Skip to content
Agentmart
Submit a tool

cue-omni-reader

Use when the user wants an external AI agent to parse or understand an HTTP(S) URL or an authorized local document, audio, or video source through Cue Omni…

Skill by sensedeal

IndexedMIT★ 6
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/sensedeal-cue-omni-reader

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use when the user wants an external AI agent to parse or understand an HTTP(S) URL or an authorized local document, audio, or video source through Cue Omni Reader.

Use cases: documents, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Python package

Network hosts it talks to 2

  • codeload.github.com · found in install command
  • github.com · found in code

Files it touches

  • ~/.claude/skills/cue-omni-reader
  • /home
  • /Users

What's inside

Skill name
cue-omni-reader
Description
Use when the user wants an external AI agent to parse or understand an HTTP(S) URL or an authorized local document, audio, or video source through Cue Omni Reader.
License
MIT
Metadata
version: 0.5.4
Instructions
73 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (40)

  • README.md
  • README.zh-CN.md
  • SKILL.md
  • SKILL.zh-CN.md
  • 2026-08-08-baseline-pressure.md · docs/verification-reports/
  • 2026-08-08-bridge-cli-audit.md · docs/verification-reports/
  • 2026-08-08-claude-code.md · docs/verification-reports/
  • 2026-08-08-codex-cli.md · docs/verification-reports/
  • 2026-08-08-gemini-cli.md · docs/verification-reports/
  • 2026-08-08-hermes.md · docs/verification-reports/
  • 2026-08-08-skill-pressure.md · docs/verification-reports/
  • 2026-08-08-workbuddy.md · docs/verification-reports/
  • 2026-08-11-content-only-compat.md · docs/verification-reports/
  • 2026-08-13-bridge-1.1.3-published.md · docs/verification-reports/
  • 2026-08-14-bridge-1.2.0-published.md · docs/verification-reports/
  • 2026-08-14-bridge-1.2.1-published.md · docs/verification-reports/
  • 2026-08-15-bridge-1.2.2-published.md · docs/verification-reports/
  • 2026-08-17-bridge-1.3.0-published.md · docs/verification-reports/
  • 2026-08-18-bridge-1.3.1-published.md · docs/verification-reports/
  • 2026-08-18-bridge-1.3.2-published.md · docs/verification-reports/
  • 2026-08-18-bridge-1.3.3-published.md · docs/verification-reports/
  • 2026-08-19-bridge-1.4.1-published.md · docs/verification-reports/
  • 2026-08-26-bridge-1.5.5-published.md · docs/verification-reports/
  • 2026-08-30-bridge-1.6.0-published.md · docs/verification-reports/
  • 2026-09-04-bridge-1.7.1-published.md · docs/verification-reports/
  • 2026-09-06-bridge-1.7.2-published.md · docs/verification-reports/
  • 2026-09-08-bridge-1.7.3-published.md · docs/verification-reports/
  • 2026-09-09-bridge-1.8.0-published.md · docs/verification-reports/
  • 2026-09-22-bridge-1.8.1-published.md · docs/verification-reports/
  • 2026-09-23-bridge-1.8.2-published.md · docs/verification-reports/
  • 2026-09-23-bridge-1.8.3-published.md · docs/verification-reports/
  • 2026-09-28-bridge-1.8.4-published.md · docs/verification-reports/
  • 2026-10-06-bridge-1.8.5-published.md · docs/verification-reports/
  • 2026-10-07-bridge-1.8.6-published.md · docs/verification-reports/
  • 2026-10-17-bridge-1.8.7-published.md · docs/verification-reports/
  • README.md · docs/verification-reports/
  • compatibility.md · references/
  • setup.md · references/
  • sync_bridge_pin.py · scripts/
  • test_skill_regression.py · scripts/

SKILL.md (first lines)

# Cue Omni Reader

> 中文版见 [`SKILL.zh-CN.md`](SKILL.zh-CN.md) — complete Chinese translation.

Use official Omni MCP tools to parse the source, then complete the user's task. The MCP package and active tool schemas are authoritative; this skill has no parser or protocol driver.

## Core flow

1. **Preserve the source.** Only HTTP(S) strings are URLs. Pass the user's source string directly to `parse`. Do not pre-read or base64-encode local content; never use `file://`, localhost, or a public temporary upload service.
2. **One provider, one first call.** Use `parse` as the only first call for both HTTP(S) URLs and local paths; do not ask the user to choose a local, remote, upload, or URL mode. Bridge is the same provider; remote-only cannot read local files.
3. **Bootstrap only with consent.** Before installing the Bridge or expanding an allowed root, obtain user confirmation and add only the minimum required directory. Follow [`references/setup.md`](references/setup.md)'s exact pin, run `doctor`, then reload/reconnect the MCP server. Run `doctor --json --silent-check` on first use; if `version_check.status=outdated`, tell the user installed→latest and upgrade only on confirmation. Never ask for an API key in chat. If already inside an allowed root, do not ask for another confirmation.
4. **Call the active schema.** Obey the active `parse` schema. If the schema exposes `wait`, use `wait: false` for long media or a large document; the source-only Bridge returns a recoverable operation. Send one of `source`/`url`. `grounded`/`layout` work for files and URLs where capabilities advertise them; `UNSUPPORTED_DETAIL`/`DETAIL_CAPABILITIES_UNAVAILABLE` is final. Do not race synchronous and asynchronous submissions.
5. **Read either response channel.** Prefer `structuredContent` when available; if only `content[].text` exists, parse compact JSON. Completed inline content may be exact Markdown; non-inline states are compact JSON. A generic success is not a completed result. Append only `result.text`; never append the JSON wrapper.
6. **Preserve one operation.** On `processing`, save the `operation_id` and poll `get_parse_status` at the returned timing/`wait_ms`. Recover the existing operation before resubmitting. A lost ID is an ambiguous timeout; explain duplicate-work/billing risk and get confirmation.
7. **Choose continuation and delivery automatically.** Choose continuation tools from the structured result; never present the tool list as a menu for the user. Use this decision table:

   ```text
   Answer directly → inline when present; otherwise `read_result`
   Find one section → `read_outline` → `read_result(cursor)`
   Read all content → `read_result` until no `next_cursor`
   Deliver a file → `save_result`
   ```

   Use `result_delivery="artifact"` for saving, section navigation, multiple documents, or strict context control. For `result.kind=artifact`, the preview is not complete; read until `next_cursor` is absent. `read_outline` does not require `save_result`. Text output is Markdown and may retain headings, lists, GFM tables, or raw HTML tables; it lacks grounding/layout sidecars, not all structure. For sidecars, request `detail=grounded`/`layout` (a `kind: bundle`). An empty outline means no recognized headings, not that the text has no structure. For multiple sources, use bounded concurrent independent `parse` calls and keep handles separate.
8. **Finish and clean up.** Continue the user's original task after parsing. For a summary, assemble the complete result first; do not truncate. For parse only, return complete Markdown or a file and optionally offer a summary. Keep artifacts through the task, then `discard_result` unless retained. Claim deletion only after discard or cleanup is confirmed.

## Operation states

| State | Action |
|---|---|
| `processing` | Continue the same operation and report authoritative progress. |
| `completed` | Consume the inline or artifact result. |
| `cleanup_pending` | Use an available result; do not claim deletion or resubmit. |
| `failed` | Surface the structured error; retry only when `retryable=true` and state permits. |
| `canceled` | Report confirmed cancellation, billing, and cleanup facts. |
| `expired` | Explain expiration; get confirmation before new work. |

For any state not recognized, preserve the operation; do not resubmit or claim completion, cancellation, billing, or cleanup.

Same-source retries reuse only recoverable or retained-completed records; failed, canceled, expired, or out-of-window retries may create and bill a new operation. On a reusable record, `auto`→`artifact` strengthens delivery without a second operation. If the user asks to stop an active operation, call `cancel_parse` with the saved ID. Discard is not cancellation; you cannot promise cancellation avoided charges.

## Unknown client capabilities

Classify Tasks, Roots, host timeout, and cwd/workspace only from direct client evidence, never from a successful parse.

- **Tasks unknown** → use ordinary `parse` plus bounded `get_parse_status` polling.
- **Roots unknown** → use process cwd and explicit roots only.
- **Host timeout unknown** → preserve Bridge's bounded 20-second status wait.
- **Cwd/workspace unknown** → do not widen authorization or guess the active workspace.

## Truthful error and billing handling

A tool-level error is not an MCP disconnection. Preserve authentication, billing, parser, retryability, operation, and cleanup facts. Report billing facts returned for this operation; retry only when `retryable=true`. Never estimate charges or rates.

On a remote-only surface, `get_billing_details` returns the same measurement for a settled operation. When a completed result includes `billing.units`, explain this operation's charge from those counters. The closed set is `document_class_pages_completed`, `primary_image_class_units_completed`, `additional_image_class_units_completed`, `cjk_codepoints_completed`, `other_codepoints_completed`, `normalized_units_completed`, and `normalized_pages_completed`, plus `audio_ms_completed` or `video_ms_completed` only when present. Zeros are real measurements. Missing `units` is not a failure: remote URL results stay credits-only, and a details read may be absent. Do not invent units, recompute `credits_charged`, or add normalized counts to pages. Never copy billing text into `result.text` or a saved file.

Run `npx -y @cueai/omni-reader-mcp@1.8.9 doctor --json` first. `CUBE_UNAVAILABLE` is pre-upload control-plane failure; post-grant failure is secure upload stage; `CUBE_PROTOCOL_ERROR` is contract mismatch. Use reported facts only; never publish internal hosts or ports.

- `OMNI_NOT_ENTITLED` / HTTP 403 is the account-entitlement signal.
- `DIRECT_UPLOAD_DISABLED` (legacy) or `DIRECT_UPLOAD_UNAVAILABLE` means the direct-upload route/capability is unavailable, not that the account is disabled or text-only.
- `DETAIL_CAPABILITIES_UNAVAILABLE` means grounded/layout is not advertised; text remains Markdown and may retain headings, lists, and tables.
- `UNSUPPORTED_DETAIL` means the requested representation/profile is unavailable; do not retry unchanged or describe the account as text-only.
- `BRIDGE_UPGRADE_REQUIRED`: install the latest published `@cueai/omni-reader-mcp` release and retry once. If already running the latest published release, do not reinstall or retry; run `doctor --json` and ask the service operator to verify Bridge admission.

## Protocol boundary

Remote-only exposes exactly `parse`, `get_parse_status`, `cancel_parse`, and `get_billing_details`. Bridge exposes the same four plus `read_result`, `read_outline`, `discard_result`, and `save_result`. Only `parse` is a first call; the other seven are continuation, lifecycle, and billing-transparency primitives, not user modes. `get_billing_details` is read-only and never billed: for a settled remote operation it returns the closed `omni.billing_details.v1` measurement, and absence is a closed BILLING_DETAILS_* error, never invented units. Artifact tools stay Bridge-local. Evidence: [`references/compatibility.md`](references/compatibility.md).

## Granted credits

New users can try Omni; current policy comes from the live server and `doctor`. Details: [`references/setup.md`](references/setup.md).

From sensedeal/cue-skills, MIT.

More from sensedeal/cue-skills

Signals

Stars
6
Forks
0
Open issues
0
Last commit (this folder)
Oct 17, 2026 (11 days ago)
Last release
Oct 9, 2026 (19 days ago)

From GitHub (fetched Oct 24, 2026).

Details

Version
0.5.4
License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from sensedeal

Cite this page

cue-omni-reader skill. Agent Mart. https://agentmart.dev/skills/sensedeal-cue-omni-reader (last updated 2026-10-24)