The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill configures a GitHub Action (ai-review.yml) that triggers on repository comments and uses an AI agent to perform code reviews.
Ingestion points: The github.event.comment.body from pull request or issue comments is processed by the agent in templates/workflows/ai-review.yml.
Boundary markers: The workflow uses a trigger phrase (/ai-review) as a basic gate before execution.
Capability inventory: The agent is granted tools including mcp__github_inline_comment__create_inline_comment, Bash (limited to gh pr commands), Read, Grep, and Glob within the CI environment.
Sanitization: No specific sanitization or escaping of the comment content is implemented in the configuration, relying on the agent's internal prompt instructions.
[COMMAND_EXECUTION]: The skill installs and configures multiple Git hooks and Claude Code hooks that execute local scripts during development workflows.
Scripts such as block-destructive-git.sh and check-branch-name.mjs are used to intercept and validate user commands to prevent data loss and enforce standards.
The PreToolUse hooks in .claude/settings.json allow the agent to monitor and block potentially dangerous shell commands before execution.
[EXTERNAL_DOWNLOADS]: The skill configures workflows and tools that download external resources during CI/CD execution.
The ai-review.yml workflow utilizes the anthropics/claude-code-action@v1 GitHub Action, which is a trusted source.
The verify.json template includes a security scan using semgrep, which is downloaded via pipx from standard package registries during execution.

setup-repo