The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to read session transcripts from ~/.claude/projects/ and ~/.codex/sessions/. While these files contain sensitive interaction history, the analysis found no network modules or exfiltration patterns. All processing is local, and the resulting HTML report is self-contained with no external dependencies.
[SAFE]: To protect user privacy, the skill implements a comprehensive redaction system in scripts/lib/redact.ts. It uses regular expressions and entropy-based detection to identify and mask API keys, tokens, and private keys from providers such as AWS, GitHub, OpenAI, and Anthropic before they are included in the report.
[SAFE]: The skill mitigates potential Indirect Prompt Injection or Cross-Site Scripting (XSS) risks. Since transcripts may contain malicious content, the code consistently applies HTML entity escaping (via the esc() function in scripts/lib/tokens.ts) to all data interpolated into the HTML report.
[COMMAND_EXECUTION]: The skill uses node:child_process to facilitate opening the generated report in the user's default browser (using platform-specific commands like open, start, or xdg-open). This is a standard and benign use of command execution for a local reporting tool.
[EXTERNAL_DOWNLOADS]: The skill documentation suggests using npx tsx for users with older versions of Node.js. While this involves a runtime download, it targets the well-known and trusted NPM registry for a standard developer utility.

session-report