Repository navigation
Decision Log
KepCup 开发助理 edited this page Oct 24, 2026
·
1 revision
Source: docs/design/README.md (full D1–D102 table)
English summary of product-critical decisions. Numbers match the Chinese design README.
| ID | Decision |
|---|---|
| D1 | Single-user app; Bots are private to that user |
| D3 | Delete conversation → keep long-term memory; chat history gone |
| D18 | Delete Bot ≈ remove contact: messages remain (name→id), loops abort, id never reused; memory/wiki/skills/workspace cleared |
| D20 | Delete conversation voids promises in that chat; wiki materials kept |
| ID | Decision |
|---|---|
| D2 / D75 | New message → next reply turn; Bot chooses inject_task / cancel_task / start_task
|
| D5 / D92 | Project teams: no @ → previous speaker continue-check → lead gateway (self / silent route / none); not all-members broadcast |
| D4 / D92 | Bot↔Bot chaining → collaboration episode + circuit breakers (12 hops / 8 ping-pong), not token budget |
| D48 | Final reply auto-sends; mid-turn send_message; silence via skip_reply; @ only via structured params |
| D86 | Send is immediate (no pending-send queue); one message = one trigger batch |
| ID | Decision |
|---|---|
| D8 | Private chat facts default into shared user profile; sensitive / “only tell you” excluded |
| D9 / D92 | Bot wikis fully isolated; project wiki is the only shared knowledge channel |
| D7 / D100 | User-intent authored skills auto-enable after sandbox trial; reflection suggestions stay disabled until user enables |
| D97 | Skill suggestions need verifiable evidenceRefs (≥2 completed runs) |
| D101 | Lessons/self-notes inject date + up to 2 evidence message ids |
| ID | Decision |
|---|---|
| D11–D12 | Local client; macOS / Windows / Linux |
| D21–D23 | pi loop; Electron; Svelte 5 + shadcn-svelte |
| D24–D26 | Encrypted SQLite stack; data dir ~/.kepcup/
|
| D27 / D15 | Sandbox: srt (macOS/Linux), private WSL2 (Windows); enhanced tiers Lima/Podman |
| D28–D30 | Workdir bound to conversation; write lease; shadow-git checkpoints |
| D35–D41 | Soft path gateway + sandbox; revoke interrupts; per-command confirm fallback; unattended mode with audit; fail-closed for unsandboxed commands that touch app data |
| D49 / D67 / D78 | Ephemeral runs: mark interrupted, no auto-resume; durable journal not implemented; tool_effects ledger is implemented |
| D99 | Unattended never auto-approves profile changes |
| ID | Decision |
|---|---|
| D58 / D89 | Missing config → inline setup / conversational settings cards |
| D65 / D73 / D74 | MCP + Connected Apps (OAuth); open platform = standards, not invented protocols — Connected Apps largely not implemented |
| D70 / D71 | Butler + A→B delegation (not SubAgent); Butler never joins project teams |
| D72 | External ACP agents as task engines; reply turns stay built-in — not implemented |
| D75 / D91 / D93 | Supervisor+tasks implemented; capability mounting implemented; task harness P0–P4 implemented |
| D83 / D34 | Optional KepCup supplier overlay — designed, not implemented |
| D84 | Mail channel P0 implemented (device acceptance pending) |
| D92 | Project teams implemented (2026-10-18) |
| D94 | macOS Share Extension P0 implemented |
| D96 | Remote control — designed, not implemented |
| D102 | Headless server — designed (auth revised to password for direct mode) |
For the complete table and wording, see docs/design/README.md in the repository.
KepCup
Core
- 01-Conversation
- 02-Execution
- 03-Bot
- 04-Memory
- 05-Wiki-and-Skills
- 30-Supervisor-and-Tasks
- 40-Capability-Mounting
- 43-Task-Loop-Harness
Runtime
- 06-Isolation-and-Storage
- 07-Local-Runtime
- 08-Workdir
- 09-Tech-Stack
- 10-Sandbox
- 11-Storage
- 13-Permissions
- 24-Durable-Execution
Product
- 12-UI-Layout
- 33-Narrow-Layout
- 39-Workbench
- 41-Project-Team
- 42-Task-Run-View
- 45-Bot-File-Cards
- 47-File-WebApp-Viewer
Capabilities
- 14-Models-and-Browser
- 15-Interactive-Execution
- 16-Capability-Models
- 17-MinerU-Document-Parsing
- 18-Inline-Setup
- 19-Work-Path-and-Group-Setup
- 20-Conversation-Media
- 21-Web-Search
- 22-File-Skill-Routing
- 23-MCP-and-SubAgent
- 25-Capability-Tools
- 26-Voice-Input
- 31-Sensors
- 32-Bot-Widgets
- 35-Mail-Channel
- 36-Wiki-Multimodal-Ingest
- 38-Conversational-Settings
Team & agents
- 27-Butler-and-Delegation
- 28-External-Agents-ACP
- 37-Bot-Asset-Handover
- 41-InkCop-Local-Collab
- 48-Bot-Template-Export
Cloud & remote