The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest content from GLOSSARY.md and GLOSSARY-MAP.md to guide its language. This creates a potential surface for indirect prompt injection if those files contain malicious instructions. However, the risk is negligible as the skill performs no dangerous actions (network access, file writes, or command execution) and is restricted to styling the agent's response output.
Ingestion points: GLOSSARY.md and GLOSSARY-MAP.md referenced in SKILL.md and TRADUCAO.pt-BR.md.
Boundary markers: None (relying on standard agent context handling).
Capability inventory: No subprocess calls, file-system writes, or network operations detected across all files; the skill is restricted to model instructions.
Sanitization: None (the skill assumes the referenced glossary files are trusted project components).

wait-what