Platform Capabilities Reference
Detailed capability matrix for AI coding tools — what each platform supports and how ContextDocs integrates.
Last researched: March 2026. The AI coding tool landscape evolves rapidly — verify against each tool’s latest release notes.
Capability Matrix
| Capability | Claude Code | Gemini CLI | Copilot | Cursor | Codex CLI | OpenCode | Cline | Windsurf | Aider |
|---|---|---|---|---|---|---|---|---|---|
| Context files | |||||||||
| AGENTS.md auto-load | via @import | configurable | yes | yes | native | native | fallback | yes | no |
| Context file | CLAUDE.md | GEMINI.md | copilot-instructions.md | .cursor/rules/ | AGENTS.md | AGENTS.md | .clinerules/ | .windsurf/rules/ | CONVENTIONS.md |
| @import syntax | yes | yes | no | no | no | no | no | no | no |
| Extensibility | |||||||||
| Skills (SKILL.md) | yes | yes | yes | yes | yes | yes | merged w/cmds | yes | no |
| Custom agents | yes | yes (exp) | yes | yes (v2.4) | yes | yes | read-only | limited | no |
| Rules directory | .claude/rules/ | via extensions | .github/instructions/ | .cursor/rules/ | no | partial | .clinerules/ | .windsurf/rules/ | no |
| Plugin/extension model | plugin.json | gemini-extension.json | .github/ plugins | marketplace | skills pkg | config | MCP self-install | MCP marketplace | no |
| MCP servers | yes | yes | yes | yes | yes | yes | yes | yes | no |
| Slash commands | yes | yes | via .prompt.md | via skills | yes (25+) | yes | merged w/skills | via workflows | yes (15+) |
| Hooks | |||||||||
| Session start | yes | yes | yes (preview) | no | yes (exp) | no | no | no | no |
| Pre-tool use | yes | yes (BeforeTool) | yes (preview) | yes | no | no | yes | enterprise | no |
| Post-tool use | yes | yes (AfterTool) | yes (preview) | yes | no | no | yes | enterprise | no |
| Session end / Stop | yes | yes | yes (preview) | yes | yes (exp) | no | no | no | no |
| User prompt submit | yes | no | yes (preview) | yes | no | no | yes | no | no |
| Total hook events | 12 | 11 | 8 | 4+ | 2 | partial | 3 | enterprise | 0 |
Per-Tool Details
Claude Code
Context loading: Auto-loads CLAUDE.md from project root and ancestor directories. Does NOT auto-load AGENTS.md — use @AGENTS.md import in CLAUDE.md. Rules in .claude/rules/*.md are auto-loaded.
Extension model: Full plugin system via .claude-plugin/plugin.json. Supports skills, agents, hooks, rules, commands, MCP servers. Installed globally or per-project.
Hooks: 12 event types — PreToolUse, PostToolUse, Stop, SessionStart, SessionEnd, UserPromptSubmit, SubagentStop, PreCompact, Notification, and more. Four handler types: command (bash), http (webhook), prompt (single-turn LLM), agent (agentic verifier).
ContextDocs support: Full — all skills, hooks, agents, rules, and commands.
Gemini CLI
Context loading: Auto-loads GEMINI.md from workspace and parent directories. Configurable via settings.json to load AGENTS.md as fallback (context.fileName: ["AGENTS.md", "GEMINI.md"]). Supports @file.md import syntax.
Extension model: Full extension framework via gemini-extension.json. Extensions bundle MCP servers, context files, slash commands, skills, agents, hooks, and policy rules. Installable from npm or local directories.
Hooks: 11 events — SessionStart, SessionEnd, BeforeAgent, AfterAgent, BeforeModel, AfterModel, BeforeToolSelection, BeforeTool, AfterTool, PreCompress, Notification. JSON stdin/stdout communication. Exit code 2 blocks execution.
Agents: Experimental (requires experimental.enableAgents: true). Local and remote subagents. Agent definitions as .md files in agents/ directory.
ContextDocs support: Tier 1 — skills, hooks, agents, and policies all portable. Extension package planned.
GitHub Copilot
Context loading: Auto-loads .github/copilot-instructions.md (repo-wide) and .github/instructions/*.instructions.md (path-scoped with applyTo glob). Also loads AGENTS.md from anywhere in the tree (nearest takes precedence). Custom agents in .github/agents/*.agent.md.
Extension model: Plugins with 5 primitives — agents, skills, hooks, MCP servers, and instructions. Marketplace available. /plugin install owner/repo syntax.
Hooks: 8 events (Preview) — sessionStart, sessionEnd, userPromptSubmitted, preToolUse, postToolUse, agentStop, subagentStop, errorOccurred. Configured in .github/hooks/*.json. preToolUse can approve or deny.
Skills: Open Agent Skills standard in .github/skills/<name>/SKILL.md or .claude/skills/. Cross-tool compatible (VS Code, CLI, coding agent).
ContextDocs support: Tier 1 — skills, hooks, agents, and instructions all portable. Package planned.
Cursor
Context loading: .cursor/rules/*.mdc (modern, replacing deprecated .cursorrules). Also loads AGENTS.md with directory-scoped overrides. Four activation modes: Always Apply, Apply Intelligently, Apply to Specific Files (globs), Apply Manually.
Extension model: Plugin marketplace (since v2.4, Feb 2026). Plugins package skills, subagents, MCP servers, hooks, and rules. /plugin install owner/repo syntax.
Hooks: 4+ events — stop, beforeSubmitPrompt, PreToolUse, PostToolUse.
Agents: Subagents since v2.4. Custom subagents with own context, prompts, tool access, and models. Can load resources from ~/.claude/{skills,agents}.
ContextDocs support: Tier 1 — skills, hooks, agents, and rules all portable. Plugin planned.
Codex CLI
Context loading: AGENTS.md auto-loaded natively — walked from git root to CWD (hierarchical). Also checks ~/.codex/AGENTS.md (global). Max combined size 32 KiB default.
Extension model: Open Agent Skills standard. Skills scanned from .agents/skills at CWD, repo root, $HOME, /etc/codex/, and bundled. Activated via /skills, $skillname mention, or auto-selected by description.
Hooks: Experimental — only SessionStart and Stop events. No PreToolUse/PostToolUse yet.
Agents: Yes. Subagents spawn on explicit request. Each runs independently with own model and tool access. Configured in [agents] section of config.toml.
ContextDocs support: Tier 2 — skills and agents work, hooks limited to session events.
OpenCode
Context loading: AGENTS.md from project root (primary). CLAUDE.md as fallback (Claude Code compatibility). ~/.config/opencode/AGENTS.md (global). Additional files via instructions field in opencode.json.
Extension model: Custom agents in opencode.json or .opencode/agents/*.md. Claude Code skill directory compatibility (.claude/skills/). Can be disabled via OPENCODE_DISABLE_CLAUDE_CODE=1.
Hooks: Partial. Does not natively run Claude Code JSON hooks. Open issue for compatibility.
Agents: Four built-in (Build, Plan, General, Explore) + custom agents. Per-agent tool permissions, model overrides, step limits.
ContextDocs support: Tier 2 — skills and agents work, no automated hooks.
Cline
Context loading: .clinerules file or .clinerules/ directory at project root. Auto-detects fallbacks: .cursorrules, .windsurfrules, AGENTS.md. Supports path-based conditional activation via paths: YAML frontmatter.
Extension model: MCP self-install. Hooks in .clinerules/hooks/ (project) or ~/Documents/Cline/Rules/Hooks/ (global).
Hooks: 3 events — PreToolUse, PostToolUse, UserPromptSubmit. JSON stdin/stdout. macOS and Linux only (no Windows).
Agents: Subagents (disabled by default). Read-only — can explore but cannot write or run destructive commands.
ContextDocs support: Tier 2 — skills (merged with commands), hooks (3 events), read-only agents.
Windsurf
Context loading: .windsurf/rules/*.md (workspace, 12,000 chars per file). .windsurfrules at root (legacy). AGENTS.md at root and subdirectories. Four activation modes: always_on, model_decision, glob, manual.
Extension model: MCP Marketplace for one-click installs. Skills in .windsurf/skills/<name>/SKILL.md. Workflows invoked via slash commands.
Hooks: Enterprise only — hooks on prompts for logging/policy, hooks on responses for auditing. Not developer-extensible.
Agents: Limited. Cascade is the primary agent. Fast Context subagent for code search. No custom subagent system.
ContextDocs support: Tier 3 — skills and rules work, no hooks or agents for non-enterprise.
Aider
Context loading: No auto-loaded context file. Uses CONVENTIONS.md loaded via --read flag or .aider.conf.yml config. No native AGENTS.md support (feature requested).
Extension model: None. Slash commands only (15+ built-in). No skills, hooks, or agents. No MCP (PR pending).
ContextDocs support: Tier 4 — AGENTS.md loadable via --read AGENTS.md. Use bin/context-verify.sh for health scoring.
ContextDocs Feature Portability
| ContextDocs Feature | Portable? | Notes |
|---|---|---|
| Signal Gate principle | Yes | Tool-agnostic methodology |
| AGENTS-first architecture | Yes | AGENTS.md widely supported |
| Line budgets | Yes | Measurable standards, any tool |
| ai-context skill | Yes | SKILL.md is cross-tool standard |
| context-verify skill | Yes | Checks are tool-agnostic |
| context-verify CLI | Yes | Bash script, no plugin needed |
| context-guard hooks | Partial | Need per-tool hook adapters |
| context-updater agent | Partial | Agent format differs per tool |
| Rules (auto-loaded) | Partial | Directory varies per tool |
| Plugin manifest | No | Claude Code specific |
| settings.json hooks | No | Claude Code specific |
| @AGENTS.md import | Partial | Claude Code + Gemini CLI only |
Sources
| Tool | Documentation | Last Verified |
|---|---|---|
| Claude Code | code.claude.com/docs | March 2026 |
| Gemini CLI | geminicli.com/docs | March 2026 |
| GitHub Copilot | docs.github.com/copilot | March 2026 |
| Cursor | cursor.com/docs | March 2026 |
| Codex CLI | developers.openai.com/codex | March 2026 |
| OpenCode | opencode.ai/docs | March 2026 |
| Cline | docs.cline.bot | March 2026 |
| Windsurf | docs.windsurf.com | March 2026 |
| Aider | aider.chat/docs | March 2026 |
