Back to skill

Security audit

Shipyard review

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, mainly because its instructions use bare `npx tsc`, which may download and run an unexpected package, and the review gate may pass despite some reported issues.

Consider avoiding installation until you can verify that the project already has a trusted local TypeScript compiler and understand which environment and network access the skill can use. The scan is incomplete: Skillspector analyzed 8 of 14 components, with partial inspection of 6 others, and its tool-misuse analysis was degraded. AIG completed without findings, but that does not resolve the coverage gaps.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
Findings (13)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk explicitly instructs its agents not to edit files or change git state. Its two phases inventory features and existing content, then propose exact copy changes. Although agent delegation and repository inspection are compatible with the declared review process, this chunk implements a materially narrower, content-focused workflow and does not implement the advertised skeptic review, non-overlapping fix waves, dead-code removal, fix re-review, production checks, microphone testing, or stacked pull requests. No unrelated resource access or exfiltration is evident.

Content

No source excerpt is available for this finding.

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 28)May include surrounding context.

md
- Review and fix in a fresh worktree made from the latest `origin/main`, never in the student's own folder. Give every helper the worktree's absolute path and tell it never to read another checkout.
- Helper agents: your strongest model for sign-in, data rules, live audio, AI costs and the database, and for the final yes or no; a cheaper one for the other areas and for mechanical work (builds, test runs, searches).
- Never change the deploy build pipeline (build script, deploy workflow, hosting settings) during a review or a cleanup.
- Never open, print or search `~/.shipyard/secrets.env`, `.env.local` or any `.env` file other than `.env.example`.
- Commits, pushes, pull requests and merges: ask first, once per action. Never push to `main`. Merge one pull request at a time, and only after the one before it deployed and passed its smoke test.
- A helper's "fixed" or "passed" is a lead, not proof. Re-check before saying done, and say what was not checked.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

This is not an MCP server reference, but bare npx tsc can download and execute the registry package named tsc when a local TypeScript compiler is absent. Because the skill supports arbitrary apps without first requiring a local compiler, its baseline check can unexpectedly execute third-party code with the agent's permissions.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 47)May include surrounding context.

md
9. **Docs.** README against the code, the open-items doc (decisions, leftovers, what shipped), the app's `AGENTS.md` rules, `.env.example` with every variable the code reads.
10. **Ship.** Read `references/ship.md`. With a yes: commit (urgent fixes first), split into stacked pull requests if the change is over 150 files, push and open them. Merge one at a time, with a merge commit, each only with a yes and after the one before deployed and passed its smoke test.
11. **Live test** (apps that record from a microphone). After the live part deploys, run the voice test in `references/ship.md` and check the saved result. Tell the student a test record was created.
12. **Share.** Write what this review taught into the app's `AGENTS.md` and `~/.shipyard/learnings.md`, and tell the student what is left and why.

`references/lessons.md` holds the rules every app should follow after a review like this. Use it in steps 3, 5 and 7.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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/rls-speed.mjs (reported line 15)May include surrounding context.

js
console.error('usage: SUPABASE_ACCESS_TOKEN=sbp_... node rls-speed.mjs <project-ref>');
  process.exit(2);
}
const api = `https://api.supabase.com/v1/projects/${ref}`;
const headers = { Authorization: `Bearer ${token}`, 'Content-Type': 'application/json' };
const sql = readFileSync(path.join(path.dirname(fileURLToPath(import.meta.url)), 'rls-speed.sql'), 'utf8');

The review gate exits successfully for some reported advisor findings, contradicting the documented exit behavior.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The script reports three advisor categories but fails the gate only for policy-query rows or auth_rls_initplan lints. An automated review relying on its documented exit behavior can therefore pass despite reported indexing or policy-performance problems, potentially leaving availability and resource-exhaustion risks unresolved; this does not itself bypass row-level authorization.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx playwright'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

README and llms.txt copy is constrained to English regardless of the user's selected languages.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The schema instruction labels README and llms.txt as English-only and requires empty strings for other languages. Although users can select output languages, this file-specific restriction overrides that choice without explicit opt-in or a documented justification.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The fixer is instructed to execute bare npx tsc without checking for an installed TypeScript compiler. If the local executable is missing and installation is accepted, npm can fetch and execute the unrelated registry package named tsc; this is a package-resolution risk rather than an MCP version issue.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The default final-check command uses npx tsc even though this workflow does not establish that a local TypeScript dependency exists. On a project without that binary, npm may offer to download the tsc package, exposing the agent environment to unintended third-party execution.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

Every fix batch is told to run bare npx tsc, but the workflow does not verify that TypeScript is installed locally. In an app without that executable, npm can fetch the registry package named tsc, introducing unintended code execution if the installation proceeds; the MCP classification is inaccurate.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx tsc'.

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The sweep's final validation again uses npx tsc without requiring a locally installed compiler. This can resolve to a downloaded registry package rather than the intended TypeScript binary, extending package-execution risk to the cleanup phase.

Content

No source excerpt is available for this finding.