The command line

The plugin registers three subcommands with Obsidian’s own CLI, so a review can be driven from a terminal, a script, or an agent.

obsidian editor-ai-daemons:review --file "Some Note" --format text
obsidian editor-ai-daemons:status --file "Some Note" --format text
obsidian editor-ai-daemons:cancel --file "Some Note"

Requires Obsidian 1.12.2 or newer, on desktop — the same floor the plugin itself requires, since 1.12.2 is the release that added the API.

Not to be confused with CLI backends, which are the other direction: an AI agent running on your machine as a backend.

editor-ai-daemons:review

Review a note with the configured AI editors.

Flag Value Notes
--file <path> Required. Vault path, with or without .md, or link text
--editors <ids-or-names> Comma-separated; overrides a matching rule. Default: the note’s own pool
--format json | text Default json
--confirm-large flag Confirm reviewing a note above the size warning threshold

It runs through the exact same pipeline as the Review current note command, so it refuses for the same reasons and honours the same rules, exclusions and size guard. --editors is a choice you made, so it wins over a rule that assigns a reviewer to the note — rules only supply the default. An unknown or disabled name fails the whole call rather than silently reviewing with fewer editors. It waits for the run to settle — including the panel scorecard when the run is a panel run — and then prints one document.

If the note is open in a markdown view, the snapshot comes from the live editor buffer (unsaved edits included), the run is bound to that view, and the findings show up on the rail, in the highlights and in the panel just like a review you started by hand. If the note is not open, the saved file is used and the run is discarded once the output is shaped.

JSON output 

{
    "ok": true,
    "file": "Some Note.md",
    "findings": [
        {
            "id": "…",
            "editor": "Concision Editor",
            "severity": "suggestion",
            "quote": "in order to",
            "critique": "…",
            "edits": [
                {
                    "op": "replace",
                    "text": "to",
                    "anchor": { "from": 412, "to": 423, "state": "anchored" }
                }
            ],
            "anchor": { "from": 412, "to": 423, "state": "anchored" }
        }
    ],
    "skips": [{ "editor": "Fact Checker", "reason": "no model configured" }],
    "summaryByEditor": { "Concision Editor": "…" },
    "panel": null,
    "error": null
}

anchor is null when the quote could not be located in the note. panel carries the scorecard — verdict, rationale, per-member verdicts, ranked top fixes with the quote they point at, dissent, and the members that produced nothing — on a panel run, and is null otherwise.

Text output

One line per finding:

[suggestion] Concision Editor 412-423: "in order to" — Three words doing one word's work -> [replace] to
Skipped Fact Checker: no model configured

An unanchored finding prints unanchored where the offsets would be; a stale one prints 412-423 (stale). No findings. when there are none. A panel run appends its verdict, one line per member, the ranked fixes and each disagreement.

Error codes

Errors are typed so a script can branch on them, and the messages are status-only — nothing model- or backend-derived is echoed.

Code Meaning
bad-args A flag is missing or unparseable — distinct from the note not existing
file-not-found No markdown note at that path
excluded The note is privacy-excluded
rule-disabled A binding rule switched the plugin off for it; names the rule
needs-confirmation Above the size threshold; re-run with --confirm-large
no-editors Nothing could run; skips says why for each
panel-unavailable The assigned panel cannot run
backend-error The run failed
timeout The run timed out

editor-ai-daemons:status

Show the state of the AI review run for a note. Read-only: it starts nothing and spends nothing.

Flag Value Notes
--file <path> Required
--format json | text Default json

Text output is a headline plus the same finding lines as review:

Run in progress (2 running, 1 done) — 4 findings so far

or Run settled (3 done) — 7 findings, or No run for Some Note.md. when nothing is tracked. The JSON document carries run.settled, one entry per editor with its status and (redacted) error, the findings in the same shape as review, and the per-editor summaries.

editor-ai-daemons:cancel

Cancel the in-flight AI review of a note.

Flag Value Notes
--file <path> Required

JSON only. {"ok":true,"file":"…","cancelled":true} when something was cancelled, or cancelled: false with a reason of no-run or already-settled. Both of those are successes: asking to cancel nothing is not an error.

Scripting notes

  • Every subcommand always answers with parseable output, including on failure. Nothing throws through the CLI surface.
  • --format text is for humans and greps; --format json is for programs. The finding line format is identical between review and status.
  • A CLI review of an open note behaves as a real review of that note: cancel it with editor-ai-daemons:cancel, or from the rail.
  • Batch use: prefer notes that are not open. Runs on open notes stay in memory so their highlights keep working.

Next