The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes language-specific syntax and type checks using local system tools. Examples include running python -c for AST parsing, npx tsc for TypeScript verification, and cargo check for Rust projects.
[REMOTE_CODE_EXECUTION]: Code implementation is delegated to the mcp__codex__codex tool, which executes in a restricted workspace-write sandbox. This behavior is consistent with the skill's primary purpose as a developer automation tool.
[EXTERNAL_DOWNLOADS]: The agent interacts with external task management platforms including JIRA (Atlassian), Linear, and Azure DevOps (Microsoft) via official Model Context Protocol (MCP) servers to read ticket data and post logs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from ticket descriptions and external specification documents. The risk is mitigated by several architectural safeguards:
- Ingestion Points: Context is loaded from ticket summaries and spec files in Step 1 of the workflow.
- Boundary Markers: Tasks are structured using a rigid 7-section delegation template to constrain engine behavior.
- Capability Inventory: The skill has access to the local filesystem (git), MCP-based tracker updates, and code execution via Codex.
- Sanitization & Review: All generated code undergoes a multi-stage review loop (Step 5) including syntax checks, automated test suites, a vulnerability gate for dependency changes, and a mandatory code review pass before any commits are finalized.

task-executor