The Agent Skills Directory

[SAFE]: The skill is designed for ICP signal discovery and account rescoring. It uses deterministic Python scripts to perform lexical analysis on provided datasets.
[COMMAND_EXECUTION]: The Quick Start section in SKILL.md suggests installing the deepline CLI tool via npm install -g deepline. This is a vendor-provided tool consistent with the skill's purpose and name.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (website content and job descriptions) from CSV inputs.
Ingestion points: Data enters through the website and jobs columns of input CSV files processed by scripts/analyze_signals_v2.py.
Boundary markers: The skill does not interpolate raw external data into LLM prompts; instead, it performs local lexical matching and statistical analysis.
Capability inventory: Scripts are limited to file system I/O for reading CSVs and writing JSON results. The replay-score.play.ts script interacts with the deepline dataset API.
Sanitization: scripts/analyze_signals_v2.py applies NFKC normalization to all input text and uses boundary-aware regex matching to avoid misinterpretation of content.
[DATA_EXFILTRATION]: No unauthorized network operations or exfiltration patterns were detected. The skill instructions explicitly advise keeping customer data, rules, and receipts within the local project directory.

deepline-scoring