skills/gharib89/skills/ship/Gen Agent Trust Hub

ship

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's core logic involves interpreting untrusted data from issue descriptions and comments as implementation requirements and 'effective specs'. This creates a vulnerability surface where external users can influence the agent's planning and execution.
  • Ingestion points: scripts/read-issue.sh (reads issue body and comments) and scripts/poll-pr.sh (reads PR review comments).
  • Boundary markers: The skill uses structured markdown sections to isolate criteria, but does not explicitly instruct the agent to ignore potentially malicious embedded instructions in the untrusted text.
  • Capability inventory: The skill can execute arbitrary shell commands, perform host write operations (PR creation, merging), and modify repository content.
  • Sanitization: Basic character escaping and HTML normalization are performed for data sent to host APIs.
  • [DYNAMIC_EXECUTION]: The skill is designed to execute arbitrary shell commands defined in a repository's profile (docs/agents/ship.md) during the preparation phase.
  • Evidence: scripts/prepare.sh executes the Bootstrap: line from the profile using bash -c "$bootstrap". This allows a project to define environment-specific setup tasks that are run by the agent.
  • [COMMAND_EXECUTION]: Multiple scripts facilitate the execution of arbitrary commands for the purpose of testing, probing the environment, or proving fixes.
  • Evidence: scripts/run-file.sh verbs probe and prove (via revert-red.sh) execute shell commands provided as arguments to evaluate their outcomes.
  • Evidence: scripts/merge.sh and scripts/open-pr.sh construct and output shell command lines for the agent to execute directly.
  • [EXTERNAL_DOWNLOADS]: The skill downloads an installation script from a remote server to set up host-specific command-line tools.
  • Evidence: scripts/host/ado.sh downloads the Azure CLI installation script from https://aka.ms/InstallAzureCLIDeb.
  • Context: The download targets an official Microsoft domain for a standard administrative purpose.
  • [PRIVILEGE_ESCALATION]: The tooling installation logic utilizes sudo to acquire necessary permissions for system-wide package installation.
  • Evidence: scripts/host/ado.sh invokes sudo -n bash to execute the Azure CLI installation script.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 11:13 AM