The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs standard system utilities such as bash, grep, sed, awk, and jq to perform repository analysis, resolve version metadata, and manage the file-based task board artifacts. These operations are scoped to the project directory and the plugin's own installation path.
[INDIRECT_PROMPT_INJECTION]: The skill features a documentation sweep phase that reads existing repository files (e.g., TODO.md, ROADMAP.md) to migrate them into the new task board. This creates a surface for indirect prompt injection, which is addressed by partitioning the work among subagents with explicit scoping instructions and acceptance criteria.
[DYNAMIC_EXECUTION]: As its primary function, the skill dynamically generates and installs new AI agents, skills, and markdown-based rules by populating local reference templates with data derived from the repository analysis. This code generation is confined to the .claude/ directory of the target repository.
[DATA_EXPOSURE]: The skill accesses the repository structure and configuration files to derive 'domains' and 'exclusions'. It identifies and proposes moving secrets from CLAUDE.md to a gitignored CLAUDE.local.md as a security best practice, rather than exfiltrating them.

task-board-setup