The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses session logs stored in ~/.claude/projects/. These files contain complete conversation histories, which may include sensitive code, configuration data, or private information discussed in previous sessions. While necessary for the skill's primary function, this involves access to highly sensitive developer data.
[COMMAND_EXECUTION]: Uses subprocess.run() to execute external tools. This includes invoking npx ccusage to compare reporting results with a well-known community tool for usage tracking.
[EXTERNAL_DOWNLOADS]: Requires the Playwright library and triggers the installation of the Chromium browser (provided by Microsoft) to enable CSS-based PDF report generation.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays text content from conversation logs, which are considered untrusted data.
Ingestion points: Reads JSONL files containing user and assistant messages from the ~/.claude/projects/ directory in scripts/claude_usage.py.
Boundary markers: Extracted message content is displayed without delimiters or instructions to ignore embedded commands.
Capability inventory: File system access and shell command execution across multiple scripts.
Sanitization: Conversation content is extracted and printed without sanitization or filtering of potentially malicious instructions that may have been logged during previous sessions.

claude-usage