The Agent Skills Directory

[SAFE]: The skill is instructional and provides templates that follow industry-standard security practices, such as using @env for sensitive credential management and @assert for mandatory input validation.- [COMMAND_EXECUTION]: The workflow DSL includes a @shell.exec directive for running local commands. The documentation specifies that this feature is restricted to an allowlist of commands (e.g., echo, sleep, date, jq, curl) and performs automatic variable escaping to mitigate shell injection risks.- [EXTERNAL_DOWNLOADS]: Templates for data pipelines and service integrations use directives like @call mcp.fetch_data and @call mcp.http_post to interact with remote endpoints. These are documented as standard features for ETL and API integration use cases.- [PROMPT_INJECTION]: The skill provides templates for multi-agent orchestration that process user-provided inputs. It addresses potential indirect prompt injection risks by recommending quality review cycles, confidence-based thresholds, and human-in-the-loop handoffs for low-confidence results. Evidence Chain: 1. Ingestion points: templates/data-pipeline.md (data_source_url), templates/agent-research.md (research_topic). 2. Boundary markers: Absent. 3. Capability inventory: shell execution, network/filesystem MCP tools, and agent delegation. 4. Sanitization: Framework documentation claims automatic escaping for shell commands; prompt interpolation relies on high-level logic and quality review steps.

workflows-creator