The Agent Skills Directory

[DATA_EXPOSURE]: The skill manages an ingest token (OBSERVE_TOKEN) but correctly instructs the agent to use shell environment variables rather than handling the raw secret in the conversation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted output from Kubernetes and Helm commands. It implements a robust wrap helper using nonces and character sanitization (via sed and tr) to prevent terminal injection and indirect prompt injection attacks. Evidence: references/untrusted-output.md defines a wrapper that sanitizes ANSI escapes and control characters.
[COMMAND_EXECUTION]: The skill generates shell commands for the user to execute manually. It explicitly prohibits the agent from running these commands directly in its sandboxed shell to ensure user oversight and safety. Evidence: SKILL.md states 'Do NOT run any of the commands in this skill from the agent shell.'
[EXTERNAL_DOWNLOADS]: The skill fetches Helm charts and documentation from the vendor's official resources. Evidence: SKILL.md references https://observeinc.github.io/helm-charts and https://docs.observeinc.com.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses platform-specific dynamic context commands (!) to ensure the agent loads the latest instructions and referenced files. Evidence: SKILL.md calls !observe skill view ....

setup-k8s-collection