████████████████████████████████████████████████████████████████████████████████████████████ █▌ ▐█ █▌ ▐█ █▌ ▐█ █▌ I8 ▐█ █▌ I8 ▐█ █▌ 88888888 ▐█ █▌ I8 ▐█ █▌ ,gggg,gg ,gggg,gg ,ggg, ,ggg,,ggg, I8 ,gggggg, gg gg ▐█ █▌ dP" "Y8I dP" "Y8I i8" "8i ,8" "8P" "8, I8 dP""""8I I8 8I ▐█ █▌ i8' ,8I i8' ,8I I8, ,8I I8 8I 8I ,I8, ,8' 8I I8, ,8I ▐█ █▌ ,d8, ,d8b,,d8, ,d8I 'YbadP' ,dP 8I Yb,,d88b,,dP Y8,,d8b, ,d8I ▐█ █▌ P"Y8888P"'Y8P"Y8888P"888888P"Y8888P' 8I 'Y88P""Y88P 'Y8P""Y88P"888 ▐█ █▌ ,d8I' ,d8I' ▐█ █▌ ,dP'8I ,dP'8I ▐█ █▌ ,8" 8I ,8" 8I ▐█ █▌ I8 8I I8 8I ▐█ █▌ '8, ,8I '8, ,8I ▐█ █▌ 'Y8P" 'Y8P" ▐█ █▌ ▐█ █▌ ▐█ █▌ ▐█ ████████████████████████████████████████████████████████████████████████████████████████████
One source of truth for prompts. Format-aware sync to every agent CLI.
agentry is a terminal UI application that manages prompts for 11 agent CLIs from a single unified source. Instead of maintaining separate prompt files scattered across .claude/CLAUDE.md, .continue/, .codex/AGENTS.md, and so on, agentry lets you write prompts once and sync them everywhere -- automatically converting between each agent's native file format.
| Agent | CLI Binary | Config Dir | Prompt File | Format |
|---|---|---|---|---|
| Claude Code | claude |
.claude |
CLAUDE.md |
Plain Markdown |
| Continue | continue |
.continue |
prompts/ |
XML Tag + MD |
| Gemini CLI | gemini |
.gemini |
GEMINI.md |
Plain Markdown |
| Codex | codex |
.codex |
AGENTS.md |
Plain Markdown |
| Amp | amp |
.amp |
AGENTS.md |
Plain Markdown |
| OpenCode | opencode |
.opencode |
AGENTS.md |
Frontmatter + MD |
| Firebender | firebender |
.firebender |
rules/ (.mdc) |
MDC |
| OpenClaw | openclaw |
.openclaw |
AGENTS.md |
Plain Markdown |
| DeepAgents | deepagents |
.deepagents |
AGENTS.md |
Plain Markdown |
| Antigravity | antigravity |
.antigravity |
SKILL.md |
Frontmatter + MD |
| Warp | warp-cli |
.warp |
AGENTS.md |
Frontmatter + MD |
- One source of truth -- Write prompts once in
~/.agents/prompts/, sync to all agents - 5 format converters -- Automatic conversion between PlainMD, FrontmatterMD, MDC, XmlTagMD, and LobsterYAML
- Agent detection -- Parallel discovery of installed agents, versions, config dirs, and skills
- Sync engine -- Plan, preview, and execute prompt distribution across agents (copy, symlink, or skip)
- Skill hub -- Browse, install, and update community skills from Git repositories
- OpenClaw workspace -- Discover and manage OpenClaw
.lobsterworkflows - ACP orchestration -- Agent Communication Protocol for multi-agent routing
- Agent audit -- Health scores (0-100) from 24 diagnostic checks, with
--fixmode gated by a fail-closed shell allowlist and a self-evolving feedback loop - Keyboard-first TUI -- Five-tab dashboard with a persistent keymap bar; editing opens your
$EDITOR(nvim/vim/vi)
agentry understands five distinct prompt formats and can convert between them:
| Converter | Format | Used By |
|---|---|---|
PlainMarkdownConverter |
Plain Markdown | Claude Code, Gemini CLI, Codex, Amp, OpenClaw, DeepAgents |
FrontmatterMdConverter |
YAML frontmatter + Markdown body | OpenCode, Antigravity, Warp |
MdcConverter |
MDC (frontmatter with globs) | Firebender |
XmlTagMdConverter |
Frontmatter + XML tag wrappers | Continue (<expertise>, <base_rules>) |
LobsterYamlConverter |
Lobster YAML workflows | OpenClaw |
Each converter implements the FormatConverter trait with parse() and serialize() methods, enabling lossless round-tripping and cross-format conversion through the UnifiedPrompt intermediate representation.
The terminal UI features an intro animation with ASCII art and a progress bar while agents are detected in parallel, followed by a 5-tab dashboard:
| Tab | Key | Description |
|---|---|---|
| Agents | 1 |
Browse detected agents, versions, skills, and config paths; when the openclaw agent is selected, the detail panel shows a Workspaces section for managing OpenClaw workspace docs and .lobster workflows |
| Prompts | 2 |
Manage prompts; editing opens your $EDITOR (nvim/vim/vi) |
| Skills | 3 |
Browse and install skill hub entries |
| Sync | 4 |
Plan and execute prompt sync (plan auto-loads on tab entry) |
| Audit | 5 |
Run agent health audits and review findings (auto-runs on first entry) |
Global keys (all tabs):
| Key | Action |
|---|---|
q |
Quit |
? |
Toggle help overlay |
Tab / Shift+Tab |
Switch tabs forward/backward |
1 -- 5 |
Jump directly to tab (resets list selection) |
j / k / Arrow keys |
Navigate list |
A persistent keymap bar is always visible at the bottom of the TUI, showing the current tab's active keys. Per-tab keys:
| Tab | Key | Action |
|---|---|---|
| Agents | Enter |
Install agent (or edit OpenClaw workspace doc when openclaw is selected) |
| Agents | u |
Update agent |
| Agents | r |
Remove agent |
| Agents | v |
List versions (Enter installs the selected version) |
| Agents | ← / → |
Cycle install method |
| Agents (openclaw) | c |
Create workspace (runs openclaw setup) |
| Agents (openclaw) | a |
Add agent (shows the openclaw agents add command) |
| Prompts | Enter / e |
Edit prompt in $EDITOR |
| Prompts | n |
New prompt |
| Prompts | d |
Delete prompt |
| Skills | Enter / i |
Install skill |
| Skills | u |
Update skill |
| Skills | r |
Remove skill |
| Skills | g |
Open skill GitHub repository |
| Sync | s |
Execute selected mapping |
| Sync | S |
Execute all non-skip mappings |
| Audit | r |
Re-run audit |
| Audit | f |
Cycle severity filter (All / Critical / Warning / Info / Suggestion) |
| Audit | Enter |
Open finding file / show remediation |
Run agentry without arguments to launch the TUI, or use subcommands for non-interactive mode:
# Detect installed agents
agentry detect
# Sync a specific prompt to all agents
agentry sync --prompt software-architect
# Sync all prompts (dry run first)
agentry sync --all --dry-run
agentry sync --all
# List installed and available skills
agentry skills list
# Install a skill
agentry skills install deploy-to-vercel
# Update all installed skills
agentry skills update
# Remove a skill
agentry skills remove deploy-to-vercel
# List discovered prompts
agentry prompts list
# Browse OpenClaw workspaces
agentry openclaw workspaces
# Run a full agent health audit
agentry audit
# Audit a single agent
agentry audit --agent claude-code
# Machine-readable report (CI/agentic gate)
agentry audit --json
# Apply auto-fixes
agentry audit --fix --yesagentry is organized as a Cargo workspace with 8 crates:
agentry/
├── Cargo.toml # Workspace root
├── crates/
│ ├── agentry-core/ # Core types, models, and format converters
│ ├── agentry-agents/ # Agent specs, detection, and registry
│ ├── agentry-sync/ # Sync planner and executor
│ ├── agentry-skills/ # Skill hub, lockfile, and installer
│ ├── agentry-openclaw/ # OpenClaw workspace discovery and docs
│ ├── agentry-acp/ # ACP protocol and message router
│ ├── agentry-audit/ # Agent health audit engine and remediation
│ └── agentry-tui/ # Terminal UI (binary crate: `agentry`)
└── tests/
└── integration/ # Integration tests
agentry-tui
├── agentry-core
├── agentry-agents
├── agentry-sync
├── agentry-skills
├── agentry-openclaw
└── agentry-audit
agentry-agents ──> agentry-core
agentry-sync ──> agentry-core, agentry-agents
agentry-skills ──> agentry-core
agentry-openclaw ──> agentry-core, agentry-agents
agentry-acp ──> agentry-core, agentry-agents, agentry-skills
agentry-audit ──> agentry-core, agentry-agents, agentry-sync, agentry-skills, agentry-openclaw, agentry-acp
UnifiedPrompt-- Canonical representation of a prompt from any format, with frontmatter, body, XML tags, scope, and source formatAgentSpec-- Static specification of a known agent (binary name, config dir, prompt filename, format)DetectedAgent-- Runtime detection result (installed status, version, skills, symlinks)SyncPlan/SyncMapping-- Planned sync actions with status tracking (UpToDate, Missing, Outdated, Conflict)SkillEntry-- A skill from the hub with source repo and install stateAppConfig-- User configuration at~/.agents/agentry.toml
The quickest way to install. The installer downloads the right build for your platform and places the binary in your Cargo home bin directory ($CARGO_HOME/bin):
curl -LsSf https://github.com/AndlerRL/agentry/releases/latest/download/agentry-tui-installer.sh | shOnce the crate is published to crates.io:
cargo install agentry-tuiNote: This requires the
agentry-tuicrate to be published to crates.io. To install directly from the repository in the meantime:cargo install --git https://github.com/AndlerRL/agentry --bin agentry
Download the archive for your platform from GitHub Releases. Each archive is also available directly at https://github.com/AndlerRL/agentry/releases/latest/download/<archive>:
| Platform | Archive |
|---|---|
| macOS (Apple Silicon) | agentry-tui-aarch64-apple-darwin.tar.xz |
| macOS (Intel) | agentry-tui-x86_64-apple-darwin.tar.xz |
| Linux (x86_64, glibc) | agentry-tui-x86_64-unknown-linux-gnu.tar.xz |
| Linux (ARM64, glibc) | agentry-tui-aarch64-unknown-linux-gnu.tar.xz |
| Linux (x86_64, static/musl) | agentry-tui-x86_64-unknown-linux-musl.tar.xz |
Each archive extracts to a versioned directory (agentry-tui-v<version>-<target>/) containing the agentry binary:
# Example: macOS (Apple Silicon)
curl -LO https://github.com/AndlerRL/agentry/releases/latest/download/agentry-tui-aarch64-apple-darwin.tar.xz
tar xf agentry-tui-aarch64-apple-darwin.tar.xz
cd agentry-tui-v*/
mv agentry ~/.local/bin/macOS Gatekeeper: If you see "cannot be opened because it is from an unidentified developer", run:
xattr -d com.apple.quarantine ~/.local/bin/agentry
git clone https://github.com/AndlerRL/agentry.git
cd agentry
cargo build --release
# Binary at target/release/agentry
cp target/release/agentry /usr/local/bin/Linux prerequisites: OpenSSL is vendored and built from source, so no system OpenSSL packages are needed. The vendored OpenSSL build does require
perlandmake:# Debian/Ubuntu sudo apt install perl make # Fedora dnf install perl makeWindows is not yet supported.
agentry reads configuration from ~/.agents/agentry.toml:
[project_dirs]
dirs = ["~/Development"]
[sync_defaults]
dry_run = true
conflict_strategy = "overwrite" # overwrite | keep | merge | diff
[extra_skill_sources]
repos = []| Phase | Status | Scope |
|---|---|---|
| Phase 1 | Complete | Core scaffolding, agent detection, TUI shell with intro animation and dashboard |
| Phase 2 | Complete | Prompt editor, unified prompt management, format conversion |
| Phase 3 | Complete | Sync engine -- planner, executor, dry-run, conflict resolution, project-level sync |
| Phase 4 | Complete | Skill hub -- browse, install, update, lockfile management, symlink creation |
| Phase 5 | Complete | OpenClaw workspace discovery and .lobster workflow management |
| Phase 6 | Complete | ACP protocol -- multi-agent orchestration, capability routing, .lobster generation |
| Phase 7 | Complete | Polish, testing, error handling, docs, CI improvements, release |
- Rust -- Language
- ratatui -- Terminal UI framework
- crossterm -- Cross-platform terminal manipulation
- tokio -- Async runtime
- serde -- Serialization framework (JSON, YAML, TOML)
- clap -- CLI argument parsing
- git2 -- Git operations for skill hub
This project is licensed under the Apache License 2.0.