The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads sensitive application state, including chat logs (logs/nanoclaw.log), session databases (inbound.db, outbound.db), and project-specific token usage files (.jsonl). This data is periodically transmitted via HTTP POST to a local dashboard service at 127.0.0.1. While the communication is confined to the local loopback address, it involves the movement of sensitive conversational and execution data out of the core application process.
[INDIRECT_PROMPT_INJECTION]: The dashboard implementation creates an attack surface for indirect prompt injection by aggregating untrusted data from external user messages and system logs.
Ingestion points: dashboard-pusher.ts reads content directly from log files and SQLite databases within the v2-sessions directory.
Boundary markers: No explicit delimiters or boundary markers are applied to the collected message content or log lines before they are bundled into the JSON snapshot.
Capability inventory: The skill possesses extensive file system read access for application data and the ability to perform network operations to local ports.
Sanitization: The pusher logic includes a regular expression to remove ANSI escape sequences from logs, but it does not perform semantic sanitization or filtering of the ingested message data.

add-dashboard