📥 Feedback
Actionable inbox items for Clara to process. The "proactive sticky notes" of the portal.
Why Feedback Exists
Feedback was created as part of the Entity Architecture refactoring (A82, April 2026). The old
comments table was overloaded — serving as both permanent history AND an actionable inbox. This
caused confusion:
The original problem: "Previously I want to leave a comment, but actually what I really mean
is I want to leave a feedback, that will be collected and processed. Then the comment (actually feedback) will
be gone when we run Clara." — Adrian
The solution: split into Comment (permanent, backward-looking) and Feedback (actionable, forward-looking).
Schema
D1 feedback table
Proactive inbox items. Clara's sweep pulls from here.
idTEXT PK — UUID
contentTEXT NOT NULL — The
feedback message
workspaceTEXT — Optional workspace
routing
priorityTEXT — urgent | high |
normal | low (default: normal)
statusTEXT — open | processed |
parked | archived (default: open)
targetTEXT — Optional target
entity
linked_typeTEXT — task | report |
job (optional)
linked_idTEXT — ID of linked
entity (optional)
sourceTEXT — portal | clara |
claudebot | agent (default: portal)
processed_atTEXT — Timestamp set
once when processed (immutable)
processed_noteTEXT — Clara's
one-liner ("Created A67" or "Not actionable")
created_atTEXT — ISO
timestamp
updated_atTEXT — ISO
timestamp
Lifecycle
open
→
processed
open
→
parked
→
open
(promote)
any
→
archived
Ownership Rules
| Transition | Who | Notes |
|---|---|---|
open → processed |
Clara only | She's the one doing the processing. Sets processed_at + processed_note. |
* → parked |
Adrian only | Conscious decision: "I've read this, I'm choosing not to act now." |
* → archived |
Adrian only | "I don't care about this anymore." Not the same as processing. |
parked → open |
Adrian only | Promotes parked feedback back to Clara's queue. |
"Processed" is immutable: Once Clara marks feedback as
processed, it cannot be
reopened. The outcome lives elsewhere — if Clara created a task from it, the task carries it forward. If she
decided it wasn't actionable, that's captured in processed_note.
Agent-Posted Feedback
When agents (not humans) post feedback, it arrives as parked — not open. This
prevents agents from presumptively filling Clara's queue.
Inversion: Agents say "here's a thought" (parked). Adrian decides whether it's worth Clara's
time by promoting to
open. The source field disambiguates the origin.
API
| Method | Endpoint | Description |
|---|---|---|
GET |
/api/feedback |
List. Filter: ?status=open, ?workspace=agentic |
POST |
/api/feedback |
Create. Body: { content, workspace?, priority?, source? } |
PATCH |
/api/feedback?id=xxx |
Update status/content. Body: { status: "processed", processed_note: "Created A87" } |
DELETE |
/api/feedback?id=xxx |
Hard delete (prefer PATCH to archive) |
Where It Appears
| Location | Context |
|---|---|
| Inbox page | Global feedback view — all open feedback across workspaces |
| Task modal → Feedback tab | Feedback scoped to a specific task (target: task:workspace:id) |
| Report viewer → Annotations drawer | Feedback on reports — "This report needs updating" or "Finding is outdated" |
Naming Decisions
| Choice | Alternatives Rejected | Why |
|---|---|---|
| Feedback | Notes, Inbox Items | Feedback implies someone should act. "Notes" feels passive. |
| Processed | Considered, Evaluated | Unambiguous. "This item has been through the pipeline." |
| Open | Ready | "Ready" collides with task status. "Open" is universal. |
| Parked | Draft | "Draft" implies incomplete writing. "Parked" = consciously deferred. |