The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process code snippets from a public API (snippets.nativescript.org).
Ingestion points: Snippets are ingested via GET /api/snippets/{slug} and GET /api/snippets?category=... in SKILL.md.
Boundary markers: The skill instructions do not specify boundary markers or instructions to the agent to treat the fetched code as data only, potentially leading to the agent following instructions contained within a malicious snippet.
Capability inventory: The skill has the capability to perform network requests (curl) and potentially execute code based on the patterns found in snippets (as the skill is intended for developer assistance).
Sanitization: There is no mention of sanitizing or escaping the code field retrieved from the external API before presenting it to the agent or the user.
[DATA_EXFILTRATION]: While the skill communicates with a legitimate vendor service (nativescript.org), the POST functionality for sharing snippets involves sending code to an external server.
The instructions explicitly warn not to include secrets, but the automated nature of the skill could lead to accidental exposure of environment variables or credentials if they are present in the code the user asks to share.
The skill also suggests saving editTokens to a local file (.nativescript-snippets-tokens.json). While it recommends gitignoring this file, the management of these tokens introduces a small surface for credential exposure if the local environment is compromised.

ns-snippets