The Agent Skills Directory

Indirect Prompt Injection Surface: The skill reads and processes natural language content from project files to drive de-risking logic, which could be used to influence the agent if the project files are externally modified. * Ingestion points: The skill reads evidence and vision state from .amplifier/revisioner/vision.md and .amplifier/revisioner/data/<id>/findings.md. * Boundary markers: No specific boundary markers or 'ignore embedded instructions' warnings are present when reading findings content. * Capability inventory: The skill executes local Python scripts (classify.py, archive_assumptions.py, stamp_vision.py) and has the capability to write to the project's ledger and vision files. * Sanitization: The skill uses yaml.safe_load for parsing structured data but relies on the agent to interpret the natural language content of evidence files.- Local Script Execution: The skill executes bundled Python scripts to perform state analysis and file management. These scripts are restricted to the skill's own directory and use standard libraries for processing project data.

vision-check-in