The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill uses WebFetch to ingest content from external URLs (e.g., llvm.org, rust-lang.github.io, eslint-config.antfu.me) which are not within the defined trusted scope. This content is used to guide the agent's code modifications and reviews. Since the skill also has capabilities to execute local scripts and modify the filesystem, a malicious update to these remote guides could lead to unauthorized system changes.
Ingestion points: SKILL.md (Step 2 fetches external guide content).
Boundary markers: Absent.
Capability inventory: scripts/setup-commitlint.sh can install packages, write files, and change permissions.
Sanitization: Absent.
Persistence Mechanisms (HIGH): The scripts/setup-commitlint.sh script creates a .git/hooks/commit-msg file. Git hooks are a form of persistence that automatically execute specified commands (in this case, npx --no -- commitlint) whenever git operations occur.
External Downloads (MEDIUM): The script scripts/setup-commitlint.sh performs automated installation of external Node.js packages (@commitlint/cli, @commitlint/config-conventional) via the local package manager. These dependencies are not pinned or verified by the skill.
Command Execution (MEDIUM): The skill script scripts/setup-commitlint.sh executes several shell commands to modify the repository environment, including chmod +x to change file permissions for the created git hook.

coding-standards