The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute a series of cawplan CLI commands, such as qa-insights testrail execution failures and defects create-ticket. These commands are used to interact with TestRail and manage CawPlan defect tickets through a structured workflow.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external TestRail results (e.g., test case titles and result comments) to generate defect descriptions.
Ingestion points: The agent extracts data from TestRail results via the failures and draft commands, which include potentially attacker-controlled text in the comment field.
Boundary markers: While the skill mandates human-in-the-loop confirmation gates (--confirm, --dry-run), it does not define explicit prompt delimiters to isolate untrusted external text from the agent's core instructions.
Capability inventory: The skill has the capability to write temporary JSON configuration files to /tmp and execute write operations to the CawPlan ticketing system.
Sanitization: The instructions focus on refining the content for brevity and clarity but do not specify filtering or sanitization steps to prevent the execution of instructions embedded within the TestRail comments.

cawplan-defect-ticket