Back to skill

Security audit

Shipyard ops

Security checks for vulnerabilities and agentic risk

Overview

A simple request about logs can steer this skill into a much broader app upgrade without a clearly stated choice to approve that expansion.

Explicitly limit work to your chosen app and task, and require a separate plan and approval before any broader upgrade. Review the full Shipyard plugin and generated workflows before enabling automation. Protect backups because they contain users' personal data.

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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (4)

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 23)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. Scripts handle `CRON_SECRET` and tokens; they never print them. Never put a secret in a command line or a file in the repo.
- Plain English, short messages, one question at a time with your recommended answer. Explain any new word in five words or fewer ("a pull request: a proposed change to review").
- Changing repo settings, adding workflows, merging, and anything that costs money: ask first, wait for a clear yes, once per action.
- Commits and pushes are the student's call. Use a branch and a pull request, never push straight to `main` once the app is live.

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/backups.md (reported line 14)May include surrounding context.

md
`node "${CLAUDE_PLUGIN_ROOT}/scripts/backup-export.mjs" <appDir>`

It saves every table the app lists as user data, plus profiles and the tables from added modules, one JSON file each, with a `manifest.json` (row counts, project id, database version). Folder: `~/.shipyard/backups/<app>/<date and time in UTC>/`. It prints the folder; read it to the student with the privacy warning. It needs the Supabase access token in the keys file (already there after setup). Never open or print the backup files in the chat, `manifest.json` included. The secrets guard blocks reading anything under `~/.shipyard/backups/` (reading, searching, `cat`, `jq`, `node -e`); listing folder names with `ls` and deleting old ones with `rm` still work, and the script prints the row counts itself.

Add `--include-auth` to also save the sign-in list: ids, emails, sign-up dates and whether the email was confirmed. No passwords, phone numbers, names or profile data.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/automerge.md (reported line 27)May include surrounding context.

md
## Set it up

1. `node "${CLAUDE_PLUGIN_ROOT}/scripts/ops-setup.mjs" <appDir> --only automerge,deps` writes `.github/workflows/automerge.yml` and `.github/dependabot.yml` (npm, weekly, grouped, at most 5 open pull requests).
2. The script reads the repo settings first and shows what it wants to change. Ask the student, then say yes for them: turn on "Allow auto-merge", turn on "Automatically delete head branches", and require the `Tests` check on `main`.
3. Branch, pull request (this one touches `.github/`, so it waits for the student's own review, which is correct), merge.
4. Try it: make a tiny text change on a branch, open a pull request with the label `automerge`, watch `gh pr checks --watch`. It should merge after green. Then try one that touches a migration and confirm it gets `needs-review`. Ask before opening test pull requests.

Single-word triggers such as "errors", "logs", and "alerts" are too broad without an explicit app-operations context requirement.

Low
Category
Trigger Abuse
Confidence
93% confidence
Finding

These terms commonly appear in debugging questions and other requests that do not ask to configure ongoing app operations. The activation description lists them independently without requiring an explicit operations intent, so unrelated requests could invoke this skill.

Content

No source excerpt is available for this finding.