The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses highly sensitive user activity logs, including screen usage data and audio transcripts, by invoking the luci CLI to generate daily summaries.
[DATA_EXFILTRATION]: Distilled summaries of user activity are sent to external connectors as specified in local configuration files (~/.Life/CLAUDE.md or ~/.Life/AGENTS.md). This allows sensitive personal data to be transmitted to external endpoints.
[COMMAND_EXECUTION]: The skill executes shell commands (e.g., date, bash, and PowerShell snippets) to calculate Unix timestamps and interacts with the host system via the luci CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external sources during the distillation process.
Ingestion points: Raw text from screen captures (luci usage) and audio transcripts (luci transcript) in SKILL.md Step 2.
Boundary markers: The instructions mandate an "objective" tone and a specific YAML/Markdown schema, but there are no technical delimiters to prevent the agent from interpreting text within transcripts as new instructions.
Capability inventory: File system read/write access in the user home directory, shell command execution, and network communication through configured external connectors.
Sanitization: No sanitization or filtering is performed on the transcript or screen data before processing.
[PERSISTENCE]: The skill modifies the luci-config.json application configuration file to inject a notification object (first-distill), which ensures a specific message is displayed to the user when the Luci application is next opened.

distill-my-life