The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and interprets untrusted data from external Linear issues, which could be used to facilitate indirect prompt injection attacks.\n
Ingestion points: The skill retrieves the description and acceptance criteria of Linear issues using the mcp__plugin_linear_linear__get_issue tool to generate a test plan.\n
Boundary markers: The skill instructions do not define any delimiters or strict formatting requirements to isolate the fetched data from the agent's instructions, nor do they instruct the agent to ignore potentially embedded directives within the issue body.\n
Capability inventory: The skill allows the agent to execute shell commands, read the local git history, and modify Linear ticket attachments, providing a functional path if malicious instructions are followed.\n
Sanitization: There is no explicit sanitization or validation of the fetched Linear content before it is processed by the model.\n- [COMMAND_EXECUTION]: The skill uses local shell commands to perform repository reconnaissance and handle file management.\n
Evidence: Uses git branch --show-current and git log --oneline -10 to identify issue contexts.\n
Evidence: Executes ls and cat commands to inspect project configuration files (e.g., package.json, Makefile, playwright.config.*) to identify available test harnesses.\n
Evidence: Utilizes mkdir and curl commands to manage temporary files and upload the generated test plan to the Linear upload URL.

plan-qa