The Agent Skills Directory

Data Processing Surface: The skill ingests findings and threat models from the local workspace to perform its analysis.
Ingestion points: Data is loaded from JSON files in workspace/findings/ and the markdown file workspace/kb/THREAT_MODEL.md.
Boundary markers: No explicit delimiters are specified for separating untrusted data from instructions, though the skill is instructed to independently verify all information.
Capability inventory: The skill utilizes file-writing capabilities to update findings and append to learnings.jsonl, and it may execute shell tools like jq or short scripts for these updates.
Sanitization: The skill relies on an "adversarial stance" and independent code path verification to mitigate the risk of being misled by processed data.
Command Execution for File Updates: To optimize performance and reduce token usage, the skill suggests using tools like jq or generated scripts to perform in-place edits on JSON files. This is a standard method for programmatic data manipulation within a development environment.

mantis-critic