The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation and installation instructions recommend executing a remote script by piping it directly to the shell (curl ... | bash). This pattern allows for the execution of arbitrary code from a remote source without verification.
Evidence: README.md and install.sh both reference the command curl -sL https://raw.githubusercontent.com/Hainrixz/maia-skill/main/install.sh | bash.
[COMMAND_EXECUTION]: The skill's orchestrator (SKILL.md) performs extensive local system command execution to facilitate its dashboard and server functionality.
Evidence from SKILL.md Step 9:
npm install --prefix dashboard
npm run dev --prefix dashboard -- -p 3420
python3 -m http.server PORT --directory output
lsof -i :3420 (port checking)
[EXTERNAL_DOWNLOADS]: The skill performs external downloads during installation and runtime.
Evidence: The install.sh script performs a git clone of the repository and executes npm install, which downloads numerous third-party packages.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it relies on unvetted data fetched from the web and social media to generate investment advice.
Ingestion points: Market news, asset prices, and social sentiment gathered via WebSearch and WebFetch tools (referenced in SKILL.md Step 4 and references/agent-prompts.md).
Boundary markers: While the skill enforces a JSON structure for agent responses, it lacks explicit delimiters or instructions to ignore malicious directives potentially embedded in the fetched web content.
Capability inventory: The skill has significant local capabilities, including directory creation, file writing/deletion (Step 7 in SKILL.md), and shell command execution (npm, python).
Sanitization: There is no specific sanitization or filtering of external data content described before it is processed by the "Strategy Agent" and the "Translation Agent".

investment-analysis