The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the cawplan CLI tool and local Node.js/Python scripts using the Bash tool to fetch log data and serve a local web interface. These operations are restricted to the intended QA and data labeling functionality.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface for human-input logs which may contain untrusted content.\n
Ingestion points: Log data is fetched from the CawPlan server using the cawplan CLI in scripts/build_sample.mjs.\n
Boundary markers: The skill manages data using structured JSON formats and employs HTML escaping in the web-based labeling interface.\n
Capability inventory: The skill is capable of writing sampled data to local files and hosting a local HTTP server for labeling purposes.\n
Sanitization: Preprocessing logic in scripts/classify_context.mjs performs text sanitization, including the removal of decorations and emojis, normalization of escapes, and truncation of rune lengths to align with production safety rules. The labeling editor in assets/label-editor.html uses an escapeHtml function to prevent script injection when displaying log content.

cawplan-internal-qa-coding-humaninputs-sample