Skip to main content

Coding Agent Toolkit

Comparative reference · reviewed 2026-08-01

Nine durable capabilities. Three fast-moving products. One evidence model that separates documented behavior from operational proof.

Evidence standard · reviewed 2026-08-01

Compare documented capability, not feature names.

Capability pages compare documented product behavior. Deep implementation lenses retain product-specific examples where portability would be misleading. Documentation evidence is not runtime conformance evidence.

Native
The official product documents a first-party implementation on this surface.
Partial
The product covers part of the capability or carries a material documented limit.
External
The capability depends on an external tool, extension, or integration.
No documented equivalent
The reviewed official sources did not document an equivalent as of the review date.
Unknown
The evidence reviewed was insufficient to make a defensible classification.

Product projections

Follow a product without creating three competing taxonomies.

Each product view is generated from the same capability and source registry used below. Product names never become the information architecture.

Claude CodeAnthropic's agentic coding environment across terminal, IDE, and hosted execution surfaces.cli · ide · cloudOpenAI CodexOpenAI's coding agent across local CLI and IDE clients, ChatGPT, and cloud execution.cli · ide · web · cloudGitHub CopilotGitHub's coding-agent family across Copilot CLI, editors, GitHub surfaces, and cloud agents.cli · ide · web · cloud

Capability index

The nine durable decisions

Start with the problem you need to solve, then compare product behavior at the exact CLI, IDE, web, or cloud surface where the work will run. An “Unknown” marks a dated evidence gap only for the surface beside it, never a product-wide verdict.

  1. 01Project InstructionsPersistent repository guidance, scope, precedence, and instruction architecture
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE · Cloud/NativeWeb/Unknown
    Copilot
    CLI · IDE · Cloud/NativeWeb/Partial
    4 implementation lenses
  2. 02Lifecycle HooksEvent-driven automation, validation, and the limits of hook-based enforcement
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE/NativeCloud/No documented equivalentWeb/Unknown
    Copilot
    CLI · Cloud/PartialIDE · Web/Unknown
    4 implementation lenses
  3. 03Reusable SkillsProgressively disclosed instructions, scripts, resources, and invocation policy
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE/NativeWeb · Cloud/Unknown
    Copilot
    CLI · Cloud/NativeIDE · Web/Unknown
    4 implementation lenses
  4. 04Agents & SubagentsDelegated context, specialization, isolation, and coordinator-managed work
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE/NativeWeb · Cloud/Partial
    Copilot
    CLI · Cloud/NativeIDE · Web/Unknown
    4 implementation lenses
  5. 05Multi-Agent CoordinationParallel delegation, coordination topology, shared state, and peer communication
    Claude
    CLI/PartialIDE · Cloud/Unknown
    Codex
    CLI · IDE/PartialWeb · Cloud/Partial
    Copilot
    CLI/NativeIDE · Web · Cloud/Unknown
    4 implementation lenses
  6. 06MCP IntegrationExternal tools and context through Model Context Protocol, with surface-specific limits
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE/NativeWeb · Cloud/Unknown
    Copilot
    CLI/NativeCloud/PartialIDE · Web/Unknown
    4 implementation lenses
  7. 07Commands & WorkflowsInteractive commands and invocable workflows for repeatable operator intent
    Claude
    CLI/NativeIDE · Cloud/Unknown
    Codex
    CLI/NativeIDE · Web · Cloud/Unknown
    Copilot
    CLI/NativeIDE · Web · Cloud/Unknown
    4 implementation lenses
  8. 08Settings & PolicyConfiguration scope, permissions, model choice, and managed controls
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE/NativeWeb · Cloud/Unknown
    Copilot
    CLI/NativeIDE · Web · Cloud/Unknown
    4 implementation lenses
  9. 09Persistent MemoryCross-session recall, storage boundaries, provenance, and stale-context risk
    Claude
    CLI · IDE/NativeCloud/Unknown
    Codex
    CLI · IDE/NativeWeb · Cloud/Partial
    Copilot
    CLI/NativeIDE · Web · Cloud/Unknown
    4 implementation lenses

Implementation depth

Four lenses, with product scope stated explicitly

Every lens is capability-first and current across the reviewed product corpus. Product classifications remain documentation findings until independently runtime-tested.

  1. 01Mental ModelThe mechanism, constraints, and decision boundary behind the capability.
  2. 02PlaybookA concrete implementation path, including the product-specific details.
  3. 03CompositionsHow the capability combines with other controls into a delivery system.
  4. 04PitfallsFailure modes, false assurances, and operational limits to test.

Where This Connects

Products change faster than the engineering problems. The Patterns library names the portable techniques that these capabilities implement.