Claude Buddy Card

ReviewAudited by ClawScan on May 10, 2026.

Overview

The skill is mostly coherent, but it asks the agent to extract and use your Claude Code OAuth token from macOS Keychain, which is high-impact account access for a trading-card generator.

Review this before installing. It appears to generate the promised Buddy card, but it requires the agent to read your Claude Code Keychain credential and use your Google/Gemini API key. If you proceed, avoid untrusted proxies, use a limited Google API key, and consider running or approving the credential-reading commands manually.

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

The agent will handle a Claude account bearer token; if run in an untrusted shell, logged accidentally, or routed through an unsafe proxy, that token could expose account access.

Why it was flagged

The skill instructs the agent to retrieve a Claude Code OAuth access token from macOS Keychain and use it for an authenticated account-profile request. This is disclosed and related to deriving the Buddy ID, but it is high-impact credential/session access and is not reflected in the declared requirements.

Skill content
Run this to extract the user's Claude OAuth token ... CREDS=$(security find-generic-password -s "Claude Code-credentials" -w 2>/dev/null) ... TOKEN=...accessToken ... curl ... -H "Authorization: Bearer $TOKEN" "https://api.anthropic.com/api/oauth/profile"
Recommendation

Only use this if you are comfortable granting the agent access to your Claude Code Keychain credential. The skill should declare this credential requirement clearly and ask for explicit confirmation before reading or using the token.

What this means

Your Google/Gemini API key will be used to call Google's image API and may incur provider-side logging or quota usage.

Why it was flagged

The script reads a Google/Gemini API key from the environment and sends it to the image-generation provider. This is expected for the stated feature and the README discloses the setup, but users should know a provider key is required.

Skill content
const API_KEY = process.env.GOOGLE_API_KEY || process.env.GEMINI_API_KEY; ... headers: { "Content-Type": "application/json", "x-goog-api-key": API_KEY! }
Recommendation

Use a dedicated API key with appropriate limits, and avoid setting custom API base URLs unless you trust them.

What this means

The skill may run a local command that sends the image prompt and Google API key to the configured image API endpoint.

Why it was flagged

The image script can execute the local `curl` binary as a network fallback when proxy environment variables are present. This is purpose-aligned and not shown as shell-injectable, but it is still external command execution.

Skill content
const result = execFileSync("curl", curlArgs, { maxBuffer: 100 * 1024 * 1024, timeout: 310000 });
Recommendation

Review proxy and API endpoint environment variables before running, especially on shared or managed machines.

What this means

If Bun is missing, running the fallback may execute code fetched from the package registry at that time.

Why it was flagged

The fallback command can download and execute the `bun` package through `npx` without a pinned version. This is a documented setup convenience, but it adds supply-chain dependency risk.

Skill content
If bun is not installed, use: `npx -y bun ${SKILL_DIR}/scripts/generate-image.ts ...`
Recommendation

Prefer installing Bun from a trusted source yourself, or pin and verify the package/version before using the `npx -y` fallback.

What this means

A user may underestimate that the skill handles an OAuth token, even though the token is used for an Anthropic account-profile request.

Why it was flagged

The privacy section emphasizes local handling and Google-only prompt submission, while the SKILL instructions separately use a Claude OAuth token to call Anthropic's profile API. The detailed instructions disclose this, but the privacy summary may understate credential handling.

Skill content
Privacy

- Your account UUID **never leaves your machine**
- The skill reads from YOUR local Keychain only
- Only the image prompt is sent to Google's API (no personal data)
Recommendation

Read the command steps before running; the author should update the privacy section to explicitly mention the Anthropic profile request and OAuth-token handling.