The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses uvx to fetch and execute the lintro package from an external registry during the review process. It also relies on external binaries like cursor-agent and the claude CLI.
[COMMAND_EXECUTION]: The skill executes multiple shell commands, including git, grep, sed, and uvx, to manipulate configuration and run the review engine.
[PRIVILEGE_ESCALATION]: The skill instructs the agent to use git update-index --skip-worktree on .lintro-config.yaml. This technique hides local modifications to the configuration file from git status, which can lead to configuration changes persisting without being visible to the user during normal repository inspection.
[PRIVILEGE_ESCALATION]: The skill uses cursor-agent --trust to grant workspace trust non-interactively. This bypasses the manual security prompt designed to ensure users intentionally trust the code in the directory before allowing an AI agent to operate on it.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads the contents of the current git branch and its diff against the base branch (SKILL.md).
Boundary markers: No explicit delimiters or boundary markers are used to isolate the git diff content from the agent's instructions.
Capability inventory: The skill possesses capabilities to execute shell commands, download and run remote packages, and interact with external AI model providers.
Sanitization: There is no evidence of sanitization or validation of the branch diff content before it is processed by the AI models.

lintro-review