The Agent Skills Directory

[SAFE]: The skill defines a standard software engineering workflow for debugging and performance analysis. The instructions are process-oriented and do not contain malicious commands or injection patterns.
[DATA_EXFILTRATION]: The skill includes a dedicated 'Redact' section that strictly instructs the agent to replace secrets with <REDACTED> and maintain credentials in environment variables rather than displaying them in captured artifacts or command outputs. This is a proactive security measure to prevent accidental data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data such as logs, traces, and HAR files provided by the user to diagnose issues.
Ingestion points: GLOSSARY.md, ADRs, logs, HAR files, and core dumps.
Boundary markers: Instructions explicitly tell the agent to ask the user for access and to redact credentials before processing artifacts.
Capability inventory: The agent can execute local shell commands (e.g., curl, git bisect), run tests, and write code for harnesses.
Sanitization: Mandated manual redaction of sensitive headers and secrets in captured data.
[COMMAND_EXECUTION]: The skill involves executing local commands for testing and reproduction, such as test runners, curl against development servers, and git bisect. These are standard development tools used within the user's project context and do not involve unauthorized access or privilege escalation.
[SAFE]: The provided bash script scripts/hitl-loop.template.sh is a benign utility designed to facilitate structured communication between the agent and the human user during manual reproduction steps.

diagnosing-bugs