The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes platform-specific commands (open on macOS, xdg-open on Linux, and start on Windows) to automatically open the generated HTML report in the user's default browser.
[COMMAND_EXECUTION]: The skill runs a bundled Python script (scripts/compute_stats.py) using Bash to perform data aggregation and metric calculations on local JSONL files. The script is documented to use only the Python standard library.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of conversation logs stored in ~/.claude/projects/**/*.jsonl.
Ingestion points: Reads all .jsonl files within a user-defined window (e.g., 60 days) to extract metrics and representative snippets.
Boundary markers: The instructions explicitly direct the agent to HTML-escape all narrative text inserted into the report to prevent cross-site scripting (XSS) or layout breakage.
Capability inventory: The skill has Read, Write, and Bash capabilities, which are used to generate, save, and open the local report.
Sanitization: The Python script includes a redaction mechanism (REDACT_PATTERNS) that uses regular expressions to replace sensitive information such as file paths (/Users/...), email addresses, and API tokens (e.g., sk-..., ghp_...) with placeholders in the evidence snippets used by the agent.

cc-canary-html