The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to prioritize inferred intent from the conversation over explicit arguments provided by the user. The instructions state: "Infers intent from the conversation; arguments optional" and "Arguments only narrow the target; their absence never blocks." This behavior could lead the agent to ignore user-provided constraints or safety parameters in favor of its own interpretation of the task state.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by requiring the agent to ingest and act upon untrusted external data sources.
Ingestion points: The agent is directed to read "real state from the source of truth," which includes "task output / logs / transcript" and "monitor status" in SKILL.md under the "Steps" and "Active-verification protocol" sections.
Boundary markers: There are no instructions for the agent to use delimiters or ignore potentially malicious instructions embedded within the logs or transcripts it reads.
Capability inventory: Based on the data processed, the agent has the capability to "resume", "rerun", or "kill-and-restart" background processes, as documented in SKILL.md step 3.
Sanitization: The skill lacks any requirement for sanitizing, escaping, or validating the content of the artifacts before the agent classifies and acts upon them.

keep-going