The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project code which is an untrusted data source.
Ingestion points: Reviewing and building module-federation/core plugin architecture as described in SKILL.md.
Boundary markers: None present in the analyzed skill files.
Capability inventory: The skill is capable of reviewing and building code architecture.
Sanitization: No specific sanitization or filtering of external content is documented in the instructions.
[PROMPT_INJECTION]: The instructions include a directive to ignore specific configuration metadata from a referenced file.
Evidence: The skill instructs the agent that the disable-model-invocation frontmatter in the shared skill is Claude-specific metadata and does not apply to the current Codex configuration.

sokra-mode