The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to suggest or facilitate the installation of external code via the command npx skills add wondelai/skills --global. This command fetches code from an unverified repository associated with the 'wondelai' account.
[REMOTE_CODE_EXECUTION]: By encouraging the global installation of packages from an untrusted source, the skill enables potential remote code execution on the user's host environment.
[COMMAND_EXECUTION]: The skill documents and promotes shell commands for software installation that execute with high privileges, creating a vector for system compromise if the remote repository is malicious.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8) and contains metadata inconsistencies.
Ingestion points: The agent is instructed to read content from docs/MARKETING.md, docs/OFFER.md, docs/STRATEGY.md, docs/METRICS.md, and docs/CUSTOMER.md to inform its execution logic.
Boundary markers: There are no defined delimiters or instructions to ignore embedded commands within the files being processed.
Capability inventory: The skill can invoke other local skills, modify the filesystem in the docs/ folder, and execute shell commands.
Sanitization: Content retrieved from project documentation is used to drive agent behavior without validation or sanitization, potentially allowing external data to hijack the agent's flow.
Metadata inconsistency: The metadata author field ('wondelai') does not match the verified author context provided ('manish31-oss').

grow-business