The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from user-provided MATLAB source files to generate documentation and the toolbox specification. * Ingestion points: MATLAB source files (.m, .mlx) are scanned for H1 help text and arguments blocks as described in references/phase-1-scope.md and references/phase-4-documentation.md. * Boundary markers: The skill does not explicitly define delimiters or warnings for external content in the generated README or specification artifacts. * Capability inventory: The agent can execute shell commands (gh CLI), run MATLAB code (evaluate_matlab_code), and write files to the project directory. * Sanitization: There is no explicit sanitization described for content extracted from source code comments. * Mitigation: The skill enforces strict human-in-the-loop checkpoints at every phase, requiring the user to approve all generated content before it is persisted or executed.
[DYNAMIC_EXECUTION]: The skill generates a buildfile.m automation script based on project discovery and executes it via the buildtool utility. This involves script generation and execution based on internal templates (scripts/buildfile-template.m) and requires explicit user approval of the build plan in Phase 5 before execution in Phase 7.
[COMMAND_EXECUTION]: During the publication phase (references/phase-8-publish.md), the skill may invoke the GitHub CLI (gh) to create releases. This integrates with a well-known service and is gated by a mandatory confirmation prompt before the command is executed.

matlab-package-toolbox