The Agent Skills Directory

[COMMAND_EXECUTION]: The skill and its provided templates facilitate the execution of shell commands to manage background processes, build software artifacts, and invoke LLM orchestration tools.
Evidence: The assets/workflow-script-template.sh script executes mise start, bun run build, and npx claude-flow. The SKILL.md file recommends using uv run jelmore for LLM invocations and provides patterns for log monitoring using grep and process management.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external software packages from the NPM registry to function.
Evidence: README.md and references/claude-flow-commands.md instruct the user to install claude-flow@alpha via npm install -g or execute it via npx.
[PROMPT_INJECTION]: The skill implementation pattern creates a surface for indirect prompt injection by reading requirements from a local file and including them in LLM prompts.
Ingestion points: TASK.md is read and its content is assigned to the objective variable in assets/workflow-script-template.sh and SKILL.md.
Boundary markers: Absent; input is concatenated directly into the prompt string without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The agent has capabilities for subprocess execution (via claude-flow swarm), file system access, and background process management in the assets/workflow-script-template.sh script.
Sanitization: None observed; the content of the task file is used verbatim in the prompt construction.

creating-workflows-using-claude-flow