The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes user-provided shell commands to observe their runtime behavior and generate telemetry data for analysis.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests telemetry data (spans, logs) and writes 'lessons' to LoreKit. Evidence of mitigation includes ingestion points via dash0 spans query and local exporters, defined boundary markers in the receipt format, a capability inventory restricted to necessary CLI tools, and a sanitization process in rules/lessons.md that drops secret-bearing content.
[SAFE]: The skill uses standard OTel protocols and vendor tools for their intended purposes. Analysis found no evidence of malicious patterns, obfuscation, privilege escalation, or persistence. The author provides clear documentation on resource attribute usage and cardinality safety.

observe-run