@athenode/cli

The official command-line interface for Athenode — manage your project's specifications, track their status, and record implementation results, all without leaving your terminal.

Install

npm install -g @athenode/cli

Getting started

Run the interactive setup once in your project:

athenode init

This walks you through connecting the CLI to your Athenode project and stores a local, private config file so you don't have to re-authenticate every time.

Already have a project token? Add it directly:

athenode auth add-token <token>

What you can do

• Browse and search specifications — list, filter, and full-text search the specifications in your project.
• Create and edit specifications — draft new specifications and update their title, summary, content, or status right from the command line.
• Track implementation plans — attach or read back the implementation plan for any specification.
• Record execution results — log what changed when a specification is implemented.
• Manage clarifying questions — add and answer the questions attached to a specification during planning.

Run athenode --help or athenode <command> --help at any time to see everything available. Every data-returning command always prints raw JSON, so output is easy to pipe into another tool.

Environment field mapping

When you export agents/commands to a coding environment (Claude Code, Codex, or Cursor), each metadata field is mapped per-environment rather than copied verbatim: some rename to that environment's own field name, some only pass through under specific values, and some have no equivalent at all and are dropped. The tables below are the current, field-by-field source of truth for that mapping.

Unless noted otherwise, "dropped" for Cursor means the field isn't a native top-level frontmatter field there — it still survives, nested under a catch-all metadata: key, rather than being lost. "Dropped" for Claude Code/Codex means the field is genuinely omitted from the exported file.

Agent fields (.claude/agents/*.md / .codex/agents/*.toml / .cursor/agents/*.md)

| Field (our catalog) | Claude Code | Codex | Cursor | | ---------------------------- | ----------------- | -------------------------------------------------- | -------------------------------------------------------------- | | description | description | description | description | | tools | tools | dropped — no allow-list field in Codex's custom-agent schema | dropped — derives readonly: true when none of Write/Edit/Bash are granted | | model | model | model, only when the value is exactly inherit | model, only when the value is exactly inherit | | color | color | dropped — no such field in Codex | dropped — no such field in Cursor's subagent schema | | maxTurns | maxTurns | dropped — not a documented custom-agent field | dropped — no such field | | mcpServers | mcpServers | mcp_servers, per-entry shape transformed¹ | dropped — no per-agent MCP config in Cursor's schema | | hooks | hooks | dropped — schema compatibility with Codex's hooks unconfirmed | dropped — no such field | | disallowedTools | disallowedTools | dropped — no such field | dropped — no such field | | permissionMode | permissionMode | dropped — no per-agent equivalent (approval_policy is session-wide) | dropped — no such field | | skills | skills | dropped — Codex's skills.config is path/enabled pairs, not a name list | dropped — no such field | | memory | memory | dropped — Codex's memories is session-wide config, not this concept | dropped — no such field | | background | background | dropped — no such field | is_background (renamed) | | effort | effort | model_reasoning_effort (renamed; max → xhigh) | dropped — no such field | | isolation | isolation | dropped — no git-worktree concept in Codex custom agents | dropped — no such field | | initialPrompt | initialPrompt | dropped — no equivalent | dropped — no such field |

¹ mcpServers → Codex mcp_servers: bare-string "reference by name" entries are dropped (no such concept in Codex); stdio configs keep command/args/env as-is; http/sse/ws configs keep url, rename headers → http_headers, rename timeout → startup_timeout_sec; headersHelper/alwaysLoad/oauth have no Codex equivalent and are dropped.

Two Codex-only fields have no backend catalog row yet (nothing in the UI sets them): sandbox_mode and, historically, model_reasoning_effort before effort was wired to it above.

Command/skill fields (.claude/commands/*.md / .codex/skills/<name>/SKILL.md / .cursor/skills/<name>/SKILL.md)

Cursor commands (.cursor/commands/*.md) support no frontmatter at all, so our command catalog is exported as a Cursor skill instead (.cursor/skills/<name>/SKILL.md), the same folder-per-skill shape Codex already uses.

| Field (our catalog) | Claude Code | Codex | Cursor (skill) | | ---------------------------- | ----------------- | -------------------------------------------------- | ------------------------------------------ | | description | description | description | description | | allowed-tools | allowed-tools | allowed-tools² | dropped — not one of Cursor's 5 skill fields | | argument-hint | argument-hint | argument-hint² | dropped | | model | model | model, only when the value is exactly inherit | dropped | | disable-model-invocation | disable-model-invocation | dropped — Codex's equivalent lives in a separate agents/openai.yaml file, not SKILL.md | disable-model-invocation | | when_to_use | when_to_use | when_to_use² | dropped | | arguments | arguments | arguments² | dropped | | user-invocable | user-invocable | dropped — no equivalent | dropped | | disallowed-tools | disallowed-tools | disallowed-tools² | dropped | | effort | effort | model_reasoning_effort (renamed; max → xhigh) | dropped | | context | context | dropped — no equivalent | dropped | | agent | agent | dropped — no equivalent (only meaningful with context: fork, itself dropped) | dropped | | hooks | hooks | dropped — schema compatibility unconfirmed | dropped | | paths | paths | paths² | paths — one of Cursor's 5 documented skill fields | | shell | shell | dropped — no equivalent | dropped |

² Codex's own docs only explicitly document name/description in SKILL.md itself. These fields are passed through best-effort on the (unconfirmed) assumption that Codex follows the same shared agentskills.io convention Claude Code does — an unrecognized field is simply ignored there, not an error.

Two Codex-only skill fields have no backend catalog row yet: license and compatibility.