The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a Python utility (scripts/discover-themes.py) to manage theme discovery. This script dynamically loads and executes an external component (validate-theme-manifest.py) using importlib.util. The path to this external script is constructed at runtime from environment variables and the local directory structure, representing a dynamic loading pattern from computed paths.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It reads and parses theme.md and manifest.json files from both standard and workspace directories. Content from these files, such as theme names and descriptions, is extracted using regular expressions and used directly to generate options for the AskUserQuestion tool and as parameters for downstream skills.
Ingestion points: theme.md and manifest.json in directories defined by CLAUDE_PLUGIN_ROOT and COGNI_WORKSPACE_ROOT.
Boundary markers: No delimiters or instructions to ignore embedded content are used when processing the text extracted from theme files.
Capability inventory: Access to bash for script execution, interactive AskUserQuestion, and standard file manipulation tools.
Sanitization: Metadata from theme files is utilized directly without sanitization or validation against unexpected instructions.
[COMMAND_EXECUTION]: The discovery script implements an auto-discovery mechanism that scans the user's filesystem (including home and cloud storage directories) for .workspace-config.json files if configured paths are stale, which involves broad read-access to the local environment.

pick-theme