The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute various shell commands to build and run feedback loops for bug reproduction.
Evidence: Phase 1 of SKILL.md lists CLI invocations, HTTP scripts via curl, and browser automation scripts (Playwright/Puppeteer) as core methods.
[PRIVILEGE_ESCALATION]: The skill documentation describes operations that typically require elevated system privileges, such as filesystem mounting.
Evidence: SKILL.md provides instructions for using mount --rbind when creating throwaway test harnesses.
Safe Practice: The skill explicitly includes defensive measures, such as the use of mount --make-rprivate and verification of mount propagation states, to mitigate the risk of host system disruption.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, which creates a vulnerability surface for indirect prompt injection.
Ingestion points: The skill reads logs, HAR files, and error messages provided by the user or captured from the environment.
Boundary markers: There are no specific delimiters or markers defined in the instructions to isolate untrusted data from the agent's instructions.
Capability inventory: The agent has wide-ranging capabilities, including shell command execution, file system access, and network operations.
Sanitization: The skill mandates the redaction of secrets from all output, which provides strong protection against data exposure but does not fully sanitize the input content against adversarial instructions.

diagnosing-bugs