The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads highly sensitive personal data from the Wispr Flow SQLite database at ~/Library/Application Support/Wispr Flow/flow.sqlite. This file contains the complete history of the user's voice dictations, including full transcripts, timestamps, and the applications where dictation occurred. While the skill is intended for local analytics, the exposure of such broad personal data to an AI agent is a high-sensitivity operation.- [COMMAND_EXECUTION]: The script scripts/wispr_dictionary.py executes the pgrep system command using subprocess.run to check if the Wispr Flow application is active before performing database modifications. This is a functional requirement for data safety but represents an execution of local system commands.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It ingests untrusted text data (user dictations) from the local database via scripts/extract_wispr.py and interpolates this content into prompts for LLM analysis without sufficient isolation.
Ingestion points: Reads from flow.sqlite (dictation history) in scripts/extract_wispr.py.
Boundary markers: None. The extracted text samples are passed to the LLM context for qualitative analysis without delimiters or instructions to disregard embedded commands.
Capability inventory: scripts/wispr_dictionary.py has capabilities to modify the local SQLite database and execute subprocesses; scripts/extract_wispr.py can write to the local file system.
Sanitization: No filtering, escaping, or validation is performed on the formattedText or asrText fields before they are provided to the LLM for analysis.

wispr-analytics