The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions as a dispatcher that ingests arbitrary natural language phrases from the user and reads project-level configuration files to determine which sub-skills to invoke. This creates an attack surface where crafted input could potentially misdirect the agent's actions.
Ingestion points: User-supplied text queries and the content of the AGENTS.md file in the local repository.
Boundary markers: The instruction set includes a mandatory clarification step (Step 2) where the agent must confirm its interpretation of the user's goal if the request is ambiguous.
Capability inventory: The skill is authorized to call iskron_* platform tools, execute git and gh CLI commands, and dynamically load other functional modules (e.g., writing, design, vahta).
Sanitization: No explicit input sanitization or filtering of the user's natural language strings is specified before they are processed by the routing logic.
[COMMAND_EXECUTION]: The skill routinely executes local shell commands, specifically git status --short and interactions with the GitHub CLI (gh), to verify repository state and process external issue data for the product-roadmap and reconcile routes.
[PERSISTENCE]: Through the iskronify and vahta modules, the skill can establish long-term automation within a project, such as setting up git hooks or 'duty' modes where the agent acts autonomously on a task queue.

iskron