The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by directing the agent to read and execute instructions located at the end of a long 'work package prompt' generated by a command. If the work package content is controlled by a malicious actor, it could override the agent's behavior.\n
Ingestion points: The output of the spec-kitty agent workflow review command.\n
Boundary markers: Absent. The skill instructions explicitly tell the agent to 'scroll to the BOTTOM to see the completion commands', which encourages following instructions from untrusted data.\n
Capability inventory: The agent has access to shell execution for git commands, the spec-kitty CLI tool, and local Python scripts (e.g., .kittify/scripts/tasks/tasks_cli.py).\n
Sanitization: No sanitization or validation of the work package content is mentioned.\n- [COMMAND_EXECUTION]: The skill relies on the execution of multiple shell commands and local Python scripts to manage project tasks and transitions.\n
Evidence: The skill uses spec-kitty agent workflow review $ARGUMENTS, spec-kitty agent tasks move-task, and python3 .kittify/scripts/tasks/tasks_cli.py to automate the review process.\n
The $ARGUMENTS placeholder and the requirement to use a temporary file path provided within the external work package prompt create potential surfaces for command injection if the agent uses unsanitized input from the task data.

spec-kitty-review