Back to skill

Security audit

Shipyard apis

Security checks for vulnerabilities and agentic risk

Overview

The skill’s instructions fit its purpose of adding outside services, with sensible key safeguards and some dependency and privacy caveats.

Use this with a trusted full Shipyard installation because its helper scripts are supplied separately. Keep keys out of chat, provision only necessary keys and set spending caps. Include Upstash in privacy disclosures, avoid caching sensitive responses and keep private cache entries isolated by user.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (10)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 15)May include surrounding context.

md
## Rules (always)

- You do the work. Never hand the student a command you could run.
- Never ask for keys in the chat and never read `~/.shipyard/secrets.env` or `.env*` files. Keys go in with `node "${CLAUDE_PLUGIN_ROOT}/scripts/secrets-init.mjs"`; `secrets-check.mjs` shows names with a tick or a cross, never values.
- Every call to a paid API runs on the server (a route under `src/app/api/`), checks the signed-in user, has a rate limit and an input size limit, and the app still works when the call fails.
- Plain English, one question at a time with your recommended answer.
- Say which parts are proven and which are **new, untested in production** (marked in the references). Do not present the second kind as tested.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/api-keys.md (reported line 3)May include surrounding context.

md
# API keys: where to get them, names, storage, checks

All keys live in the student's keys file `~/.shipyard/secrets.env` (one `NAME=value` line each, on their computer only). Add a missing one with `secrets-init.mjs` (it adds the empty line with a hint; the student pastes the value), check with `secrets-check.mjs --live`, and give it to an app with `app-keys.mjs <appDir> NAME`. Apps read them from `process.env.NAME` on the server only. Nothing here goes in `NEXT_PUBLIC_*` names, the repo or a chat.

| Service | Env var | Where to get it | Cheap check (`--live`) |
|---|---|---|---|

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/ai-router.md (reported line 10)May include surrounding context.

md
| Provider | Env var | Base URL | Good for |
|---|---|---|---|
| OpenRouter | `OPENROUTER_API_KEY` | `https://openrouter.ai/api/v1` | one key, hundreds of models, cheap ones first |
| Groq | `GROQ_API_KEY` | `https://api.groq.com/openai/v1` | very fast, free tier, Whisper speech-to-text |
| OpenAI | `OPENAI_API_KEY` | `https://api.openai.com/v1` | best structured output and tools |
| NVIDIA | `NVIDIA_API_KEY` | `https://integrate.api.nvidia.com/v1` | free hosted open models for trials (rate-limited) |
| Gemini | `GEMINI_API_KEY` | `https://generativelanguage.googleapis.com/v1beta/openai` | free tier, long inputs, images (OpenAI-compatible endpoint) |

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/ai-router.md (reported line 11)May include surrounding context.

md
|---|---|---|---|
| OpenRouter | `OPENROUTER_API_KEY` | `https://openrouter.ai/api/v1` | one key, hundreds of models, cheap ones first |
| Groq | `GROQ_API_KEY` | `https://api.groq.com/openai/v1` | very fast, free tier, Whisper speech-to-text |
| OpenAI | `OPENAI_API_KEY` | `https://api.openai.com/v1` | best structured output and tools |
| NVIDIA | `NVIDIA_API_KEY` | `https://integrate.api.nvidia.com/v1` | free hosted open models for trials (rate-limited) |
| Gemini | `GEMINI_API_KEY` | `https://generativelanguage.googleapis.com/v1beta/openai` | free tier, long inputs, images (OpenAI-compatible endpoint) |

The setup instructs the agent to send user identifiers and cached content to Upstash without a third-party data disclosure warning.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The setup directs transmission of user IDs or IP addresses and cached answers to an external service without requiring disclosure or a sensitivity check. Hashing cache keys does not protect cached values, and prohibiting private data only under shared keys still permits undisclosed storage of private answers under user-specific keys.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/voice.md (reported line 7)May include surrounding context.

md
## Speech to text: Deepgram (`DEEPGRAM_API_KEY`)

- Pre-recorded audio: `POST https://api.deepgram.com/v1/listen?model=nova-3&smart_format=true&language=<code>`, header `Authorization: Token <key>` (Token, not Bearer), body = the audio bytes with its `Content-Type`. Text is at `results.channels[0].alternatives[0].transcript`.
- Cheaper fallback with a free tier: Groq Whisper (`POST https://api.groq.com/openai/v1/audio/transcriptions`, OpenAI format, model `whisper-large-v3-turbo`).
- Live microphone streaming needs a WebSocket and a short-lived token made on the server; never send the main key to the browser.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/voice.md (reported line 8)May include surrounding context.

md
## Speech to text: Deepgram (`DEEPGRAM_API_KEY`)

- Pre-recorded audio: `POST https://api.deepgram.com/v1/listen?model=nova-3&smart_format=true&language=<code>`, header `Authorization: Token <key>` (Token, not Bearer), body = the audio bytes with its `Content-Type`. Text is at `results.channels[0].alternatives[0].transcript`.
- Cheaper fallback with a free tier: Groq Whisper (`POST https://api.groq.com/openai/v1/audio/transcriptions`, OpenAI format, model `whisper-large-v3-turbo`).
- Live microphone streaming needs a WebSocket and a short-lived token made on the server; never send the main key to the browser.

## Text to speech: ElevenLabs (`ELEVENLABS_API_KEY`)

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/voice.md (reported line 13)May include surrounding context.

md
## Text to speech: ElevenLabs (`ELEVENLABS_API_KEY`)

- `POST https://api.elevenlabs.io/v1/text-to-speech/<voice_id>?output_format=mp3_44100_128`, header `xi-api-key: <key>`, JSON `{"text": "...", "model_id": "eleven_multilingual_v2"}`. The answer is the MP3 bytes. List voices with `GET /v1/voices`.
- Charged per character: cap text length per request (for example 1,000 characters) and cache the audio for repeated text (Supabase Storage).

## Rules

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/web-search.md (reported line 7)May include surrounding context.

md
## Fallback order

1. **Brave Search** (`BRAVE_API_KEY`): own index, free monthly allowance, good for general results. `GET https://api.search.brave.com/res/v1/web/search?q=<query>&count=10`, header `X-Subscription-Token: <key>`. Results in `web.results[]` (`title`, `url`, `description`).
2. **Serper** (`SERPER_API_KEY`): Google results, cheap per search. `POST https://google.serper.dev/search`, header `X-API-KEY: <key>`, JSON body `{"q": "<query>", "num": 10}`. Results in `organic[]` (`title`, `link`, `snippet`).
3. **SerpAPI** (`SERPAPI_API_KEY`): most engines (Google, Bing, news, maps), most expensive. `GET https://serpapi.com/search.json?engine=google&q=<query>&api_key=<key>`. Results in `organic_results[]` (`title`, `link`, `snippet`). The key is in the URL, so never log request URLs.

Bare-name triggers such as "Brave", "NVIDIA", and "Gemini" do not constrain activation to API integration tasks.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

The description explicitly says to use the skill for these names without requiring an API-related request. They also refer to browsers, graphics hardware, or nontechnical subjects, so unrelated requests could activate a skill intended to configure application services and keys.

Content

No source excerpt is available for this finding.