This repository is the canonical Papers rebuild and the only active Papers product line from this point onward. It is a clean rebuild of Papers' narrow definition (see below) — not a cleaned-up copy of the legacy implementation.
If you are a coding agent on a narrow task, read AGENTS.md first, follow your task prompt, and read only the files it names. If you are doing broad orientation, read DOCS/PROMPTER/ and DOCS/PAPERS_REPO_DECISION_AUDIT_V1.txt. In all cases the creator’s live direction in the current session is the highest authority.
Papers has two separate local directories and two separate GitHub repositories. Treat them very differently.
| Directory | GitHub repo | Role | |
|---|---|---|---|
| Legacy (v0) | Papers are papers\REAL |
Futahua/papers-are-papers |
Archived parts shelf and historical record. Reference only. |
| Canonical (rebuild) | Papers are papers\REAL2 |
Futahua/papers-real2 |
Active product line. All new work happens here. |
- REAL = legacy Papers v0. A well-engineered but wrongly-shaped implementation: a safety-hardened cockpit for one specific agent runtime (Hermes), wrapped around a generic agent chat workbench, with Papers vocabulary painted onto a disabled menu. It is retained as a salvage source and a lesson. It is not the base for ongoing work.
- REAL2 = the canonical rebuild. Its authority is the rebuild documentation in
DOCS/, governed above all byDOCS/PROMPTER/PAPERS_FOUNDER_BRIEF.txt. Its current product state is mapped inDOCS/PAPERS_RECOVERY_MAP_V1.txt— read that document before any product expansion or new build prompt.
The split is deliberate and structural. Its purpose is to stop future work from drifting back into “a generic agent workbench with Papers vocabulary painted on top” — which is exactly what legacy v0 became. Keeping the rebuild in a separate repo means REAL2 cannot silently inherit v0’s assumptions, session model, provider-centric surfaces, or Hermes-as-identity coupling.
- Do not continue development inside the legacy repo (
Futahua/papers-are-papers/ REAL). - Do not merge REAL2 into the legacy repo, and do not merge legacy into REAL2.
- No wholesale migration from REAL. Do not copy the legacy codebase into REAL2, and do not scaffold REAL2 by refactoring the old repo in place.
- Salvage is deliberate, narrow, and documented. REAL may be mined for specific proven pieces (see the audit’s salvage map) — but only by explicit decision, slice by slice, never as a bulk import.
- Authority is the rebuild docs, not legacy source. When code, legacy implementation, or old plans disagree with the governing docs, the governing docs win.
- Future sessions and implementation work target REAL2 only.
- The first implementation scaffold in REAL2 must be designed from the REAL2 docs themselves, not copied from the legacy app shell and renamed.
- REAL2 must not collapse into an engine cockpit, provider console, or generic chat workbench with Papers language layered on top. If a proposed change moves the product in that direction, it is the wrong change.
Papers is a lightweight personal layer over Windows. Its universal governing responsibility:
Papers persistently identifies, organizes, exposes, enters, tracks, leaves, and switches Backpacks — one active Backpack at a time.
Backpacks may be literally anything the creator designates. One active Backpack at a time is the only universal runtime assumption. No universal assumptions may be made about a Backpack’s interface, contents, furniture (dashboard, Desk, notes, files, chat, history, artifacts), AI integration, lifecycle, or state model — those are each Backpack’s own concern.
Papers itself must not prescribe: one global Papers AI; permanent World-level memory; universal Backpack context; AI in every Backpack; one provider or runtime; or any particular Backpack furniture as mandatory.
Reuse-first is hard law. Before any new capability is built, the discipline is: desired creator behavior → existing products → reuse investigation → smallest missing Papers responsibility → implementation.
The current REAL2 application code in app/ is experimental proving substrate — it demonstrates one form a Backpack can take (things, notes, conversation, history, a Desk), not a prescription for what all future Backpacks must contain. See DOCS/PAPERS_RECOVERY_MAP_V1.txt for the forensic classification of what the current implementation is, what is reusable, what is quarantined assumption, and what must be resolved before expansion continues.
Papers must never collapse into an engine cockpit, provider console, or generic chat workbench with Papers language layered on top.
Founder statement (CANONICAL): DOCS/PROMPTER/PAPERS_FOUNDER_BRIEF.txt (PAPERS_FOUNDER_BRIEF_V4).
- The creator’s live direction in the current session — explicit instruction given now always outranks older documents, including AI-written planning docs.
DOCS/PROMPTER/PAPERS_FOUNDER_BRIEF.txt(PAPERS_FOUNDER_BRIEF_V4) — CANONICAL, clean statement of what Papers fundamentally is.DOCS/PAPERS_RECOVERY_MAP_V1.txt— forensic drift history, current implementation classification, stopping point, and restart gate. Authoritative on those specific operational-recovery subjects; does not redefine product identity. Read before any product expansion or new build prompt.- Later explicit correction documents that say they supersede the founder brief on a specific point.
- Operational rebuild documents —
DOCS/PROMPTER/PAPERS_AGENT_CONSTITUTION_V1.txtand other corrected operational docs. - Historical evidence and legacy repo history —
DOCS/PAPERS_REPO_DECISION_AUDIT_V1.txt,DOCS/PAPERS_NEXT_BUILD_ORDER_V1.txt, legacy repo, old plans — weak evidence, often contaminated or superseded.
Supporting behavioral docs:
DOCS/PROMPTER/PAPERS_PROMPTER_HANDOFF.txtDOCS/PROMPTER/PAPERS_PROXY_OPERATING_MANUAL.txtDOCS/PROMPTER/PAPERS_SESSION_CLOSEOUT_PROTOCOL.txt
These do not outrank the founder brief. They exist to keep future sessions, agents, and handoffs aligned with it.
Slice 1 and Slice 2A are implemented in app/. The current code is experimental proving substrate — it demonstrates one Backpack form that has been built and tested, not a prescription for what all future Backpacks must contain.
Product expansion is paused. DOCS/PAPERS_RECOVERY_MAP_V1.txt documents the stopping point, the implementation classification, and the conditions (Section I of that document) that must be satisfied before another build prompt is authorized.
No new slices should be started until the restart gate in DOCS/PAPERS_RECOVERY_MAP_V1.txt Section I is cleared with the creator.
Present in the repository now: founder brief (V4, canonical); recovery map; repo decision audit (historical); agent constitution; build order (historical substrate-sequencing record); ontology clarification; Desk document (implementation-specific); prompter/proxy/closeout docs; delta log; application code in app/.
Vocabulary: per DOCS/PAPERS_ONTOLOGY_CLARIFICATION_V1.txt, Papers, Backpack, and Tool are the only active product vocabulary. "World" and "room" are historical implementation labels in the current code and persisted data — not product ontology.
Current app state (see app/README.md and DOCS/PAPERS_DELTA_LOG.txt for the full record): one Backpack implementation with truthful file/folder references, durable AI-written notes with provenance (revisable in place under the creator's direction, guarded, with the replaced text kept in the revision trail), a scoped conversation, a navigable activity history, a landing surface showing the Backpack's own state, and a Desk (active work surface) — an optional experiment in this Backpack form only, not universal Backpack anatomy. The AI speaks through one Papers-native engine seam with interchangeable runtime backends. Unavailability is reported honestly, never faked. None of this is claimed as the definition of what a Backpack must be — see DOCS/PAPERS_RECOVERY_MAP_V1.txt Section F for classification.
REAL2/
├── README.md
├── AGENTS.md
├── .gitignore
├── app/ (the Papers application — see app/README.md)
└── DOCS/
├── README.md
├── PAPERS_RECOVERY_MAP_V1.txt (current product state and restart gate)
├── PAPERS_REPO_DECISION_AUDIT_V1.txt
├── PAPERS_NEXT_BUILD_ORDER_V1.txt
├── PAPERS_ONTOLOGY_CLARIFICATION_V1.txt
├── PAPERS_BACKPACK_DESK_V1.txt
├── PAPERS_DELTA_LOG.txt
└── PROMPTER/
├── PAPERS_FOUNDER_BRIEF.txt
├── PAPERS_PROMPTER_HANDOFF.txt
├── PAPERS_PROXY_OPERATING_MANUAL.txt
├── PAPERS_SESSION_CLOSEOUT_PROTOCOL.txt
└── PAPERS_AGENT_CONSTITUTION_V1.txt
- Governing founder / proxy / rebuild documentation
- The eventual REAL2 implementation, following
DOCS/PAPERS_NEXT_BUILD_ORDER_V1.txtonce implementation is authorized - Narrow, explicitly justified salvage from REAL, only when a specific REAL2 slice calls for it
- A copied or renamed legacy app shell
- Bulk imports from REAL
- A generic Tauri/React chat workbench scaffold created “just to get started”
- Provider-setup work, engine-management surfaces, or session plumbing treated as the product’s center of gravity
The legacy repo Futahua/papers-are-papers remains the archived Papers v0 line. It is retained for:
- historical record
- targeted salvage of proven pieces
- negative reference for contamination patterns to avoid in REAL2
It is not the active product line, and it is not the implementation base for the rebuild.