The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs runtime compilation of system-level source files.
scripts/supervisor.mjs compiles the scripts/linux-owner.c helper using the system C compiler (/usr/bin/cc) when running on Linux.
scripts/windows-job.ps1 utilizes the PowerShell Add-Type command to compile C# code defined in scripts/windows-job.cs into the execution environment on Windows.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands and process spawning to manage tasks.
scripts/bootstrap.mjs and scripts/server.mjs use node:child_process to execute configuration updates and delegate implementation tasks.
The supervisor script on Windows invokes PowerShell using the -EncodedCommand flag to orchestrate process trees.
[PERSISTENCE]: The skill modifies host-level configuration to ensure its services are always available.
scripts/bootstrap.mjs updates the user's config.toml (typically located in ~/.codex/config.toml) to register the codex-background MCP server, causing it to be loaded automatically by the agent host.
[PRIVILEGE_ESCALATION]: The skill performs actions that require access outside the standard sandbox environment.
The installation process involves writing to configuration files and registries located in the user's home directory, outside of the immediate project workspace.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for processing untrusted instructions embedded in external files.
Ingestion points: The skill reads task "briefs" from the briefs/ directory and approved workspace roots.
Capability inventory: The system can spawn registered relay scripts, manage process groups, and perform file system operations.
Boundary markers: The server implements strict path validation (via within checks) to ensure briefs and workspaces remain within configured artifact and project roots.
Sanitization: Input paths are canonicalized and verified against an explicit registry of approved delegate skills and workspace roots.

codex-background