The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow is designed to ingest and analyze untrusted external data such as crash artifacts, input vectors, and source code to identify vulnerabilities, which creates a potential surface for indirect injection.
Ingestion points: SKILL.md specifies that the agent should establish target versions, build flags, input vectors, and crash artifacts for triage and reproduction.
Boundary markers: The instructions do not mandate the use of delimiters or specific prompts to ignore instructions that might be embedded within the crash artifacts or target files.
Capability inventory: The skill files provided are limited to markdown instructions and YAML configuration; no scripts, tools, or executable code are included in the skill package.
Sanitization: There is no evidence of sanitization, filtering, or escaping logic for the external content processed during the triage and assessment phases.
[NO_CODE]: The skill consists entirely of instructional markdown and configuration metadata. It does not ship with any scripts, binaries, or automated tools, which limits the risk of direct execution or unauthorized command runs by the skill itself.

exploit-development