The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading local documentation files to derive technical constraints.
Ingestion points: Reads from docs/decisions-*.md and current conversation context as specified in SKILL.md.
Boundary markers: None identified. The skill lacks explicit instructions to ignore or delimit potentially malicious content within the decision documents.
Capability inventory: The skill has the capability to write to the file system (creating or updating docs/decisions-*.md) based on the outcomes of its analysis.
Sanitization: No sanitization or validation of the input file content is mentioned, allowing for potential prompt injection through the documentation it is tasked to read.

architect