Oracle

ReviewAudited by ClawScan on May 1, 2026.

Overview

The skill is transparent about using an external Oracle CLI to send selected project files to another model, with disclosed but important risks around package trust, account/API use, remote browser automation, and stored sessions.

This skill appears coherent and purpose-aligned. Before using it, verify the @steipete/oracle package, preview exactly which files will be sent, exclude secrets and private data, confirm API-mode cost consent, secure any remote browser host, and delete stored sessions that contain sensitive project context.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

Running the command gives the external CLI normal local process access for the selected workflow.

Why it was flagged

The documented workflow executes an external npm package on demand, and the command does not pin a package version. This is purpose-aligned, but users are trusting the referenced CLI package.

Skill content
npx -y @steipete/oracle --help
Recommendation

Verify the npm package identity and publisher before use; consider pinning a known version such as @steipete/oracle@<version>.

What this means

Project files, prompts, errors, and other attached context may be uploaded to third-party model providers.

Why it was flagged

The core workflow sends selected local files and prompts to an external model through API or browser automation. This is disclosed and central to the purpose, but it crosses a data boundary.

Skill content
Oracle bundles your prompt + selected files into one “one-shot” request so another model can answer with real repo context (API or browser automation).
Recommendation

Use dry-run and files-report first, attach only necessary files, and exclude or redact secrets, private customer data, credentials, and proprietary material that should not leave your environment.

What this means

If API mode is used, requests may be billed to the user's provider account.

Why it was flagged

The skill may use a local OpenAI API key when present, but it discloses this behavior and requires explicit consent for cost-incurring API runs.

Skill content
Auto-pick: uses `api` when `OPENAI_API_KEY` is set, otherwise `browser`. ... API runs require explicit user consent before starting because they incur usage costs.
Recommendation

Confirm which engine will be used before running, require explicit approval for API mode, and use a scoped or dedicated API key where possible.

What this means

If misconfigured or exposed, a remote browser host could allow unintended access to browser automation tied to a signed-in account.

Why it was flagged

The optional remote browser mode can expose a browser automation service on all network interfaces, protected by a token. It is disclosed and optional, but it involves a signed-in machine and network-reachable control path.

Skill content
Remote browser host (signed-in machine runs automation): Host: `oracle serve --host 0.0.0.0 --port 9473 --token <secret>`
Recommendation

Use remote browser hosting only when needed, bind to a trusted interface when possible, use a strong unique token, restrict network access, and stop the service after use.

What this means

Sensitive prompts, file summaries, or review outputs could remain on disk in Oracle session storage.

Why it was flagged

The workflow stores sessions locally for later reattachment and rendering. This is useful and disclosed, but stored prompts or review context may persist after the task.

Skill content
Stored under `~/.oracle/sessions` (override with `ORACLE_HOME_DIR`). ... Attach: `oracle session <id> --render`
Recommendation

Avoid including secrets in prompts or attachments, set ORACLE_HOME_DIR to an appropriate location if needed, and clean up old sessions that contain sensitive context.