The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation references official installation scripts and binaries from the trusted domain claude.ai (owned by Anthropic). These references are part of the standard setup process for the tools described.
[REMOTE_CODE_EXECUTION]: Scripts provided in cicd-setup.md, gitlab-ci.md, and self-hosted-environments-deploy.md demonstrate downloading and executing the official Claude CLI installation script. These actions target verified vendor sources and are intended for deployment automation.
[COMMAND_EXECUTION]: Multiple files across the scripts/ and samples/ directories provide shell commands for cloud environment configuration, git operations, and Kubernetes/Docker runner deployments. These commands are standard administrative instructions for deploying a CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill describes CI/CD integrations that ingest untrusted data from external sources such as GitHub/GitLab comments and Slack messages.
Ingestion points: Data enters the agent context via GitHub issue/PR comments, GitLab merge request events, and Slack channel mentions.
Boundary markers: The provided workflow samples utilize explicit trigger phrases (e.g., @claude) and conditional checks to limit when the agent processes input.
Capability inventory: The configuration samples allow the agent to use Bash, Read, Edit, and Write tools within CI environments.
Sanitization: The documentation provides extensive mitigation guidance, including the use of network egress restrictions, private network isolation for gateways, and restrictive permission modes (e.g., acceptEdits) to control agent behavior.

anthropic-claude-code-deploy