The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (init_agent.py and validate_agent.py) to automate the scaffolding and structure validation of new agents. These scripts use standard Python libraries and are located within the skill package.
[SAFE]: The validate_agent.py script utilizes yaml.safe_load() to process agent frontmatter, which is a recommended security practice to prevent arbitrary code execution during the parsing of YAML data.
[DATA_EXFILTRATION]: The skill performs local file system operations, including reading and writing markdown files within standard locations like ~/.claude/agents/ and .claude/skills/. These operations are limited to the user's configuration paths and are required for the skill's primary management tasks.
[PROMPT_INJECTION]: The skill contains a minor surface for indirect prompt injection as it searches and reads existing agent and skill files to check for conflicts and domain patterns. 1. Ingestion points: SKILL.md Phase 1, Step 6 uses Glob to read existing files. 2. Boundary markers: Absent. 3. Capability inventory: init_agent.py performs file writes and SKILL.md initiates script execution. 4. Sanitization: Absent for the contents of the scanned files. This risk is considered negligible as it is inherent to the skill's primary purpose of agent management.

create-agent