The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and executes Model Context Protocol (MCP) server packages from the npm registry using npx. The packages include tools from trusted or well-known organizations such as Microsoft (@azure-devops/mcp), Upstash (@upstash/context7-mcp), and Bytebase (@bytebase/dbhub).
[REMOTE_CODE_EXECUTION]: Dynamically resolves and pins package versions from the npm registry and executes them via npx to run MCP servers. It also references Microsoft's official MCP endpoint for .NET documentation.
[COMMAND_EXECUTION]: Generates and executes local shell scripts in 'scripts/agent-hooks/' to enforce repository policies. These scripts provide a 'Secret read-guard' to prevent unauthorized file access and a 'Dangerous-command blocker' to intercept risky shell commands.
[PERSISTENCE]: Registers the installed hook scripts in '.claude/settings.json', ensuring that the security guardrails and tools persist across future agent sessions within the repository.
[PRIVILEGE_ESCALATION]: Grants execute permissions ('chmod +x') to the locally generated hook scripts. This is the standard procedure for installing executable hooks in this environment.
[DYNAMIC_EXECUTION]: Assembles and writes shell scripts from provided templates at runtime. These scripts implement the core hook logic and are verified for syntax before being registered.
[SAFE]: The skill implements a robust 'Secret read-guard' specifically designed to detect and block attempts to read sensitive files like .env, SSH keys, and cloud credentials.
[SAFE]: Configures an 'Audit Log' that records tool inputs locally. The skill proactively mitigates data exposure by ensuring the log directory is added to '.gitignore' before any data is recorded.

instrument-agent-dotnet