The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A high-entropy Cloudflare API token (8uS4nHflVYMGq6m6YysHWQLKRVZMk83A-Z0gQOtg) is hardcoded within references/docker_patterns.md. Although described as a 'commented-out value', its presence in a documentation file constitutes a credential leak.
[DATA_EXFILTRATION]: The skill references multiple sensitive file paths that store credentials and cryptographic keys, providing a roadmap for potential data exfiltration:
~/.cloudflared/cert.pem (Cloudflare tunnel certificate)
~/.config/zshyzsh/secrets.zsh (Unencrypted shell secrets file)
credentials.json (Cloudflare tunnel credentials)
~/.ssh/ (Implicitly referenced via shell configuration parity)
[REMOTE_CODE_EXECUTION]: The skill recommends installing the Mise tool manager using an unverified remote script (curl https://mise.run | sh) in both references/zshyzsh_patterns.md and references/shell-context-independence.md. While Mise is a well-known tool, executing unpinned remote scripts is a significant security risk.
[INDIRECT_PROMPT_INJECTION]: The skill defines a large surface area for indirect prompt injection by instructing the agent to manage external Docker configurations, repository structures, and Project documentation.
Ingestion points: Docker compose.yml files, repository contents, and Obsidian vault documents.
Boundary markers: None present; the agent is encouraged to interact directly with these files.
Capability inventory: The skill allows for significant filesystem operations, Docker container management, and network configuration (Cloudflare API).
Sanitization: No sanitization or validation logic is defined for the external configuration files processed by the skill.

delonet-conventions