Post a top-level comment on a post (activity_urn = urn:li:activity:<id>, or a bare id). People-facing → requires confirm=True. Returns status + the new comment URN when resolvable. Pure API (SDUI createComment, verified browserless 200). No browser.

Delete a comment. comment_id = the numeric comment id; activity_urn = the post it's on (urn:li:activity:<id>). Destructive → requires confirm=True. SAFETY GUARD: by default only the OWNER's own comments can be deleted. Deleting someone else's comment (e.g. a reply on your post) is refused unless force=True — this prevents accidentally removing a real person's comment during testing. Pure API: browserless Voyager REST DELETE (verified 204). dry_run=True builds+returns the request without sending. No browser fallback.

Return the owner's connection + pending-invitation counts.

List the owner's messaging conversations (inbox).

Read ONE job posting as a flat projection (title, company, location, employment_status, remote_allowed, listed_at, applies, views, salary, reposted, description_text). Browserless. job_id accepts a numeric id, a urn:li:fsd_jobPosting:<id> URN or a full linkedin.com/jobs/view/… URL. Unusable input returns an error WITHOUT any call. The result is IDENTITY-CHECKED: if the response describes a different job than the requested one — or names two job ids that disagree — ok=False and NO url is returned. The url is always built from the id you asked for, never from an id found in the response. A response that identifies the job but yields no readable field is also ok=False ("could not read" is not "a job without details"). Unknown values are null, never false; `salary_present` tells "salary exists but is not readable" apart from "no salary".