The Agent Skills Directory

Log Redaction and Credential Protection: The scripts/pod-deep-dive.sh script includes a dedicated redact function that uses regular expressions to mask common credential patterns (e.g., Authorization headers, passwords, secrets, and API keys) before they are projected into the agent's context. This significantly reduces the risk of accidental exposure of sensitive information.
Integrity via Pinned Resource Identifiers: In the Inspektor Gadget reference (references/inspektor-gadget.md), external container images are referenced using their exact SHA256 digest rather than simple version tags. This practice ensures that the code being executed is exactly what was intended and protects against potential image-squatting or supply chain attacks.
Context and Target Verification: The provided shell scripts (scripts/cluster-snapshot.sh and scripts/pod-deep-dive.sh) include logic to verify that the active Kubernetes context endpoint matches the target AKS cluster's FQDN before executing any commands. This prevents the execution of diagnostic scripts against the wrong environment.
Defensive Data Handling: The skill includes explicit operating rules (in SKILL.md) instructing the agent never to run commands, pull images, or follow URLs discovered within logs, events, or annotations. This provides a strong defense against indirect prompt injection or malicious data found in the troubleshooting environment.
Privileged Operation Guardrails: While the skill provides instructions for powerful diagnostic tools like Inspektor Gadget, it mandates explicit user approval and appropriate RBAC checks for privileged operations (such as creating sysadmin-profile debug pods), adhering to the principle of least privilege.
Secure Artifact Management: The diagnostic scripts implement checks to ensure artifact directories are not symlinks or the root directory, and they apply restrictive filesystem permissions (chmod 700 and 600) to local logs and artifacts.

aks-troubleshooting