The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local file system and interpolates it into the agent's session context, creating a potential surface for indirect prompt injection.
Ingestion points: Reads content from docs/adlc/memory/adr/, .adlc/memory/adr/, and .adlc/drafts/adr/ (specifically ADR-*.md files and adr.md indexes).
Boundary markers: The skill lacks explicit boundary markers or instructions for the agent to ignore potentially malicious instructions embedded within the ADR files it summarizes.
Capability inventory: The skill performs file read and write operations to manage architectural decisions (e.g., writing drafts to .adlc/drafts/adr/).
Sanitization: There is no mention of sanitization, escaping, or filtering for the content read from the ADR markdown files before it is emitted as part of the session context.

architect-boot