annotate
/annotate
Read a cueloop plan or diff and comment on it as a review-side agent, without leaving the review. Use when you are helping a collaborator poke holes in a plan under review - you annotate on their behalf; you never rewrite the plan or cast the message.
- Author
- mmurakaru ↗
- Repository
- cueloop ↗
- Licence
- Apache-2.0
- Last repository change
- 25.10.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 25.10.2026, without references to retired models or APIs.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
What it does and when to use it
Read a cueloop plan or diff and comment on it as a review-side agent, without leaving the review. Use when you are helping a collaborator poke holes in a plan under review - you annotate on their behalf; you never rewrite the plan or cast the message.
Original author description
Read a cueloop plan or diff and comment on it as a review-side agent, without leaving the review. Use when you are helping a collaborator poke holes in a plan under review - you annotate on their behalf; you never rewrite the plan or cast the message.
How it works
- Rights boundary
- Steps
SKILL.md preview
---
name: annotate
description: Read a cueloop plan or diff and comment on it as a review-side agent, without leaving the review. Use when you are helping a collaborator poke holes in a plan under review - you annotate on their behalf; you never rewrite the plan or cast the message.
---
# cueloop annotate
The agent side of the review write-surface. You read the artifact under review
and attach span-anchored comments through the same annotation API a human uses.
Your comments are attributed to the collaborator who launched you and show up in
their rail exactly as if they had typed them.
## Rights boundary
You may **read** and **comment** only. You may not rewrite the plan
(`submit-revision` is the driving agent's alone), send the Message (`resolve` is
the human's alone), or cut / edit / share (owner-only). The plan stays
single-writer: you annotate, the human curates, the message carries it back.
Pass `--role agent` on every command below. The daemon then caps this connection
to read + annotate and rejects any escalation - the boundary is enforced, not
just documented.
## Steps
1. Read the artifact and its existing annotations:
```bash
cueloop session get <id> --role agent
```
2. See the shared quick-action vocabulary for this session (the same presets the
human picks); `--session` reads the reviewed session's own repo:
```bash
cueloop actions list --session <id> --role agent
```
3. Comment on a span. **Anchor contract: `--quote` is the exact text from the
artifact you are commenting on** - copy it verbatim, not a paraphrase or a
line number. Add `--prefix`/`--suffix` (a few surrounding characters) when the
quote is not unique. Attribute every comment to the acting collaborator with
`--author` (and `--author-name` on the first comment so the rail shows a name):
```bash
cueloop session annotate <id> --role agent \
--author <collaborator-id> --author-name "<display name>" \
--quote "<exact span from the plan>" \
--body "<your comment>"
```
To use a quick action instead of a free-form body, reference it by index or
name from `actions list`:
```bash
cueloop session annotate <id> --role agent \
--author <collaborator-id> --quote "<exact span>" --action "Out of scope"
```
4. Type a comment with `--kind` for a typed annotation (`security`, `perf`, a
tool id); the default kind is `comment`.
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · mmurakaru/cueloop | 100 | Apache-2.0 |
| glebis/claude-skills | 100 | MIT |
| janburzinski/annotate | 100 | MIT |
| Wahajolla/annotator-skill | 100 | MIT |
| a8cteam51/claude-code-plugins | 100 | MIT |
| infinum/arkive | 90 | Apache-2.0 |
Database snapshot: 13.11.2026 · package version b79690a9e3f3 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 81diff
Submit working-tree changes to the cueloop changes panel. Use when the user asks to review your code changes or before committing substantial agent-authored changes. - 81refine
Analyze past cueloop Threads and propose recurring feedback writebacks through a plan Thread. Use when the user asks to learn from past reviews or run refine. - 100review
Use when the rocket router selects review, or the user asks to review, audit, assess or check a PR, branch, diff, files, plan or design without asking for changes. - 95annotate-project
Analyse a project's source code and generate a simply-annotate database file with LLM-created annotations categorised by tag: `defun` for functions and large syntactical blocks, `line` for smaller sub-blocks. File-wide narrative lives in a sibling `.org` overview file, not in an… - 81plannotator-herdr
Open a Plannotator session (annotate a Markdown file or URL, review code, annotate the last message) inside the Herdr app instead of the browser. Use when the user asks to open, review or annotate something "in herdr", "in plannotator herdr", or "in a herdr pane". - 100annotate-setup
Add AnnotateKit (visual feedback for iOS apps: drawings and comments on the running app become prompts for a coding agent) to an iOS project, including React Native and Expo apps, and connect the agent so sent annotations arrive automatically. Use when the user wants to…