Type to search the docs and updates.

Browse documentation

Global configuration reference

Setup is in Global configuration. This page holds the per-key and per-tool detail, and the list of files sync --global writes.

Shared names

A project skill or agent can share its name with one in the home. Sync writes both, and each tool decides which one it loads.

sync and doctor print one warning per shared name. It names each target where one copy hides the other, which copy wins, and the fix:

! .agnostic-ai/skills/gh-issue/SKILL.md: skill "gh-issue" also exists in ~/.agnostic-ai/skills/gh-issue/SKILL.md with different content; claude loads the global one, which exists only on this machine, gemini loads this one; to load both, rename the global one (such as gh-issue-personal), or delete it to drop it, then run `agnostic-ai sync --global`; project sync cannot select this skill instead on claude

Why sync keeps both copies:

  • Claude Code, Cline, and Amp document no project setting that picks the project copy over a home skill and keeps the other home skills.
  • Amp's amp.skills.disableGlobalAgentsSkills turns off every skill in the global agents folders.
  • So sync explains the limit and leaves your home files alone. Doctor reports it even when sync warnings are off.
  • A winning global copy changes what you run, not what others run from the repository.
  • When only project copies win, the warning says the global one is unused here. When both copies match, it says to delete one before they drift apart.

The check covers these targets:

TargetSkill with the same nameAgent with the same name
AmpGlobal wins when both sync amp: ~/.agents/skills/ masks .agents/skills/ (skills)No global agents
Claude CodeGlobal wins (skills)Project wins (subagents)
ClineGlobal wins (skills) when the project syncs cline and the home syncs cline, or any target that writes ~/.agents/skills/, which Cline also reads (config)Not checked
CodexBoth can appear in skill selectors, so no warning (skills)Not checked
Gemini CLIProject wins (skills)Not checked

Sync and doctor do not check other tools, though some document what happens:

  • Kiro prefers workspace skills over global skills.
  • Qoder IDE documents project skills taking priority over user skills. Qoder CLI documents user skills taking priority over project skills.
  • Warp prefers home skills, then higher directories, when it resolves names in the background. Its slash-command menu shows all matches, and a plain-language request gives the agent every match in scope with its path.
Which pairs the check compares
  • Targets that both the project and the home's targets write. A spec whose target, targets, or targets-exclude leaves one of them out is skipped.
  • Claude Code skill names match ignoring case, spacing, invisible characters, and fullwidth forms. Other targets compare names exactly.
  • A spec shared through a link is not a clash. Neither is a project that is the home.

To resolve a clash:

  • Layer on purpose: give the project spec its own name, such as a gh-issue-project skill beside a global gh-issue. Both then load everywhere.
  • Keep one name in both places: list it in sync.allow-global-names.
  • Silence every shared name: set sync.global-name-clash: ignore.

Default model and effort

A settings spec in the home sets each tool's default model and effort in its user settings file. model and effort each take one string for every target, or a per-target map with an optional default:

# ~/.agnostic-ai/settings/defaults.yaml
model:
  claude: opus
  codex: gpt-6-luna
effort:
  claude: high
  codex: high
TargetFileKeys
codex~/.codex/config.tomlmodel, model_reasoning_effort
claude~/.claude/settings.jsonmodel, effortLevel, permissions.defaultMode, permissions.allow, ask, deny
codebuddy~/.codebuddy/settings.jsonmodel, reasoningEffort, permissions.defaultMode, permissions.allow, ask, deny
copilot~/.copilot/settings.jsonmodel, effortLevel
qoder~/.qoder/settings.jsonmodel.name, model.reasoningEffort
qwen~/.qwen/settings.jsonmodel.name, model.reasoningEffort
vibe~/.vibe/config.tomlactive_model
kimi~/.kimi-code/config.tomldefault_model, default_plan_mode, telemetry
gemini~/.gemini/settings.jsonmodel.name
augment~/.augment/settings.jsondefaultModel, defaultThinkingLevel (Auggie 1.x)
  • Targets not in the table raise a coverage note for model and effort. So does effort on Gemini and Kimi Code.
  • Augment also takes x-augment keys, such as shell.
  • A dotted key is a nested JSON object, except for Kiro CLI's literal dotted keys. For model.name, sync sets name inside model and leaves the other keys alone.
Details: accepted effort values
TargetAccepted effort values
Claudelow, medium, high, xhigh
Copilotnone, minimal, low, medium, high, xhigh, max (each model accepts a subset)
Qoderdisabled, off, none, low, medium, high, xhigh, max
Augmentoff, minimal, low, medium, high, xhigh
Qwenlow, medium, high, xhigh, max (each provider maps the value to its active model)
CodexAny string
CodeBuddy CodeSee native fields and limits

lint --global (LINT014) and validate --global flag a value a target cannot take.

Claude Code and effortLevel:

  • ~/.claude/settings.json effortLevel is read on Opus 5, Fable 5.1, and earlier models only.
  • Opus 5.5 and later models ignore it. They need a level saved per model with /effort.
  • sync --global writes the key and prints a note.
  • In project settings, effortLevel applies to every model.

Permissions

Set Claude Code's starting permission mode in a home settings spec:

# ~/.agnostic-ai/settings/defaults.yaml
permissions:
  default-mode: acceptEdits
  • sync --global --only claude writes permissions.defaultMode in ~/.claude/settings.json, or under CLAUDE_CONFIG_DIR.
  • Accepted values: default, manual, acceptEdits, plan, auto, dontAsk, bypassPermissions.
  • A later settings spec wins.
  • Removing the field removes the managed mode, unless you changed it by hand.
  • Claude's auto and bypassPermissions need user, managed, or session settings (Claude's mode reference).

Home permissions.allow, ask, and deny rules, and the same lists under x-claude.permissions, merge into ~/.claude/settings.json rule by rule.

  • Claude Code saves standing approvals to the project's .claude/settings.local.json, which global import does not read (Claude's settings scopes).
  • CodeBuddy Code also merges user permission lists and starting modes (native limits).
  • A permissions field on any other target raises a coverage note. For Codex, set x-codex.approval_policy and x-codex.sandbox_mode instead.
Details: how permission rules merge
  • Sync appends each missing rule after the rules already there, and adopts a rule already there.
  • Sync records added and adopted rules in state/global.json.
  • Dropping a rule from the specs removes it, including a rule you wrote before adoption. Other rules keep their order.
  • --check fails while a spec rule is missing or a dropped one remains.
  • --plan and --dry-run list each rule added or removed.
  • import --global claude copies user-level rules into settings/permissions-claude.yaml, so they reach your other machines.

Other settings

  • session-retention-days sets how many days transcripts are kept, for Claude Code, CodeBuddy Code, Gemini CLI, and Qoder CLI (session retention).
  • voice configures voice dictation for tools with a native setting (voice dictation).

Which value wins

Each item wins over the one before it:

  1. settings/*.yaml in the home.
  2. local/settings/*.yaml. A same-named file merges into the shared one field by field. A new file comes after the shared ones, so its model and effort win.
  3. The project tier: a project's settings file, such as .codex/config.toml or .claude/settings.json, wins by the tool's own settings order.
  4. An agent's own model and effort, for that agent.
  5. The tool's flag for one run, such as codex -m or claude --model.

agnostic-ai explain --global settings/defaults.yaml names the file and key each target gets.

Details: how sync edits settings keys
  • Sync edits only the keys it writes.
  • Removing the spec removes only those keys and the objects sync created to hold them. An object you wrote stays, even an empty one such as "env": {}.
  • Sync deletes a file it created once the file is empty.
  • A key that already holds the value sync would write is adopted.
  • A key with another value stops the run and names the file, key, and both values. This is normal for Codex, whose /model picker saves to config.toml. --backup overwrites the key and keeps <path>.bak.
  • --dry-run lists each key a write sets or removes. --check fails on a changed key.
  • An x-<target> block sets that target's own keys in the same file. Nested objects merge leaf by leaf.
  • An x-claude key wins over the portable field it shares a key with.
  • A later spec wins key by key. null drops an earlier key.
  • Global sync writes Codex top-level scalars and arrays, such as x-codex.notify, to ~/.codex/config.toml.
  • Tables such as profiles, x-claude.hooks, and x-claude.permissions keys other than allow, ask, and deny raise a coverage note.
  • Project sync does not route x-codex into .codex/config.toml.

Approvals

approvals sets what happens when a tool wants permission for a tool call:

ModeWhat happens
askThe tool shows its prompt. The default.
reviewThe tool's own reviewer model decides routine requests. The rest still ask.
yoloEach request is approved once, unless a fixed always-ask rule matches.
# $AGNOSTIC_AI_HOME/agnostic-ai.yaml or $AGNOSTIC_AI_HOME/local/agnostic-ai.yaml
approvals:
  mode: ask            # ask | review | yolo
  tools:               # per tool, replaces mode
    codex: review
  projects:            # per project folder, replaces mode and tools
    ~/Code/sandbox: yolo
  always-ask:          # extra patterns, added to the fixed list
    - "Bash(make deploy*)"
  • review and yolo are read only from the home config. A cloned repository, a pull request, or the agent itself can write a project's files, so a project's agnostic-ai.yaml or agnostic-ai.local.yaml may set approvals.mode: ask and nothing else. Any other approvals key there is ignored, and validate and lint report it as LINT064.
  • An approvals block in local/agnostic-ai.yaml replaces the shared block whole, except always-ask: the patterns of both files add up.
  • A projects key is an absolute path or one under ~, the home folder. The deepest folder that holds the session wins; links are resolved, and on macOS and Windows case is ignored. Two keys for one folder take the stricter mode.
  • The strictest source wins, in this order: AGNOSTIC_AI_APPROVALS=ask in the session's environment, a project file that says ask, projects, tools, mode. Any value of the environment variable means ask: it can only tighten.
  • A config without approvals behaves as ask. $AGNOSTIC_AI_HOME must be an absolute path for approvals to read it.
  • Only Claude Code and Codex carry out yolo, through a hook that builtins: [approvals] and sync --global install (Approvals). Only the tools in the review table carry out review. Every other tool falls back to ask, and approvals status says so.

Review through the tool's own reviewer

When approvals.tools.<tool>, or approvals.mode without it, says review, sync --global turns on the tool's own reviewer in its user settings:

ToolReviewerKey sync --global writes
Claude CodeAuto modepermissions.defaultMode: "auto" in ~/.claude/settings.json
CodeBuddyAuto mode, an AI reviewer; not run by uspermissions.defaultMode: "auto" in ~/.codebuddy/settings.json
CodexAuto-reviewapprovals_reviewer = "auto_review" in ~/.codex/config.toml
Cursor (cursor-agent only)Auto-review, a server classifierapprovalMode: "auto-review" in ~/.cursor/cli-config.json
Devin CLI (windsurf)Smart mode, a fast model; not run by uspermission_mode: "smart" in ~/.config/devin/config.json (%APPDATA%\devin\config.json on Windows)
QoderAuto mode, a two-stage classifier; not run by usgeneral.defaultPermissionMode: "auto" in ~/.qoder/settings.json
VibeSmart-approve, a classifier; not run by usdefault_agent = "smart-approve" in ~/.vibe/config.toml
Qwen CodeAuto mode, an LLM classifier; not run by ustools.approvalMode: "auto" in ~/.qwen/settings.json
  • A home settings spec that sets the same key wins: permissions.default-mode where the tool maps it, otherwise the key under x-<tool>, such as x-codex.approvals_reviewer. A value set outside agnostic-ai stays as it is, and sync --global warns that the reviewer stays off.
  • Remove review and run sync --global again, and the key goes. approvals ask alone does not remove it.
  • The key sits in a user file, so it applies in every folder. A projects entry or AGNOSTIC_AI_APPROVALS=ask cannot turn the reviewer off for one folder or session; approvals status notes it there. A project sync never writes these keys.
  • Codex Auto-review answers only prompts Codex would show. With approval_policy = "never" there are none, so the reviewer never runs; approvals status says so when ~/.codex/config.toml sets it. It does not read a trusted project's .codex/config.toml, which can set either key too.
  • Qwen Code already runs auto mode in a trusted folder when tools.approvalMode is unset, so ask alone does not bring its prompts back; approvals status says so. Set x-qwen.tools.approvalMode: default in a home settings spec to see them. Qwen Code asks anyway in an untrusted folder and with --bare or --safe-mode, and a project's .qwen/settings.json can set the same key. Checked against the Qwen Code 0.28.0 package.
  • CodeBuddy takes defaultMode: "auto" only from user settings and the command line; a project file cannot grant it. It falls back to default when permissions.disableAutoMode is "disable" or auto mode is unavailable, and approvals status notes the first. sync --global writes ~/.codebuddy/settings.json even when CODEBUDDY_CONFIG_DIR moves CodeBuddy's settings elsewhere; status notes that too. Checked against the CodeBuddy 2.185.0 package and its permission-modes page.
  • cursor-agent reads cli-config.json from CURSOR_CONFIG_DIR, else $XDG_CONFIG_HOME/cursor, else ~/.cursor, and sync --global follows the same order. cursor-agent rewrites that file itself, so sync owns only approvalMode there. cursor-agent writes approvalMode: "allowlist" as its own default, so review replaces that value and gives the key back when review goes; any other value set by hand stays. x-cursor.approvalMode in a home settings spec wins and must be allowlist, unrestricted or auto-review, since cursor-agent sets a file with any other value aside and starts over. CURSOR_CONFIG_DIR and XDG_CONFIG_HOME must be absolute paths. Auto-review needs a classifier-capable model and team permission, or cursor-agent falls back to its allowlist. The Cursor desktop app keeps its own setting, and approvals status says to turn on Auto-review there. Checked against cursor-agent 2026.11.11, which kept the key when it rewrote the file.
  • Devin CLI reads permission_mode only from the user config, so a project cannot set it. --permission-mode replaces it for one launch. Devin's docs also show DEVIN_PERMISSION_MODE=smart devin, and approvals status notes a set variable; that the variable wins over the file is our reading, not stated. A null value means normal, so review replaces it. Your organization's policy can remove Smart mode, and sandbox sessions do not offer it; the docs do not say what Devin does with the key then. approvals status reports dangerous as approving every call. Sync owns only permission_mode in that file; x-windsurf.permission_mode in a home settings spec wins and must be normal, accept-edits, smart or dangerous. Your own deny and ask rules still come first. From the Devin CLI configuration and permissions pages; Devin CLI is not installed here.
  • Qoder reads general.defaultPermissionMode from every settings layer, so a project's .qoder/settings.json can set it too, and an untrusted folder starts in default, as does a session with no classifier model available. An older tools.approvalMode key counts as the user's mode, so review leaves it. Checked against the Qoder CLI 1.1.93 package.
  • Vibe takes a default_agent that a person sets as an opt-in, so smart-approve starts even before the mode reaches Vibe's mode picker. An organization can still exclude it, and a trusted project's .vibe/config.toml, VIBE_DEFAULT_AGENT, an admin config, or --agent replaces the user file's value. In the user file, x-vibe cannot set default_agent, so only a hand-set value wins; Vibe can also write default_agent = "ask" there itself when disabled_agents lists accept-edits. With bypass_tool_permissions = true, Vibe approves every call and the classifier never runs; approvals status notes it and a set VIBE_DEFAULT_AGENT. Checked against the mistral-vibe 3.1.3 package.
  • A running session keeps its mode until you switch it in the tool (Shift+Tab in Claude Code). A new session reads the new settings.
  • agnostic-ai logs no decision the reviewer makes. Each tool keeps its own record, and each review costs the tool's own model calls.

Run approvals status to see the mode and route per tool, and approvals ask to show every prompt again.

MCP servers

MCP specs in the home's mcps/ install servers in each supported tool's user MCP file, using the same format as project servers:

TargetFileWhere
claude~/.claude.json ($CLAUDE_CONFIG_DIR/.claude.json when set)top-level mcpServers.<name>
codex~/.codex/config.toml[mcp_servers.<name>] table
codebuddy~/.codebuddy/.mcp.jsonmcpServers.<name>; legacy precedence and native limits
vibe~/.vibe/config.toml[[mcp_servers]], keyed by name
kimi~/.kimi-code/mcp.jsonmcpServers.<name>, retaining disabled servers
cursor~/.cursor/mcp.jsonmcpServers.<name>
copilot~/.copilot/mcp-config.jsonmcpServers.<name>, with tools: ["*"] when the spec sets none
gemini~/.gemini/settings.jsonmcpServers.<name>
qoder~/.qoder/settings.jsonmcpServers.<name>
warp~/.warp/.mcp.json (Warp app only; the Agent CLI reads its own file)mcpServers.<name>: {command, args, env, working_directory} or {url, headers}
augment~/.augment/settings.jsonmcpServers.<name>
antigravity~/.gemini/config/mcp_config.jsonmcpServers.<name>: {command, args, env, cwd} or {serverUrl, headers}, with optional disabled

The Tools page in serve --global lists servers that have no home spec and were not written by sync. Each row has the import --global <target> command that creates its spec (serve).

OpenHands has no user MCP file sync can write. Sync prints a note that points at Agent Canvas Customize > MCP Servers. It also removes the servers an earlier sync recorded as its own in ~/.openhands/mcp.json (OpenHands MCP).

Details: how sync edits MCP servers
  • A hand-written server that means the same as the spec is adopted.
  • A different server with the same name stops the run. --backup overwrites it.
  • A server sync wrote is removed when its spec goes. Servers you add under other names stay.
  • In ~/.claude.json, sync edits only its own mcpServers entries and creates a missing file at 0600.
  • If ~/.claude.json changed after sync read it, sync stops without writing. Run it again.
  • A spec with disabled: true stays out of the Augment, Claude, Cursor, Copilot, and Warp user files, where a listed server is live in every project.
  • Antigravity keeps disabled servers in its user file with disabled: true.
  • When CLAUDE_CONFIG_DIR or another root variable moves a file, the next sync removes its entries from the old one.

Native discovery and shared paths

Discovery limits

  • Copilot's path is for its CLI.
  • OpenHands discovery applies to local conversations.
  • Devin custom profiles are experimental.
  • Trae needs Settings > Beta > Subagents > Enable Subagents Directory.

Configuration roots

  • CLAUDE_CONFIG_DIR, CODEX_HOME, COPILOT_HOME, CLINE_DIR, QODER_CONFIG_DIR, KIRO_HOME, JUNIE_HOME, and KIMI_CODE_HOME replace the matching ~/.<tool>/ folder.
  • GEMINI_CLI_HOME is the parent of .gemini/.
  • Everything under that folder moves together. Paths outside it, such as ~/.agents/skills/, stay put.
  • Use absolute paths.

Shared paths

Shared paths are written once:

PathTools
~/.agents/skills/codex, windsurf, amp, zed, warp, goose, openhands
~/.agents/agents/goose, openhands
~/.agents/AGENTS.mdcline, warp
~/.gemini/GEMINI.mdgemini, antigravity

If two tools would write different content to one path, the run stops before writing and names the path.

Per-tool notes

Copilot agent effort

  • Copilot CLI documents a native profile reasoningEffort field, available through x-copilot.reasoningEffort (agent fields).
  • Portable agent effort is not mapped to that CLI field.
  • A portable agent effort of low, medium, high, or xhigh goes to subagents.agents.<name>.effortLevel in ~/.copilot/settings.json. Other values raise a coverage note.
  • Sync stops when the file has comments, since a rewrite drops them. Rerun with --backup to rewrite it and keep settings.json.bak.
  • Sync also stops when a hand-set effortLevel has a different value.

Cursor

  • Cursor does not load the home-level AGENTS.md.
  • Global sync installs a sessionStart hook and a script under ~/.cursor/hooks/ that pass the instructions to Cursor. The script is POSIX shell on macOS and Linux, PowerShell on Windows.
  • It needs neither agnostic-ai, Python, nor jq, and it adds context without enforcing policy.
  • Existing sessionStart entries stay.
  • Settings: global sync writes only approvalMode to cursor-agent's cli-config.json, for approvals. Other x-cursor keys give a note.

OpenCode

  • disable-model-invocation stays in project and global skills. V2 honors it; V1 does not (OpenCode skills covers native metadata precedence).
  • V1 reads ~/.claude/CLAUDE.md only when ~/.config/opencode/AGENTS.md does not exist.
  • The current V2 migration guide documents a CLAUDE.md fallback in each directory without AGENTS.md, plus ~/.claude/CLAUDE.md when the global OpenCode AGENTS.md is absent.
  • V2 chooses per directory, so an AGENTS.md at the project root does not hide a nested CLAUDE.md.
  • In V2 2.3.1 and later, either OPENCODE_DISABLE_CLAUDE_CODE_PROMPT or OPENCODE_DISABLE_CLAUDE_CODE disables every CLAUDE.md fallback, both project and global (2.3.1 release notes).
  • Put OpenCode guidance in its AGENTS.md when syncing both tools.

Devin CLI and VS Code Copilot

  • Both read ~/.claude/ by default. Syncing claude plus windsurf or copilot delivers the same instructions through two paths.

Targets left out

No user-level instructions file:

  • Aider reaches a home instructions file only through a read: entry in ~/.aider.conf.yml.
  • Continue's one documented home setting is the rules: list in the config.yaml it rewrites itself.
  • Jules documents nothing at user level.
  • Ellipsis reads only files in the repository.

Hooks reach these tools at user level: Claude Code, CodeBuddy Code, Codex, Gemini, Qoder, Cursor, Augment, Kiro CLI V3, Copilot, Cline SDK/CLI, Grok Build, Kimi Code CLI, Qwen Code, Mistral Vibe, and Goose. The others have no user-level hook output:

  • Factory, Devin CLI, Antigravity, and Crush use hook formats that global sync does not write.
  • Junie's hooks are Early Access.
  • Amp, OpenCode, and Kilo take hooks only as TypeScript plugins. Sync writes them at project level only, for OpenCode (.opencode/plugins/) and Kilo (.kilo/plugin/).
  • OpenCode defaults to V1. A hook selects V2 with x-opencode.plugin-api: v2 on macOS and Linux.
  • On Windows, V2 refuses matching callbacks, so keep the default V1 hooks.

Global output

A dash means global sync writes nothing for that kind.

TargetInstructionsRulesHooksSkillsAgentsSession retention
claude~/.claude/CLAUDE.mdinlined~/.claude/settings.json~/.claude/skills/<name>/~/.claude/agents/<name>.mddays
codebuddy~/.codebuddy/CODEBUDDY.mdinlined~/.codebuddy/settings.json~/.codebuddy/skills/<name>/~/.codebuddy/agents/<name>.mddays
cursor~/.cursor/AGENTS.md (bridged)inlined~/.cursor/hooks.json~/.cursor/skills/<name>/~/.cursor/agents/<name>.md-
codex~/.codex/AGENTS.mdinlined~/.codex/hooks.json~/.agents/skills/<name>/~/.codex/agents/<name>.toml-
gemini~/.gemini/GEMINI.mdinlined~/.gemini/settings.json~/.gemini/skills/<name>/~/.gemini/agents/<name>.mddays
kimi~/.kimi-code/AGENTS.mdinlined~/.kimi-code/config.toml~/.kimi-code/skills/<name>/~/.kimi-code/agents/<name>.md-
vibe~/.vibe/AGENTS.mdinlined~/.vibe/hooks.toml~/.vibe/skills/<name>/~/.vibe/agents/<name>.toml-
qoder~/.qoder/AGENTS.mdinlined~/.qoder/settings.json~/.qoder/skills/<name>/~/.qoder/agents/<name>.mddays
copilot~/.copilot/copilot-instructions.mdinlined~/.copilot/hooks/agnostic-ai.json~/.copilot/skills/<name>/~/.copilot/agents/<name>.agent.md-
cline~/.agents/AGENTS.mdinlined~/.cline/hooks/<event>~/.cline/skills/<name>/~/.cline/agents/<name>.yml-
windsurf~/.config/devin/AGENTS.mdinlined-~/.agents/skills/<name>/~/.config/devin/agents/<name>.md-
amp~/.config/amp/AGENTS.mdinlined-~/.agents/skills/<name>/--
zed~/.config/zed/AGENTS.md; Windows: %APPDATA%\Zed\AGENTS.mdinlined-~/.agents/skills/<name>/--
warp~/.agents/AGENTS.mdinlined-~/.agents/skills/<name>/--
opencode~/.config/opencode/AGENTS.mdinlined-~/.config/opencode/skills/<name>/~/.config/opencode/agents/<name>.md-
antigravity~/.gemini/GEMINI.mdinlined-~/.gemini/config/skills/<name>/~/.gemini/config/agents/<name>/agent.md-
junie~/.junie/AGENTS.mdinlined-~/.junie/skills/<name>/~/.junie/agents/<name>.md-
kiro~/.kiro/steering/AGENTS.mdinlined~/.kiro/hooks/<name>.json~/.kiro/skills/<name>/~/.kiro/agents/<name>.md-
crush~/.config/crush/CRUSH.mdinlined-~/.config/crush/skills/<name>/--
factory~/.factory/AGENTS.mdinlined-~/.factory/skills/<name>/~/.factory/droids/<name>.md-
kilo~/.config/kilo/AGENTS.mdinlined-~/.kilo/skills/<name>/~/.config/kilo/agents/<name>.md-
goose~/.config/goose/.goosehintsinlined~/.agents/plugins/agnostic-ai-global/hooks/hooks.json~/.agents/skills/<name>/~/.agents/agents/<name>.md-
openhands---~/.agents/skills/<name>/~/.agents/agents/<name>.md-
trae---~/.trae/skills/<name>/~/.trae-cn/agents/<name>.md-
grok~/.grok/AGENTS.mdinlined~/.grok/hooks/<name>.json~/.grok/skills/<name>/~/.grok/agents/<name>.md-
qwen~/.qwen/QWEN.mdinlined~/.qwen/settings.json~/.qwen/skills/<name>/~/.qwen/agents/<name>.md-
augment-~/.augment/rules/<name>.md~/.augment/settings.json~/.augment/skills/<name>/~/.augment/agents/<name>.md-
  • MCP servers and settings have their own tables in the reference. Agents use the same formats as project agents (the agent example).
  • Rules are inlined into the instructions file, in the same ## Rules block as the shared body.
  • Augment has no user-level instructions file for the CLI, so its ~/.augment/rules/ entries are always on.
  • Existing instruction files keep their personal text and file permissions.
  • Sync stops if a file changes while it is preparing the update.

Path notes:

  • Paths marked ~/.config/ follow XDG_CONFIG_HOME when set, except Zed on macOS.
  • Zed on Linux and FreeBSD prefers FLATPAK_XDG_CONFIG_HOME (native instruction roots).
  • Devin agents use ~/.config/devin/agents/ on Linux and macOS and %APPDATA%\devin\agents\ on Windows.