The Agent Skills Directory

[SAFE]: The skill package does not contain any executable scripts, binaries, or hidden code. It functions purely as a set of logical prompts and reference markdown files.
[NO_CODE]: This is a non-programmatic skill. It does not execute shell commands, perform file system modifications, or invoke any external scripts or libraries.
[DATA_EXFILTRATION]: While the skill collects business-sensitive information (such as shareholding structures and investment details), it uses this information solely to generate a localized advice note for the user. No unauthorized data transmission or exfiltration patterns to external servers were found.
[EXTERNAL_DOWNLOADS]: The skill references local files for its regulatory knowledge base. It directs the agent to perform research on well-known official domains such as BSE India (bseindia.com), MCA21 (mca.gov.in), and NSE India (nseindia.com), which are standard and safe practices for legal verification.
[INDIRECT_PROMPT_INJECTION]: The skill involves analyzing external documents like financial statements and company websites, which introduces a potential surface for indirect prompt injection. However, the risk is negligible because the skill operates within a restricted environment with no access to dangerous system capabilities.
Ingestion points: 'Phase 1' instructions for analyzing investor presentations, financial statements, and company website content.
Boundary markers: No specific delimiters are used to wrap external data, but the agent's persona is focused on factual extraction.
Capability inventory: Capability is limited to web search and text generation; no system-level or destructive tools are present.
Sanitization: No explicit content sanitization is described in the instructions.

indian-foreign-investment-approval-assessment-siddhi-kudalkar