The Agent Skills Directory

[PROMPT_INJECTION]: A static detector flagged an instruction directing the agent to conceal documentation retrieval steps from the user.
Evidence: SKILL.md states: "Do not tell the user to check the docs themselves. After answering, always include the relevant link so the user can explore further."
Context: This is a benign instructional directive intended to optimize user experience by having the agent perform research instead of redirecting the user. It does not attempt to hide malicious activity or bypass safety filters.
[INDIRECT_PROMPT_INJECTION]: The skill defines an interface for processing untrusted barcode data, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Barcode strings are ingested via the camera and processed by the BarcodeFind mode and BarcodeFindTransformer classes (references/integration.md).
Boundary markers: Not explicitly defined in the instructions for the handling of scanned strings.
Capability inventory: The provided code handles barcode recognition and UI updates; it does not contain capabilities for arbitrary command execution, sensitive file system access, or unauthorized network operations.
Sanitization: The skill provides documentation for BarcodeFindTransformer (references/integration.md), which allows developers to normalize or filter scanned data before it is processed by the application logic.
[EXTERNAL_DOWNLOADS]: The skill instructions include steps to download official SDK components from established registries.
Evidence: The references/integration.md file points to com.scandit.datacapture.kmp artifacts on Maven Central and the Scandit/datacapture-kmp-spm repository on GitHub. These are standard, legitimate vendor resources for the 'scandit' author.

matrixscan-find-kmp