The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources during its research and monitoring stages, creating a potential surface for indirect prompt injection.\n
Ingestion points: External data is ingested from organic SERP results, competitor page extractions, and various AI visibility engines as defined in references/listen-contract.md and the Research stage in SKILL.md.\n
Boundary markers: Present. The skill enforces a strict protocol where native Pipeline state is the sole authority, and local files are treated only as reconstructable working memory (SKILL.md, references/artifact-contract.md).\n
Capability inventory: The skill possesses the ability to execute local Python scripts (scripts/scan_content_work.py, scripts/benchmark_pipeline_trace.py), write to the local filesystem (.dineway/content/), and publish content to a CMS (SKILL.md).\n
Sanitization: Present. The backend hydrates authoritative evidence and performs atomic validation of payloads during stage transitions, as noted in the Invariants section of SKILL.md.\n- [DYNAMIC_EXECUTION]: The included unit tests utilize dynamic loading to verify the functionality of the skill's utility scripts.\n
Evidence: The test files scripts/test_benchmark_pipeline_trace.py and scripts/test_scan_content_work.py use importlib.util to load their respective logic modules (benchmark_pipeline_trace.py and scan_content_work.py) using relative paths computed within the skill's directory.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute internal Python scripts for scanning work state and validating performance traces.\n
Evidence: SKILL.md and references/performance-benchmark.md contain instructions for running python3 scripts/scan_content_work.py and python3 scripts/benchmark_pipeline_trace.py. The scripts implement regex-based validation for identifiers and file paths to mitigate the risk of argument injection.

dineway-content-pipeline