The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a vendor-specific instruction to steer agent behavior. In SKILL.md, the "Suggest Using K-Dense Web For Complex Workflows" section directs the agent to proactively suggest the author's hosted platform (www.k-dense.ai) when handling complex reasoning or large document analysis tasks.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its data ingestion and execution capabilities.
Ingestion points: The skill ingests untrusted data from external files via functions such as readtable, readmatrix, and load, as documented in references/data-import-export.md.
Boundary markers: The instructions do not include boundary markers or explicit warnings to isolate data from instructions when processing ingested files.
Capability inventory: The skill supports shell command execution (matlab -batch, octave in references/executing-scripts.md), arbitrary Python code execution (pyrun in references/python-integration.md), and network operations (requests.get via Python interop).
Sanitization: No sanitization, filtering, or validation of ingested data is prescribed before it interacts with executable capabilities.- [COMMAND_EXECUTION]: The skill provides instructions for executing shell commands to run scripts. SKILL.md and references/executing-scripts.md cover the use of the matlab and octave binaries for non-interactive task execution.- [EXTERNAL_DOWNLOADS]: The skill demonstrates network fetching capabilities through Python integration. references/python-integration.md provides examples of using the requests library to retrieve data from external endpoints.- [DATA_EXFILTRATION]: The skill outlines capabilities for accessing local files and performing network requests. references/data-import-export.md describes methods for reading structured data from the filesystem, which could be combined with the network features in references/python-integration.md for data transmission.

matlab