The Agent Skills Directory

[SAFE]: The skill follows secure-by-default principles by identifying the hardcoding of secrets, tokens, or credentials as a major anti-pattern in its documentation (SKILL.md). It reinforces this through safety checklists (references/CHECKLISTS.md) that require verification of secret-free code.
[SAFE]: System capabilities like file writing and command execution are managed through a human-in-the-loop requirement. The provided prompt templates (references/TEMPLATES.md) explicitly instruct the agent to propose a plan before writing code and to 'stop and ask for confirmation' before performing any risky or destructive actions.
[SAFE]: The skill maintains clear scope boundaries, explicitly advising against its use for safety-critical, medical, or production-level systems, which reduces the risk of accidental exposure or misuse in high-stakes environments.
[SAFE]: The skill facilitates processing user input to generate executable scripts, which is an inherent surface for indirect prompt injection. However, this is mitigated by structured templates that use clear context/instruction delimiters and explicit safety constraints. Evidence: (1) Ingestion points: User ideas/context enter via the primary prompt and are processed into specs; (2) Boundary markers: Templates in references/TEMPLATES.md use structured headers like 'Context' and 'Instructions'; (3) Capability inventory: Subprocess/command execution and file writes are used for prototype scaffolding and running local environments (SKILL.md); (4) Sanitization: Managed via mandatory confirmation gates and explicit 'no secrets' instructions rather than automated filtering.

vibe-coding