| name | research-add-items |
| user-invocable | true |
| allowed-tools | Read, Write, Glob, Task, AskUserQuestion |
| description | Append new items (research objects) to an in-progress research outline's `outline.yaml` — sourced from your direct input, a web-search agent, or both. Use mid-`/research-outline` when you've realised the items list is incomplete (a new competitor surfaced, an important historical entry was missed, a category needs broader coverage) before running `/research-deep`, so the new items are part of the parallel deep pass instead of needing a separate run. |
Research Add Items — Append to the Items List
In-place updates outline.yaml with additional research objects.
Trigger
/research-add-items
Pipeline position
/research-outline → ► /research-add-items ◄ → /research-deep → /research-report
Reachable any time after /research-outline has produced outline.yaml, but most useful before /research-deep so the new items get researched in the same parallel pass.
Workflow
Step 1 — Auto-locate outline
Glob */outline.yaml from the current working directory. Read it to know the existing items (avoid duplicates) and the current execution config.
Step 2 — Gather candidates from two sources
These two sources are gathered sequentially (AskUserQuestion blocks waiting on the user; web search can run in the background after). Do them in this order:
AskUserQuestion: "Which items do you want to add? Any specific names?" — capture whatever the user already has in mind.
AskUserQuestion: "Should I also run a web-search agent for more candidates?" — if yes, launch a research subagent via the Task tool (subagent_type: general-purpose, run_in_background: true) seeded with the topic, the existing items list, and any items the user named in step 1 (so suggestions don't duplicate).
Step 3 — Merge, deduplicate, confirm
- Combine the user-supplied items and the subagent's suggestions.
- Deduplicate by case-insensitive name match against the existing
outline.yaml.
- Show the merged candidate list to the user.
AskUserQuestion for which to accept.
- For each accepted item, capture:
name (required), category (optional), description (optional, short).
Step 4 — Save update
Append accepted items to outline.yaml, preserving existing structure. Save in place.
Item shape (matches /research-outline)
items:
- name: <item>
category: <optional>
description: <optional, short>
Output
Updated {topic}/outline.yaml — in-place modification, user confirms before save.
Gotchas
- Duplicate detection is case-insensitive name match only.
GPT-4 and GPT 4 (different by one character) are not detected as duplicates. Eyeball the merged list before saving if you suspect near-duplicates.
- Items added after
/research-deep has already run are not auto-researched. You either need to re-run /research-deep (its Step 2 resume check will skip already-completed items and only fan out the new ones), or research the new items manually.
- The web-search subagent operates outside the conversation context. Hand it the topic AND the current items list explicitly, so its proposals don't repeat what's already there.
outline.yaml.execution.batch_size controls how many deep agents /research-deep runs in parallel. If you add many items at once, you may want to bump batch_size here too — but that's outside this skill's scope; edit outline.yaml directly or re-run /research-outline's Step 4 prompts.