The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external sources including system logs, Kubernetes metadata, and remote server responses. This data could potentially be manipulated by an attacker to include malicious instructions aimed at influencing the agent's diagnostic flow.
Ingestion points: SKILL.md identifies logs via journalctl, pod events via kubectl describe pod, and network payloads via curl as primary data sources.
Boundary markers: While the instructions emphasize evidence-based reasoning, the skill lacks explicit markers or delimiters to isolate untrusted external content from the agent's internal logic.
Capability inventory: The skill possesses significant capabilities, including network access (curl, dig), system log access (journalctl), and cluster management (kubectl).
Sanitization: There is no evidence of sanitization or filtering applied to the content of logs or remote responses before they are processed by the agent.
[COMMAND_EXECUTION]: The skill relies on the execution of various system-level utilities to gather diagnostic signals and verify service health.
Evidence: The discriminating-signal table in SKILL.md explicitly uses kubectl, dig, curl, openssl, ss, journalctl, pg_isready, and jq to inspect the environment.
[DYNAMIC_EXECUTION]: The skill employs performance profiling tools that interact with or attach to running processes to collect runtime data.
Evidence: As detailed in references/performance.md, the skill utilizes perf, py-spy, async-profiler (asprof), and dotnet-trace to attach to specific process IDs (PIDs) for CPU, heap, and allocation analysis.

debug-triage