The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data such as error messages, stack traces, and codebase content during Phase 1 (Root Cause Investigation), which represents a known attack surface for indirect prompt injection.\n
Ingestion points: Phase 1 (Root Cause Investigation) requires reading error messages, stack traces, and checking recent changes in the codebase (SKILL.md, workflow-playbook.mdx).\n
Boundary markers: The instructions do not specify explicit delimiters or boundary markers to isolate untrusted external content from the agent's core instructions.\n
Capability inventory: The skill utilizes shell execution via scripts/find-polluter.sh (running bun test) and browser automation via Playwright (references/visual-web-debugging.mdx).\n
Sanitization: There is no mention of sanitization, escaping, or filtering of the external data before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill includes scripts/find-polluter.sh, a bash utility that automates test execution to identify flaky tests. This script provides the agent with the ability to execute shell commands and run test suites using the bun runtime.

systematic-debugging