diff --git a/doc/plans/2026-05-21-roughdraft-doc-collaboration-plan.md b/doc/plans/2026-05-21-roughdraft-doc-collaboration-plan.md
new file mode 100644
index 00000000000..4da8722115d
--- /dev/null
+++ b/doc/plans/2026-05-21-roughdraft-doc-collaboration-plan.md
@@ -0,0 +1,104 @@
+# PAP-9940 — Roughdraft Learnings + Paperclip Doc Collaboration Plan
+
+## Goal
+Define how Paperclip should evolve document tooling by borrowing the best parts of Roughdraft for collaborative document review, while staying aligned with V1 constraints and our company-scoped architecture.
+
+## What we learned from Roughdraft
+
+- Local-first markdown workflow over plain `.md` files, with inline review markers persisted in file content (`comment`, `highlight`, `addition`, `deletion`, `substitution`) and agent-friendly handoff via explicit review-complete events.
+- Explicit agent collaboration contract:
+  - local open flow that yields a stable URL
+  - `review.completed`/status handoff
+  - review index JSON for machine-readability
+- A “Done Reviewing” action that transitions document state and emits completion metadata.
+- Separation between UI review ergonomics (inline/highlighted rail + split view style) and file/source-of-truth markdown persistence.
+- MCP-like external tooling path for listing open reviews, watching events, appending replies, and resolving comments.
+- Rich setup/CLI flow (open/read/status/watch) that lets agents and humans synchronize around a single document.
+
+## What to borrow now (Priority)
+
+1. **Highest value**: inline annotation model and review-index schema.
+2. **High value**: explicit review lifecycle (open → comment/suggest → complete) and status aggregation.
+3. **High value**: event/polling interface so agents can wait for reviewer completion.
+4. **Deferred**: full local CLI path (`roughdraft` command) until we finish first-party local-server UX.
+
+## What likely does not fit Paperclip without change
+
+- File-path ownership model is local filesystem centric; Paperclip docs currently live as issue/company-scoped DB entities.
+- We should avoid forcing CriticMarkup into all documents; support it as a compatibility/review layer within existing markdown storage.
+- We should not duplicate markdown file server + database state; persist as one source of truth in Paperclip docs and use derived review indexes.
+
+## Proposed Implementation Phases (post-approval)
+
+### Phase 1 — Design freeze + schema alignment (No code)
+
+- Finalize review model and decide where CriticMarkup-like marks are stored:
+  - raw markdown only, parsed annotations table, or hybrid parser + normalized `document_review_events` records.
+- Define interoperability contracts:
+  - Document review index JSON shape
+  - completion/status semantics (`open`, `in_review`, `needs_reply`, `done`)
+  - comment resolution rules and permissions
+- Confirm UX scope with product/design:
+  - inline threaded comments
+  - side rail vs modal review surface
+  - editor split/preview mode expectations
+- Define event API contract for agents:
+  - open request, update status, completion notifications
+- Add explicit UX/security acceptance criteria and non-functional requirements.
+
+### Phase 2 — API and server data model
+
+- Add server endpoints for review sessions and completion state transitions.
+- Add DB persistence for normalized annotations plus review sessions per document.
+- Add company/permission checks and immutable activity logs for all review mutations.
+- Add a compact, stable JSON review index endpoint consumed by agents and UI.
+- Add deterministic status transitions (e.g., reviewer complete, unresolved count, last update time).
+
+### Phase 3 — Rich editor and review UX in Paperclip UI
+
+- Introduce markdown-based rich editing mode with inline annotations.
+- Show rendered/markdown toggle and inline comment anchors with thread list.
+- Add “Done Reviewing” action with explicit final-state transition and optional reviewer summary.
+- Support read/write split behavior for reviewers:
+  - read mode for context
+  - edit mode for suggested replacements and highlights
+- Add diff-safe rendering that preserves base markdown as much as possible.
+
+### Phase 4 — Agent collaboration pathway
+
+- Add/extend Paperclip agent skills so agents can:
+  - open a document review session
+  - fetch pending review index data
+  - append/update comments/replies
+  - mark resolved items and report status changes
+  - wait for review completion event/timeouts
+- Add a machine-visible handoff payload (document URL/id + requested action + deadline + status).
+- Add audit-safe handling for status transitions triggered by agents.
+
+### Phase 5 — QA and rollout
+
+- Add focused integration tests for:
+  - review lifecycle endpoint transitions
+  - company scoping/security on review resources
+  - inline annotation parse/serialize roundtrip
+- Add UI interaction tests for at least:
+  - inline annotation rendering
+  - done-review transition
+  - status visibility and error handling
+- Ship with compatibility note for existing document types and migration path.
+
+## Delivery Plan (blocking dependencies)
+
+1. Confirm plan with CTO/board review.
+2. Approve schema contracts in shared types.
+3. Implement server contracts.
+4. Implement UI editor and review rail.
+5. Implement agent skill tooling and protocol.
+6. Validate with targeted QA and close by issue.
+
+## Evidence this work should satisfy issue request
+
+- Comments: full inline review flow.
+- Rich editor: markdown-aware editing and rendering with review rails.
+- Agent collaboration: explicit open/completion protocol and watch/update actions.
+- Borrowing rationale: mapped directly to Paperclip capabilities and constraints (company boundaries, contracts, existing document system).