"All" Is Not All: Systemic Semantic Inconsistency of Totality Claims¶
Evidence Level: F (directly provable from source)
Analysis Baseline:4f843556
TL;DR¶
Multiple features in Clawith claim to show "all" of something, but none actually include everything. This is not an isolated bug — it is a systemic pattern: the word "all" has never been aligned between UI and backend across the system.
1. Example 1: Activity Feed's "All Activities"¶
The Dashboard's "Recent Activity" panel shows agent activity from the AgentActivityLog table.
backend/app/models/activity_log.py — the action_type enum:
action_type: Enum(
"chat_reply", "tool_call", "feishu_msg_sent", "agent_msg_sent",
"web_msg_sent", "task_created", "task_updated", "file_written",
"error", "schedule_run", "heartbeat", "plaza_post",
)
Missing: a2a_delegated, consult, notify, task_delegated.
Any work done through A2A delegation — Agent A delegates a task to Agent B, Agent B completes it and returns the result — does not appear in the activity log. The Dashboard's "all activities" actually excludes all inter-agent collaboration.
2. Example 2: Directory's "All Agents"¶
The query_directory tool returns "the list of agents in the current tenant," but _agent_directory_conditions filters by access_mode:
- A
privatesource agent can only see same-creatorprivateagents - A
companysource agent cannot seeprivateagents
"All" actually means "the subset visible to your access_mode."
3. Example 3: custom Mode's "Custom Visibility"¶
The custom mode promises users can customize "who can see my agent." In reality:
AgentAgentRelationshipis a viewer-side configuration (agent_id= who is looking), not a viewed-entity policy- Any admin can add you from any non-private agent's perspective
"Custom" actually means "anyone can customize seeing you."
4. These Three Examples Are Not Isolated¶
| Feature | Claims | Actual |
|---|---|---|
| Activity Feed | All activities | Missing A2A delegation |
| Directory | All agents | Missing cross-access_mode agents |
custom mode |
Custom visibility | Viewer-side control, viewed has no say |
Three examples from three different modules (Activity Log, Directory, Permissions), all exhibiting the same pattern:
UI/product level: promises "all"
↓
Code level: implements "the subset visible under current local conditions"
↓
Result: the semantics of "all" are never globally defined or enforced
5. Root Cause: No Global Definition of "All"¶
In a correct system, "all" must be explicitly defined:
All activities = all action_type values
All agents = all agents in the tenant (visibility filtering is a separate layer)
All visibility = visibility policy controlled by the viewed entity
But Clawith has no such global definition. Each module interprets "all" locally, the frontend displays its own understanding, and the backend filters with its own. The word "all" is scattered across local implementations and never aligned.
6. Relationship to AI Coding¶
- Locally coherent, globally drifting semantics: Each module's "all" has its own definition in its local context (Activity Log's "all" = the types in the enum; Directory's "all" = agents visible to the current access_mode). Each module is "correct" in isolation, but "all" loses consistency across modules
- No global concept registry: When AI Coding generates code task by task, no one asks "wait, what does 'all' actually mean in this system?" — each task defines "all" within its own scope
- UI text and backend logic are never cross-validated: The frontend writes "all activities," the backend writes
select * from agent_activity_logs— both are locally correct, but no one checks "does the backend actually return everything?"