The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses local terminal history and agent conversation logs. Terminal history frequently contains sensitive data such as environment variables, API keys, and authentication tokens. This data is read into the AI's context, potentially exposing local secrets to the AI service provider.
Ingestion points: terminal and agent sources searched via sivtr_search or sivtr s in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: Processes untrusted data from terminal output and prior AI conversations. A malicious actor could inject instructions into logs (e.g., via a failed command or printed text) that the agent later retrieves and obeys.
Ingestion points: Terminal output and AI chat history retrieved in SKILL.md and references/commands.md.
Boundary markers: The skill does not mandate the use of secure delimiters or "ignore instructions" warnings for retrieved content.
Capability inventory: The agent has shell access via the sivtr CLI tool.
Sanitization: No sanitization or filtering of log content is specified.
[COMMAND_EXECUTION]: Executes shell commands using the sivtr CLI tool to search and manage local memory.
Evidence: SKILL.md specifies commands like sivtr mcp install -y and sivtr s terminal.

sivtr-memory