Select text in any chat message and attach a Google-Docs-style comment to it. Comments are highlighted inline, persisted per chat, and sendable to the prompt as one numbered review request.
Chat_comments turns an Agent Zero conversation into something you can mark up. Reviewing a long agent answer? Highlight the sentence that's wrong, comment on it, and, when you're done, send every comment to the prompt in one go so the agent addresses them together.
- Anchored comments : select a phrase in any message (yours or the agent's) → right-click → Comment. The phrase gets an inline highlight that survives re-renders via occurrence-based re-anchoring.
- General comments : chat-level notes that aren't tied to any span, for "this whole exchange went wrong" moments.
- Manage from one place : click a highlight for its popover (Edit / Delete), or open the toolbar modal to see every note; the toolbar button carries a live count badge.
- Send to prompt : collapses all comments into a single numbered instruction — anchored items quote their text — and drops it into the composer. Review the transcript, then hand the whole critique back in one turn.
- Persistence : comments are stored on the chat itself (
usr/chats/<id>/chat.json), so they survive reloads and chat switches. - Honest refusals : empty comments are rejected, and the backing service refuses malformed requests — both are asserted BDD scenarios.
A chat-top-end webui extension plus one API handler. Every create/edit/delete saves through the plugin's own endpoint; bootstrap and chat switches load the same key back and re-anchor. No external network calls, no telemetry.
Nothing to configure — the comment button appears in the chat top bar. Every behaviour below has an asserted BDD scenario, and docs/BEHAVIOUR.md shows each as a screenshot from a passing run.
Anchored comments — select text, right-click, Comment. One message can carry several; the comment records which text it refers to, so the note still makes sense later. Click a highlight to reopen, edit, or delete it.
General comments & send-to-prompt — open the toolbar modal to add a chat-level note, or hit Send to prompt to bundle every comment into one numbered request for the agent. The selection menu also offers Copy text and a direct Send to prompt.
Plugin Hub — once listed in the Plugin Index: Settings → Plugins → Chat Comments → Install.
Manual (Zip) — build it yourself, or grab the zip from the latest release:
make package # → dist/chat_comments.zipThen Settings → Plugins → Install from file → pick the zip. Or copy the plugin into your instance (usr/plugins/chat_comments/) and reload plugins.
None — zero-config, no settings screen. One internal constant: MAX_COMMENT_LENGTH = 2000 (both client and server clip to it).
git clone --recurse-submodules https://github.com/agent-zero-plugins/agent-zero-plugin-chat-comments
cd agent-zero-plugin-chat-comments
make verify # BDD static gates
python -m pytest tests -q # unit + L1 shape suite (needs tests/_testkit submodule)
make e2e # full behaviour BDD run on a nested disposable A0 (podman)Each comment persisted on a chat:
| Field | Type | Notes |
|---|---|---|
id |
str | client-generated uuid |
message_id |
str | message the highlight anchors to ("" for General) |
quoted_text |
str | the selected text ("" for General) |
occurrence |
int | which match of quoted_text in the message (0-based) |
comment |
str | the note (≤2000 chars) |
created_at |
int | unix seconds |
| Layer | Where | What |
|---|---|---|
| L1 shape | tests/test_plugin_shape.py |
static validator, deps/A0-API audits, dead hooks |
| Static gates | make verify |
feature-purity, honesty, traceability |
| Behaviour BDD | tests/e2e/features/ + steps/ |
anchored/general comments, edit/delete, persistence, refusals, e2e APIs |
Specs live in docs/spec/ (BEH-1..14). CI runs the whole pyramid on every PR, including a seam-off red-proof — a scenario that passes without the plugin fails the build.
Apache-2.0 — see LICENSE. Comments never leave the machine running Agent Zero; see SECURITY.md.