The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, including release scopes, change lists, historical failure reports, and monitoring evidence (specified in prompts/production-verification.md). This data ingestion pipeline creates a surface for indirect prompt injection, where an adversary could embed malicious instructions within release logs or change descriptions to manipulate the agent's planning output.
Ingestion points: External materials provided by the user such as release scopes, change lists, monitoring evidence, and stakeholder concerns.
Boundary markers: The skill lacks explicit instructions for using delimiters or boundary markers to isolate untrusted user data from the system prompt instructions.
Capability inventory: The skill does not explicitly request tool access in its frontmatter, though it operates within the agent's default environment which may include file and shell access.
Sanitization: There are no defined sanitization or filtering steps for the ingested external content, although the instructions in prompts/production-verification.md do require the agent to audit input for credibility and recency.

production-verification