The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external financial data, including third-party reports and web search results, which constitutes an indirect prompt injection attack surface. However, the skill provides comprehensive defensive instructions to mitigate this risk.
Ingestion points: The skill reads external data during 'Stage 1: Research' (调研), specifically from third-party sell-side reports and general web searches.
Boundary markers: It requires cross-referencing all data against official company filings (Stage 1) and project-specific guidelines (AGENTS.md).
Capability inventory: The skill utilizes shell access (Bash), file system access (Read/Write), and version control (Git).
Sanitization: It implements a mandatory privacy check using grep to ensure no local environment paths or personal identity information are included in the final output before it is committed to a repository.
[COMMAND_EXECUTION]: The skill uses shell commands for legitimate administrative and analytical tasks.
Evidence: Instructions include running date for baseline verification, grep for privacy audits, and executing a repository-local tool python3 tools/financial_rigor.py for mathematical validation.

deep-company-series