The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides and instructs the agent to execute local Python scripts (scripts/scan_content_work.py and scripts/benchmark_pipeline_trace.py) to manage local workspace state and verify workflow performance. These scripts use standard Python libraries for JSON processing, hashing, and file system operations without performing unauthorized network actions.
[DYNAMIC_EXECUTION]: The provided unit tests (scripts/test_benchmark_pipeline_trace.py and scripts/test_scan_content_work.py) utilize importlib to dynamically load and test the local pipeline scripts. This is a standard software development pattern for unit testing local modules and is restricted to static file paths within the skill directory.
[INDIRECT_PROMPT_INJECTION]: The workflow ingests external data from search results (SERP) and competitor pages during the Research and Monitor stages.
Ingestion points: External content enters the agent's context through the Research stage (dineway-content-research) and monitoring specialist jobs.
Boundary markers: The pipeline utilizes "typed stage payloads" and "canonical artifacts" (JSON/Markdown) to separate raw data from agent instructions.
Capability inventory: The agent has capabilities for local file writes in the .dineway/ directory and script execution for administrative tasks.
Sanitization: All ingested content is validated against specific stage schemas and requires human-in-the-loop approval (e.g., for the Brief stage) before being used to generate final CMS drafts.

dineway-content-pipeline