The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface. It fetches data from external sources such as Linear issue URLs (Step 1 of the smoke workflow) and uses this content to derive test plans and scenarios. If an attacker controls the content of a Linear issue, they could attempt to influence the agent's behavior during the browser-based smoke test.
Ingestion points: references/workflows/smoke.md (fetches content from mcp__claude_ai_Linear__get_issue and reads local plan files).
Boundary markers: None detected. The skill directly interpolates external content into the test planning phase.
Capability inventory: The skill possesses significant capabilities including shell command execution (mkdir, ln -s), file system writes, and full browser automation via agent-browser (Step 4 and 5 of the smoke workflow).
Sanitization: No explicit sanitization of the fetched issue description or plan content was observed before it is used to plan test scenarios.

qa