The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from git diffs and commit summaries, which are then incorporated into a prompt for the kiro-cli reviewer.\n
Ingestion points: SKILL.md resolves the review scope by reading the git diff summary and changed-file list in Step 1.\n
Boundary markers: Absent. There are no explicit delimiters or "ignore instructions" wrappers used when embedding the diff content into the generated prompt file (/tmp/kiro-review-prompt.txt).\n
Capability inventory: The reviewer is granted fs_read and execute_bash permissions via the run-kiro.sh helper, allowing it to interact with the file system and run commands based on the injected prompt.\n
Sanitization: Absent. The skill lacks sanitization or filtering of the diff content to prevent embedded instructions from influencing the reviewer's behavior.\n- [COMMAND_EXECUTION]: The skill relies on the Bash tool to execute a local helper script (helpers/run-kiro.sh) and the kiro-cli binary. While the script implements a "least privilege" mode for reviews, it executes shell commands using parameters derived from the repository environment and user-supplied arguments.\n- [DYNAMIC_EXECUTION]: The skill uses the Write tool to create a temporary prompt file (/tmp/kiro-review-prompt.txt) containing instructions for the kiro-cli tool. This file is then read by the run-kiro.sh script and passed to the CLI via stdin. This involves runtime assembly of instructions (prompts) from repository data and static templates.

kiro-review