connect-tool
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading official API documentation from various vendor websites to generate code.
- Ingestion points:
SKILL.md(Step 3b) explicitly instructs the agent to read official API docs to determine base URLs, auth headers, and endpoints. - Boundary markers: There are no explicit instructions to ignore malicious content or hidden commands within the external documentation being parsed.
- Capability inventory: Across
SKILL.mdandscripts/cli_template.py, the skill can create files (mkdir,ln -s), execute subprocesses (curl,brew,npm,pipx,python3), and perform network operations (urllib.request). - Sanitization: No specific sanitization or escaping is mentioned for the content retrieved from external API documentation before it is used to generate the new CLI tool.
- [COMMAND_EXECUTION]: The skill performs several command-line operations to install tools and verify environments.
- Evidence:
SKILL.mdusescommand -v,brew,npm i -g,pipx,uv tool install, and execution of the generated Python scripts. - [EXTERNAL_DOWNLOADS]: The skill downloads template files and instructions from GitHub.
- Evidence:
SKILL.mdusescurl -fsSL https://raw.githubusercontent.com/Melvynx/ai-sprint-skills/main/...to fetch its own configuration and scripts. Note: These are vendor-owned resources. - [PRIVILEGE_ESCALATION]: The skill may request elevated permissions during tool installation.
- Evidence:
SKILL.md(Step 3a) instructs the agent to "Ask before sudo" when installing official CLIs. - [SAFE]: The generated Python CLI template (
scripts/cli_template.py) follows security best practices for credential management, using0o600permissions (read/write only by the owner) for its config file andgetpassfor hidden input of API keys.
Audit Metadata