The Agent Engineering Kit

Tools

One kit, each tool in its own format.

The kit supports 18 named AI coding tools, plus any other agent that reads AGENTS.md. It’s written once and installed the way each tool’s own docs describe: Cursor rules, Copilot instructions and agents, Gemini CLI’s context.fileName, Codex TOML agents, and so on. Shared files such as AGENTS.md and .agents/skills/ are written once for every tool that reads them.

§01 · At a glance

What each tool gets.

✓ the installer writes it. “ref” (as reference): the tool has no file-based sub-agents, so the skills tell it to read .agent-kit/agents/<name>.md and do that review itself. “note”: it can’t be set safely from the project, so the preview tells you what to do. “—”: the tool doesn’t support it.

What each tool gets
ToolRulesSkillsAgentsFormat on editSecret guard
Claude Codeyesyesyesyesyes
OpenAI Codexyesyesyesnotenote
Cursoryesyesyesyesyes
GitHub Copilotyesyesyesnotenote
Gemini CLIyesyesyesnoteyes
Google Antigravityyesyesyesnotenote
Grok Buildyesyesas referencenot supportednote
Windsurf / Devin Desktopyesyesas referenceyesyes
Kiroyesyesyesnoteyes
opencodeyesyesyesnotenote
Kilo Codeyesyesyesnot supportednote
JetBrains Junieyesyesyesnot supportedyes
Augment Codeyesyesyesnoteyes
Clineyesyesas referencenot supportednote
Zedyesyesas referencenot supportednote
Ampyesyesas referencenot supportednote
Warpyesyesas referencenot supportednote
Aidernotenot supportednot supportednot supportedyes
Any other agent (AGENTS.md)yesyesas referencenot supportednot supported

This table is generated from the installer’s own tool profiles (the same data npx agent-engineering-kit --list-tools prints), so it can’t drift from what the installer writes.

§02 · Tool by tool

Where the files go, and why.

Only behaviour each tool’s own docs describe is used, and every profile records when its docs were last checked.

Claude Code claude-code

Rules
Reads AGENTS.md through a one-line @AGENTS.md import in CLAUDE.md.
Skills
Skills in .claude/skills (or a shared folder it also reads).
Agents
Agents in .claude/agents (or a shared folder it also reads).
Format on edit
After-edit hook in .claude/settings.json.
Secret guard
Deny rules in .claude/settings.json.
Detected by
CLAUDE.md, .claude

Docs, checked 2026-10-09: code.claude.com/docs/en/memory code.claude.com/docs/en/skills code.claude.com/docs/en/sub-agents code.claude.com/docs/en/hooks

OpenAI Codex codex

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Agents in .codex/agents (or a shared folder it also reads).
Format on edit
Codex hooks don't report which file was edited (only the patch text), so the kit doesn't install a format hook for Codex. The optional pre-commit check covers formatting.
Secret guard
Codex blocks file access through permission profiles in .codex/config.toml, which also change its whole sandbox policy. If you want it, add a profile that denies "**/.env*" yourself.
Detected by
.codex

Docs, checked 2026-10-09: learn.chatgpt.com/docs/agent-configuration/agents-md learn.chatgpt.com/docs/build-skills learn.chatgpt.com/docs/agent-configuration/subagents learn.chatgpt.com/docs/hooks

Cursor cursor

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Agents in .cursor/agents (or a shared folder it also reads).
Format on edit
After-edit hook in .cursor/hooks.json.
Secret guard
Entries in .cursorignore.
Detected by
.cursor, .cursorrules, .cursorignore

Docs, checked 2026-10-09: cursor.com/docs/context/rules cursor.com/docs/context/skills cursor.com/docs/context/subagents cursor.com/docs/agent/hooks cursor.com/docs/context/ignore-files

GitHub Copilot copilot

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .github/skills (or a shared folder it also reads).
Agents
Agents in .github/agents (or a shared folder it also reads).
Format on edit
Copilot's hook input doesn't document which file was edited, so the kit doesn't install a format hook for it. The optional pre-commit check covers formatting.
Secret guard
Copilot can't be told to skip .env from inside the repo. Use content exclusion in your GitHub repository or organization settings (not supported in VS Code agent mode).
Detected by
.github/copilot-instructions.md, .github/instructions, .github/agents, .github/prompts, .github/skills

Docs, checked 2026-10-09: code.visualstudio.com/docs/copilot/customization/custom-instructions code.visualstudio.com/docs/copilot/customization/agent-skills code.visualstudio.com/docs/copilot/customization/custom-agents docs.github.com/en/copilot/reference/customization-cheat-sheet

Gemini CLI gemini

Rules
Reads AGENTS.md once it's added to context.fileName in .gemini/settings.json.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Agents in .gemini/agents (or a shared folder it also reads).
Format on edit
Gemini CLI's hook input doesn't document the edited file's path, so the kit doesn't install a format hook for it. The optional pre-commit check covers formatting.
Secret guard
Entries in .geminiignore.
Detected by
GEMINI.md, .gemini

Docs, checked 2026-10-09: geminicli.com/docs/cli/gemini-md geminicli.com/docs/cli/skills geminicli.com/docs/core/subagents geminicli.com/docs/cli/gemini-ignore

Google Antigravity antigravity

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Agents in .agents/agents (or a shared folder it also reads).
Format on edit
Antigravity's hook input doesn't document the edited file's path, so the kit doesn't install a format hook for it. The optional pre-commit check covers formatting.
Secret guard
Set a deny rule for read_file(.env) in Antigravity's permission settings (Settings UI, or ~/.gemini/antigravity-cli/settings.json for the CLI).
Detected by
.agents/rules, .agents/workflows, .agents/hooks.json, .agents/mcp_config.json, .agent

Docs, checked 2026-10-09: antigravity.google/docs/rules antigravity.google/docs/skills antigravity.google/docs/subagents antigravity.google/docs/permissions

Grok Build grok

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .grok/skills (or a shared folder it also reads).
Agents
Grok Build's custom agent file format isn't documented yet. The skills point to .agent-kit/agents/ instead.
Format on edit
Grok Build has no documented after-edit hook; use the pre-commit checks.
Secret guard
Add a deny rule for reading .env to [permission] rules in .grok/config.toml.
Detected by
.grok

Docs, checked 2026-10-09: docs.x.ai/build/features/project-rules docs.x.ai/build/features/skills-plugins-marketplaces docs.x.ai/build/features/permissions

Windsurf / Devin Desktop windsurf

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Cascade has no file-based sub-agents. The skills point to .agent-kit/agents/ instead.
Format on edit
After-edit hook in .devin/hooks.json.
Secret guard
Entries in .codeiumignore.
Detected by
.windsurf, .devin, .windsurfrules, .codeiumignore

Docs, checked 2026-10-09: docs.devin.ai/desktop/cascade/memories docs.devin.ai/desktop/cascade/skills docs.devin.ai/desktop/cascade/hooks docs.devin.ai/desktop/context-awareness/windsurf-ignore

Kiro kiro

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .kiro/skills (or a shared folder it also reads).
Agents
Agents in .kiro/agents (or a shared folder it also reads).
Format on edit
Kiro's hook input doesn't document the edited file's path, so the kit doesn't install a format hook for it. The optional pre-commit check covers formatting.
Secret guard
Entries in .kiroignore. Kiro only uses .kiroignore after you add it to the kiroAgent.agentIgnoreFiles setting (IDE only).
Detected by
.kiro, .kiroignore

Docs, checked 2026-10-09: kiro.dev/docs/steering/ kiro.dev/docs/skills/ kiro.dev/docs/custom-agents/creating/ kiro.dev/docs/kiroignore/

opencode opencode

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .opencode/skills (or a shared folder it also reads).
Agents
Agents in .opencode/agents (or a shared folder it also reads).
Format on edit
opencode formats edited files with its own built-in formatters; configure them under "formatter" in opencode.json if needed.
Secret guard
opencode already denies reading .env and .env.* by default. If you keep secrets elsewhere, add them under "permission" > "read" in opencode.json.
Detected by
opencode.json, opencode.jsonc, .opencode

Docs, checked 2026-10-09: opencode.ai/docs/rules/ opencode.ai/docs/skills/ opencode.ai/docs/agents/ opencode.ai/docs/permissions/

Kilo Code kilo

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Agents in .kilo/agents (or a shared folder it also reads).
Format on edit
Kilo Code has no documented after-edit hook; use the pre-commit checks.
Secret guard
Add "permission": { "read": { "*.env": "deny" } } to kilo.jsonc (the installer doesn't edit files with comments).
Detected by
kilo.jsonc, kilo.json, .kilo, .kilocode

Docs, checked 2026-10-09: kilo.ai/docs/customize/agents-md kilo.ai/docs/customize/skills kilo.ai/docs/customize/custom-subagents kilo.ai/docs/customize/context/kilocodeignore

JetBrains Junie junie

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Agents in .junie/agents (or a shared folder it also reads).
Format on edit
JetBrains Junie has no documented after-edit hook; use the pre-commit checks.
Secret guard
Entries in .aiignore. Junie ignores .aiignore in Brave Mode.
Detected by
.junie, .aiignore

Docs, checked 2026-10-09: junie.jetbrains.com/docs/guidelines-and-memory.html junie.jetbrains.com/docs/agent-skills.html junie.jetbrains.com/docs/junie-cli-subagents.html

Augment Code augment

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .augment/skills (or a shared folder it also reads).
Agents
Agents in .augment/agents (or a shared folder it also reads).
Format on edit
Augment's hook "command" is a script path and its working directory isn't defined, so the kit doesn't install a format hook for it. The optional pre-commit check covers formatting.
Secret guard
Entries in .augmentignore.
Detected by
.augment, .augmentignore, .augment-guidelines

Docs, checked 2026-10-09: docs.augmentcode.com/cli/rules docs.augmentcode.com/cli/skills docs.augmentcode.com/cli/subagents docs.augmentcode.com/cli/hooks

Cline cline

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .cline/skills (or a shared folder it also reads).
Agents
Cline's custom agent file format isn't documented yet. The skills point to .agent-kit/agents/ instead.
Format on edit
Cline has no documented after-edit hook; use the pre-commit checks.
Secret guard
Cline is replacing .clineignore with a guard hook; until then, use its "Block Ignored File Access" plugin to keep it out of .env.
Detected by
.clinerules, .cline, .clineignore

Docs, checked 2026-10-09: docs.cline.bot/customization/cline-rules docs.cline.bot/customization/skills docs.cline.bot/features/subagents

Zed zed

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Zed has no file-based sub-agents. The skills point to .agent-kit/agents/ instead.
Format on edit
Zed has no documented after-edit hook; use the pre-commit checks.
Secret guard
Deny agent access to .env with agent.tool_permissions in your Zed settings.
Detected by
.rules, .zed

Docs, checked 2026-10-09: zed.dev/docs/ai/instructions zed.dev/docs/ai/skills zed.dev/docs/ai/tool-permissions

Amp amp

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Amp creates sub-agents through its plugin API, not files. The skills point to .agent-kit/agents/ instead.
Format on edit
Amp has no documented after-edit hook; use the pre-commit checks.
Secret guard
Amp has no ignore file; disable file reads of .env through a tool.call plugin or amp.tools.disable.
Detected by
.amp

Docs, checked 2026-10-09: ampcode.com/docs/customize/agents-md ampcode.com/docs/customize/skills

Warp warp

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Warp's agent profiles are set in the app, not files. The skills point to .agent-kit/agents/ instead.
Format on edit
Warp has no documented after-edit hook; use the pre-commit checks.
Secret guard
Restrict file reads in your Warp agent profile's permissions.
Detected by
WARP.md, .warp

Docs, checked 2026-10-09: docs.warp.dev/agents/cli/configuration/ docs.warp.dev/agent-platform/capabilities/skills/

Aider aider

Rules
Aider doesn't read AGENTS.md by itself. Add "read: AGENTS.md" to .aider.conf.yml (or run aider --read AGENTS.md).
Skills
Aider has no skills support.
Agents
Aider has no sub-agents.
Format on edit
Aider has no documented after-edit hook; use the pre-commit checks.
Secret guard
Entries in .aiderignore.
Detected by
.aider.conf.yml, .aiderignore

Docs, checked 2026-10-09: aider.chat/docs/usage/conventions.html aider.chat/docs/config/options.html

Any other agent (AGENTS.md) generic

Rules
Reads the kit's block in AGENTS.md.
Skills
Skills in .agents/skills (or a shared folder it also reads).
Agents
Generic agents get the agent instructions as reference files. The skills point to .agent-kit/agents/ instead.
Format on edit
Any other agent (AGENTS.md) has no documented after-edit hook; use the pre-commit checks.
Secret guard
Any other agent (AGENTS.md) has no way to block file reads.

Docs, checked 2026-10-09: agents.md/ agentskills.io/specification

§03 · Shared folders

Written once where tools overlap.

  1. A fixed order. Tools are processed in a fixed order. A tool that already reads a chosen folder is covered; otherwise its preferred folder is added. The same choices always give the same layout.
  2. Overlaps are flagged. Some tools read several folders (Cursor reads .claude/skills and .agents/skills), so with certain combinations they list a kit skill twice. The copies are byte-identical, and the preview says which tool is affected.
  3. Removing one tool keeps the files your other tools still use.
  4. Tested by format, not by hand. Only Claude Code has been run end to end. The other tools’ files follow their official docs and are checked by format tests, but haven’t each been loaded in the tool.