The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze external data sources which may contain malicious instructions embedded by an attacker.
Ingestion points: The agent is directed to read captured artifacts such as logs, HAR files, core dumps, and network traces, as well as user input provided through the scripts/hitl-loop.template.sh script.
Boundary markers: The skill does not specify the use of delimiters or clear boundaries to isolate these external data artifacts from the agent's instructions.
Capability inventory: The agent has the ability to execute shell commands (git bisect, curl), run scripts, and write to the file system to build reproduction loops.
Sanitization: While the skill explicitly requires redacting secrets from output, it does not provide mechanisms to sanitize or escape instructions that might be present in the ingested log data.
[COMMAND_EXECUTION]: The skill workflow relies on the execution of various CLI tools and automation scripts to verify bugs and performance regressions.
Evidence: SKILL.md recommends using git bisect run, curl, and headless browser scripts, and includes a bash template (scripts/hitl-loop.template.sh) intended for execution by the agent.

diagnosing-bugs