The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone an external repository from a non-trusted source and install its dependencies.
Evidence: git clone https://github.com/Tommy-yw/RunbookHermes.git followed by pip install -r requirements.txt or poetry install.
[COMMAND_EXECUTION]: Provides instructions to execute shell commands for environment configuration and service deployment.
Evidence: Commands include docker-compose up -d, uvicorn apps.runbook_api.main:app, and python -m apps.runbook_api.main.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted data from observability backends (Category 8).
Ingestion points: Data enters the context from Prometheus, Loki, and Jaeger backends, as well as Alertmanager webhooks.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are documented for the ingested observability data.
Capability inventory: The agent has the ability to execute sensitive operations such as execute_rollback, create_checkpoint, and request_approval.
Sanitization: There is no evidence of sanitization or validation logic applied to the external observability content before it is processed by the model.

runbookhermes-aiops-agent