The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of multiple shell commands via the ft utility (e.g., ft sync, ft search, ft classify) to manage bookmark data and interact with the local environment.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install a package from the public npm registry (npm install -g fieldtheory) and includes a command to install a companion macOS application (ft install app).
[DATA_EXFILTRATION]: To function without restricted API access, the tool extracts and manages sensitive session information, including browser cookies (ct0, auth_token) and OAuth tokens. This involves local access to sensitive browser database files and credential management.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data ingestion workflows.
Ingestion points: Untrusted bookmark content and enriched article text fetched from X/Twitter and stored in ~/.fieldtheory/bookmarks/.
Boundary markers: The skill provides no explicit instructions or delimiters to the agent to prevent it from following commands embedded within the synced bookmark content.
Capability inventory: The skill possesses extensive capabilities including shell command execution (execSync), file system modification (ft library update), and network access for data syncing.
Sanitization: There is no evidence of sanitization or filtering of external content before it is processed by the agent for classification or summarization.

fieldtheory-cli