The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation in modules/kuva-visualization.md instructs the agent or user to install the kuva CLI tool using cargo install kuva --features cli from a third-party GitHub repository (github.com/Psy-Fer/kuva). This represents an unverifiable external dependency from a source not recognized as a trusted organization.
[DYNAMIC_EXECUTION]: The modules/kuva-visualization.md module provides instructions for generating and executing Python code at runtime using shell heredocs (`python3
<<'EOF' ... EOF`). This pattern is used to process benchmark JSON/TSV data and pipe it to external visualization tools.
[COMMAND_EXECUTION]: The skill frequently executes shell commands to facilitate its workflow, including git diff --name-only for file discovery, cargo install for dependency management, and pytest for running regression tests. It also invokes platform-specific scripts such as plugins/gauntlet/scripts/graph_build.py.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes untrusted source code files.
Ingestion points: Target source files identified via git diff or explicit user-provided paths as described in SKILL.md (Step 1 & 2).
Boundary markers: No explicit delimiters or instructions to ignore embedded content are used when the agent reads source code for analysis.
Capability inventory: The skill has capabilities for file system access, shell command execution (git, cargo, python3), and report generation.
Sanitization: While the skill uses AST parsers to identify patterns, it lacks sanitization to prevent malicious comments or strings within the analyzed code from influencing the agent's narrative report or triggering unintended actions.

performance-review