The Agent Skills Directory

[COMMAND_EXECUTION]: The helper utility scripts/mcp-client.py executes the command string provided to the --stdio argument using subprocess.Popen(..., shell=True). This is a dangerous pattern that permits arbitrary code execution if the input is not strictly validated.- [COMMAND_EXECUTION]: The orchestration script scripts/fetch-docs.sh and scripts/fetch-raw.sh are vulnerable to command injection. They interpolate user-provided parameters like --library directly into shell command strings used to invoke mcp-client.py. By crafting a malicious library name (e.g., using escaped quotes and additional flags), an attacker can inject a new --stdio argument to override the intended command and execute arbitrary shell commands.- [EXTERNAL_DOWNLOADS]: The skill uses npx -y @upstash/context7-mcp to download and execute the Context7 MCP server from Upstash at runtime. While Upstash is a recognized service, runtime installation of external packages introduces a supply chain risk.- [INDIRECT_PROMPT_INJECTION]: The skill processes documentation fetched from an external API (Context7) and returns it to the agent.
Ingestion points: Documentation text fetched via fetch-docs.sh and mcp-client.py.
Boundary markers: Absent; the skill does not use delimiters or warnings to separate external content from instructions.
Capability inventory: The skill can execute shell scripts and arbitrary commands via the mcp-client.py wrapper.
Sanitization: Absent; the skill uses awk and grep for filtering content but does not sanitize or escape the data to prevent instructions from influencing the agent.- [DYNAMIC_EXECUTION]: The skill dynamically constructs and executes shell command strings at runtime through its internal MCP client implementation.

context7-efficient