `miro.extracts` — `--save <name>` saves without asking, and `-g` / `-r [repo]` choose the config file as they do for `assist config set` (the project `assist.yml` by default). `assist miro extract <name>` then replays that extract with no flags, resolving its paths from the repo root and reporting the config file it came from; any flag overrides the matching field
+- `assist litellm list-models [--json]` - List the model ids the configured LiteLLM proxy serves, sorted one per line, from `GET <litellm.baseUrl>/v1/models`; `--json` prints the raw response body. Requires `litellm.baseUrl` and `litellm.apiKey`
 - `assist ravendb auth add` - Add a new RavenDB connection
 - `assist ravendb auth list` - List configured RavenDB connections
 - `assist ravendb auth remove <name>` - Remove a configured connection
@@ -348,7 +349,7 @@ The Config tab of the sessions web dashboard never receives secret values: `GET
 - `assist transcript configure` - Configure transcript directories
 - `assist transcript clean <path>` - Clean any .vtt file and write the result to stdout (`--format <md|vtt>`, default `md`). Markdown chat log: `assist transcript clean ./raw.vtt > clean.md`; cleaned WebVTT with timings preserved: `assist transcript clean ./raw.vtt --format vtt > fixed.vtt`. `--timestamps` prefixes each markdown speaker turn with `[hh:mm:ss]`, so passages can be cited as ranges for `merge --select`: `assist transcript clean ./raw.vtt --format md --timestamps`
 - `assist transcript list` - List raw .vtt filenames waiting in the pick-up directory
-- `assist transcript merge <path...>` - Collapse several .vtt files into one transcript with `NOTE` provenance, rebasing cue times onto a continuous timeline (`--out <path>` to write a file instead of stdout): `assist transcript merge ./a.vtt ./b.vtt --out ./refinement.vtt`. `--select <file|->` takes keep/removed JSON (`-` reads it from stdin) naming the ranges to keep, so only those passages survive and the dropped ones are counted in the header: `assist transcript merge ./a.vtt ./b.vtt --select ./selection.json`. `--no-provenance` omits every `NOTE` — the Collapsed-from header, the per-passage source marks and the removed count — for an output going somewhere the source names and cut points should not follow: `assist transcript merge ./a.vtt ./b.vtt --select ./selection.json --no-provenance`. `--strip-profanity` deletes the profanity that carries no meaning — an intensifier before a word, `the fuck`/`the hell` after a wh-word, a standalone expletive interjection — and drops any cue that is nothing but an expletive, leaving load-bearing uses (verb, idiom, decision marker, predicate adjective, noun, reported speech) alone: `assist transcript merge ./a.vtt ./b.vtt --no-provenance --strip-profanity`
+- `assist transcript merge <path...>` - Collapse several .vtt files into one transcript with `NOTE` provenance, rebasing cue times onto a continuous timeline (`--out <path>` to write a file instead of stdout): `assist transcript merge ./a.vtt ./b.vtt --out ./refinement.vtt`. `--select <file|->` takes keep/removed JSON (`-` reads it from stdin) naming the ranges to keep, so only those passages survive and the dropped ones are counted in the header: `assist transcript merge ./a.vtt ./b.vtt --select ./selection.json`. `--no-provenance` omits every `NOTE` — the Collapsed-from header, the per-passage source marks and the removed count — for an output going somewhere the source names and cut points should not follow: `assist transcript merge ./a.vtt ./b.vtt --select ./selection.json --no-provenance`. `--widen-audience` retunes the language for a reader who was not in the call: the casual asides pitched at the people who were — an intensifier before a word, emphasis after a wh-word, a standalone interjection — are deleted, along with any cue that is nothing but one, leaving anything that carries meaning (verb, idiom, decision marker, predicate adjective, noun, reported speech) for you to judge: `assist transcript merge ./a.vtt ./b.vtt --no-provenance --widen-audience`
 - `assist transcript move <file>` - Convert a raw .vtt to a dated markdown transcript and archive the original
 - `assist voice setup` - Download required voice models (VAD, STT)
 - `assist voice start [--foreground]` - Start the voice daemon (always-on, listens for wake word)