Org Slack Later

Slack Later / Saved-Items Organizer

Turn the Slack Later queue (saved messages / bookmarks) into a categorized, mirrored, cleared-down list: read every saved item in full, sort into action items vs references vs links vs drop, copy the keepers into a durable knowledge base, then mark the saved items complete.

Related but distinct from org-slack-dm (DM threads + notification tuning). This skill is specifically the Later / saved queue.

Requirements

• A logged-in Slack session in an automatable browser (e.g. an agent-browser / Playwright session where the user has completed SSO). This is the reliable path.
• Enterprise Grid caveat: user/bot tokens (xoxc/xoxd, or a bookmarks:read app token) frequently return invalid_auth on Grid workspaces, and there is no public API for the personal "Later" queue — so browser automation is the practical path. Don't burn time on tokens first; confirm whether the workspace is Grid.
• A destination knowledge base for keepers — e.g. Notion via MCP (see ../notion-organizer/SKILL.md), Apple Notes, or a markdown vault.
• The user completes any SSO / MFA login themselves; the agent cannot.

Step 1 — Ask the User First

1. Scope — the "In progress" tab only, or also Archived / Completed?
2. Destination — where do keepers go? (Notion page/DB, Notes, markdown vault) and which sections/taxonomy?
3. Categories — default is: action-items / personal-todos, references & learnings, links (tools · posts · docs · videos), drop (ephemeral work chatter). Adjust?
4. Long-form docs — when a saved item links to a doc (Google Doc, internal wiki, ticket-tracker doc), mirror its CONTENT into its own destination sub-page, or just keep the link?
5. Clear-down — after an item is saved, mark it complete in Later (removes it from the queue; recoverable in the Completed tab)? Or leave everything in place?
6. Date range — everything, or only items newer than a cutoff?

Phase 1 — Capture (read-only)

Don't trust the count badge

The number next to the tab (e.g. "In progress 99") is often stale or a max/placeholder and does not equal the real number of saved items. Derive the true count from the list itself.

The virtualized-list trap (the core gotcha)

The Later list is a virtualized/windowed list: only ~10–20 items are in the DOM at once, and nodes are recycled as you scroll. Two failure modes to avoid:

• Synthetic events are ignored. Setting scrollTop or dispatching untrusted wheel/scroll events does not trigger Slack's lazy-load — the list stalls at the first window. Use trusted input: real CDP mouse-wheel, or focus the list and arrow-key through it (ArrowDown). Keyboard traversal reliably advances selection and forces the next page to load.
• Windowing recycles nodes. Because items unmount as you scroll, you must accumulate + de-duplicate across the whole scroll, not read one snapshot. De-dupe on a content signature (channel + author + first N chars), not the DOM node id (ids differ between captures).

Loop: capture rendered items → advance (trusted scroll/arrow) → capture again → merge unique → stop when the set stops growing AND you've reached the true bottom (the last item stays stable across several advances).

Per-item fields

For each saved item capture: channel/DM name, author, full message text, and every link (resolve link-only messages to their URL). Strip user-mention/profile links; keep real content links.

Tombstones

Expect "A message you saved was deleted" placeholders — count them separately and skip; they inflate the badge but carry no content.

Phase 2 — Classify → Mirror → Clear (with approval)

1. Classify every captured item into the agreed categories. Flag personal to-dos (items the user must act on) distinctly from reference/links (things to save) and drop (ephemeral chatter not worth keeping).
2. Present the plan — a table of item → category → destination. Get batch approval before any write. This is a dry-run gate.
3. Mirror keepers to the destination:
   • Short items / links → bullets under the right topical section (tools, reading, docs…), each annotated with source (channel/author) and the link.
   • Long-form docs linked from a saved item → create one destination sub-page per doc with the content mirrored in, grouped under a container page/toggle. Do not just paste the raw source URL if the goal is a durable copy — the source may be access-gated later.
   • Auth-gated content you can't fetch (e.g. a doc behind a scope you lack) → fall back to saving the link and note that content wasn't mirrored. Don't block the whole run on one item.
4. Mark complete — for each item actually saved, mark it complete in Later (this clears it from the queue; it's recoverable in the Completed tab). Never mark an item complete before its save is confirmed — order matters, or you lose it from the queue with nothing saved.
5. Verify — destination has the expected new entries; Later count dropped by the number completed; anything left in the queue is intentional (personal to-dos + drop items you chose to keep).

Expected blockers (surface them, don't force)

• SSO / MFA login — hand back to the user to complete in the browser.
• Access-gated linked content (403 / missing scope on a linked doc) — mirror as a link, not content.
• Moving org-private content to a personal destination (e.g. cloning a company-private repo into a personal one) — this can trip data-exfiltration / trust-boundary guards. Surface it and let the user run that step themselves rather than working around the guard.
• Enterprise Grid token auth failing — expected; use the browser session.

Reference implementation

This skill describes the contract; keep any executable helper (e.g. a capture-and-mirror script) in your own personal repo — ~/Documents/<your-handle>/code/personal-os/scripts/slack-later-to-<destination>.py or similar. Skills describe, they don't ship code. Design the helper to be idempotent on a content signature and one-way (marking complete is the only Slack-side mutation, and only after a confirmed save).

Surface-specific notes

• "Later" has three tabs: In progress, Archived, Completed. "Mark complete" moves an item to Completed (reversible) — it does not delete the underlying message.
• There is no folder/label structure inside Later — organization happens off-platform, at the destination.
• Saved items span channels and DMs; a bookmarked link-only message is common and its value is the URL.
• Bot / integration messages (deploy links, ticket-tracker notifications) are often saved — treat their linked docs like any other long-form doc.

See also

• ../org-life-organizer/_lib/taxonomy.md — shared category vocabulary
• ../org-life-organizer/_lib/patterns.md — dry-run → batch-approve → apply, journaling
• ../notion-organizer/SKILL.md — preferred destination for a structured saved-items library
• ../org-slack-dm/SKILL.md — sibling skill for DM threads + notification tuning
• ../slack-mcp-browser-auth/SKILL.md — browser-session auth for Slack when API tokens fail (Enterprise Grid)