The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes tctl and tsh binaries to query Teleport cluster state. It defines a strict allowlist of read-only subcommands and provides strong instructions to prevent the execution of any write or mutating operations based on untrusted output.
[DATA_EXFILTRATION]: The skill processes sensitive access data locally. It follows best practices for data handling, such as using mktemp for temporary scratch files and using jq for post-processing, rather than sending data to external endpoints.
[PROMPT_INJECTION]: Detection of injection-like strings in the skill documentation was found to be defensive guidance. These instructions explicitly teach the agent to identify and ignore malicious commands that might be embedded in cluster metadata like node aliases or labels.
[SAFE]: The skill demonstrates exceptional security awareness by implementing robust boundary markers for untrusted data (e.g., treating output as if enclosed in tags) and requiring literal quoting for all variables reused in follow-up queries to prevent command injection.

teleport-access-review