Feat: Add security rule for AI Agent Skill definitions

sources/rules/core/codeguard-0-ai-agent-skills.md

@@ -0,0 +1,38 @@
+---
+description: AI Agent Skill Security Guidelines for safe autonomous execution
+languages:
+- markdown
+- yaml
+- python
+- javascript
+alwaysApply: false
+tags:
+- agents
+- data-security
+---
+
+# AI Agent Skill Security Guidelines
+
+AI agent skills extend the capabilities of autonomous agents. They MUST be designed with explicit security boundaries to prevent abuse, privilege escalation, and unintended execution.
+
+### Tool Access and Constrainment
+- NEVER grant unconstrained tool access to an agent
+- Avoid "do anything" tools (e.g., raw bash execution) unless absolutely necessary and heavily sandboxed
+- Implement strict allowlists for commands, APIs, and accessible resources
+
+### Filesystem and Network Boundaries
+- Explicitly define filesystem boundaries (e.g., restrict reads/writes to a specific workspace or `/tmp` directory)
+- Do NOT allow agents to read sensitive system files (e.g., `/etc/shadow`, `~/.ssh/`)
+- Define network boundaries: restrict outbound requests to approved domains or internal services
+- Block access to internal metadata services (e.g., AWS IMDS `169.254.169.254`)
+
+### Dangerous Shell Executions
+- Parameterize shell commands; NEVER concatenate untrusted input directly into shell commands
+- If shell execution is required, run it inside a tightly constrained sandbox or container
+- Use least privilege for the execution environment (e.g., non-root user)
+- Implement timeouts for all tool executions to prevent resource exhaustion
+
+### Validation and Auditing
+- Validate all inputs and outputs of a skill against a strict schema
+- Log all actions taken by an agent skill for auditability
+- Include a human-in-the-loop (approval prompt) for high-risk actions (e.g., deleting files, pushing code, modifying infrastructure)

src/tag_mappings.py

@@ -13,5 +13,6 @@
     "privacy",
     "secrets",
     "web",
+    "agents",
 }