The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill utilizes uvx hindsight-embed, which fetches and runs a Python package from PyPI at runtime. The package hindsight-embed is not from a trusted organization or repository defined in the security framework.
COMMAND_EXECUTION (MEDIUM): Multiple shell commands are executed via uvx to manage persistent state; if the external package were compromised, it would have full access to the agent's environment.
PROMPT_INJECTION (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8). 1. Ingestion points: Data is stored via memory retain from arbitrary task outcomes and user-provided context. 2. Boundary markers: No delimiters or specific instructions are provided to the agent to treat recalled memories as untrusted data. 3. Capability inventory: The agent has shell access via uvx and other tools. 4. Sanitization: No sanitization or validation of stored memories is performed before they are synthesized by the reflect or recall commands.

hindsight-local