The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input material (e.g., from the Madman/Whirlybird phases), creating a surface where malicious instructions embedded in that content could influence the agent's behavior.
Ingestion points: Core Workflow Step 1 in SKILL.md ("Receive material").
Boundary markers: Absent; there are no explicit delimiters or instructions to ignore embedded commands in the input data.
Capability inventory: The skill defines a complex workflow involving multiple phases but does not provide its own scripts, relying on general agent tools for execution.
Sanitization: Absent; the skill does not specify any filtering or sanitization of the received content.
[METADATA_POISONING]: The author metadata field in SKILL.md identifies the author as https://github.com/dmitry, which conflicts with the provided author context of jeffallan. This inconsistency in metadata could be misleading regarding the provenance of the skill.

architect