Queue maintenance

A short, regular audit: what is waiting, what is broken, how much runway is left, and a clear list of actions, done only as far as the connection's autonomy and the person's yes allow.

Before you start

1. get_account: time zone, weekly cap, approval setting, autonomy.
2. list_channels: note needs_reconnect and queue_paused per channel.
3. list_posts once per status, with limit up to 100 (and channel_id when there are many channels):
   failed (has error)
   draft (waiting for the person's approval)
   approved (ready, not scheduled)
   scheduled (has scheduled_at, soonest first)
4. get_posting_schedule per active channel for slots and the next free slot.

Workflow

1. Failed posts first. For each, read error (or get_post). Codes are neutral for assistants: network_rejected (the network refused the content; the message says why), network_unavailable (temporary), channel_needs_reconnect, too_long, image_required, video_required, video_unsupported, weekly_limit. Fix by cause:
   Text problem: update_post, then check_post.
   Missing media: attach (propost-visual-posts, propost-video-posts).
   network_unavailable or timing: just retry.
   Reconnect: tell the person; skip that channel meanwhile. Then retry with add_to_queue or schedule_post (schedule level) once check_post is clean. A failed post can be rescheduled directly. Never publish_now as a retry unless they say yes after seeing the text.
2. Reconnect warnings. List channels with needs_reconnect, which posts are stuck on them (scheduled posts on such a channel will not go out), and ask the person to reconnect in ProPost. You cannot do it.
3. Draft triage. Group drafts by age and channel. For each: keep (ready to approve), improve (rewrite_post or update_post), or stale (old, about a past event or outdated fact). Present as a table with a proposed action. Do not decide staleness alone for anything the person wrote.
4. Approved but unscheduled. At schedule level, check_post then add_to_queue in order. On drafts-only, list them as waiting.
5. Runway check. From the scheduled list compute days covered per channel (the last scheduled_at minus now, from get_account time). Under 3 days is thin, under 1 is empty. Compare with the cadence in get_posting_schedule. If low, point to propost-weekly-content-plan and to unused ideas (list_ideas, see propost-idea-bank).
6. Reschedule. To move a post: unschedule_post (allowed at every level; it goes back to approved), then schedule_post with the new ISO time and offset, or add_to_queue. Confirm the new time in the person's time zone. If the queue is paused, say so (queue_paused).
7. Delete only with consent. Show the exact list (id, channel, first line, age) and ask "delete these N?". Then delete_post each. Deleting cannot be undone. Published posts cannot be deleted.
8. Report: counts per status, fixed, retried, what needs the person (reconnect, approvals, missing media), the runway per channel, and the next step.

Writing guidance

• Lead with the one thing that needs attention now ("2 posts failed, 1 channel needs reconnecting"), then details.
• Use the person's own words for post names: the first line, not the id. Keep ids available for follow-up.
• Be specific about why a post failed, in plain language, and say what you changed.
• Example row: Fri 09:30 | LinkedIn | "We cut onboarding..." | failed: network_rejected (text too long after edit) | shortened to 2,400 chars, rescheduled Fri 09:30+01:00.

Guardrails

• Never delete, unschedule or overwrite text without the person seeing it and saying yes (unscheduling a clearly mistimed post you are asked to move is fine).
• On drafts-only connections, editing an approved or scheduled post sends it back to draft; warn the person it now needs approval.
• Do not "retry until it works"; one fix, one retry, then report.
• Do not exceed the weekly cap; leave extras as approved and say so.
• rewrite_post is limited to 3 per post and overwrites the text; use update_post for small edits.
• Never publish_now to clear a backlog without an explicit yes per post.

If something fails

• network_rejected: read the post error, fix the text or media, retry later.
• network_unavailable: wait and retry once.
• channel_needs_reconnect: ask them to reconnect; skip meanwhile.
• already_scheduled: unschedule first. not_scheduled: it is not scheduled, nothing to undo. already_published: leave it.
• nothing_to_change: edit equals current text; skip.
• time_in_past / time_too_far: choose 1 minute to 1 year ahead.
• queue_paused / no_slots: resume or add slots (propost-posting-schedule-optimizer) or use schedule_post.
• weekly_limit, autonomy_limit, rate_limited: see ../CONVENTIONS.md.