The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the internet, which creates an attack surface for indirect prompt injection where malicious instructions embedded in search results could be executed by the agent.
Ingestion points: Untrusted data enters the agent context through search results fetched via the WebSearch tool (file: SKILL.md).
Boundary markers: Absent. The instructions do not specify the use of delimiters or provide "ignore embedded instructions" warnings for the fetched content.
Capability inventory: The agent has access to Bash, Write, and Task tools, which could be misused if an injection attack is successful.
Sanitization: Absent. Content from search results is processed and summarized without sanitization or validation.
[COMMAND_EXECUTION]: The skill setup involves executing a local Python script provided with the skill to manage its configuration.
Evidence: The Setup section in SKILL.md instructs the user or agent to run python <skill-dir>/scripts/skillctx-resolve.py. This script interacts with a local configuration file located at ~/.config/skillctx/config.json to resolve and set skill variables.

p-news-briefing