coder-cli

name	coder-cli
description	Expert knowledge of the current coder CLI and coder-node architecture. Use for questions about implemented commands, memory and skill flows, auth, release flow, and .coder state handling. Distinguishes current commands from roadmap-only ideas.

Skill: coder CLI Architecture

Overview

coder is currently a Go CLI focused on:

• memory retrieval and storage
• skill retrieval and ingestion
• local session checkpointing
• project state helpers around .coder/STATE.md
• install, auth, update, and release-adjacent maintenance

It is not currently a built-in LLM chat or workflow runner. Any document that describes coder chat, coder review, coder debug, or phase-execution commands should be treated as roadmap unless those commands are registered in cmd/coder/main.go.

coder-node is the backend service. It owns memory, skills, auth, and storage. The CLI is a thin client except for some local session and .coder/ state commands.

---

Current Top-Level Commands

The current binary registers only these commands:

install
update
list
version
check-update
self-update
login
token
skill
memory
remove
session
progress
next
milestone

Always anchor command explanations to cmd/coder/main.go.

---

Command Groups

Setup And Maintenance

Command	Purpose
coder install	Install project or global profiles
coder update	Refresh installed profile files
coder list	Show available profiles
coder remove global	Remove globally installed coder-managed files
coder version	Print CLI build information
coder check-update	Check GitHub for a newer release
coder self-update	Replace the local binary with the latest release

Connection And Auth

Command	Purpose
coder login	Configure connection to coder-node
coder token show	Show current auth identity/token summary
coder token rotate	Rotate saved access token

Knowledge Commands

Command	Purpose
coder skill search	Search ingested skills
coder skill ingest	Ingest local, embedded, or GitHub skills
coder skill list	List stored skills
coder skill info	Show a skill and its chunks
coder skill delete	Remove a skill
coder skill cache	Pull/list/clear cached skill files
coder skill index	Generate local skills_index.json
coder memory store	Store memory with lifecycle metadata
coder memory search	Search lifecycle-aware memory
coder memory recall	Re-recall memory and compute keep/add/drop decisions
coder memory active	Inspect the current active memory recall state
coder memory verify	Refresh verification metadata
coder memory supersede	Link old and replacement versions
coder memory audit	Report lifecycle issues
coder memory list	List recent memories
coder memory delete	Delete one memory
coder memory compact	Compact and optionally re-vector memory

Local Context And Project State

Command	Purpose
coder session save/resume/list/show/delete/export	Manage local working sessions
coder progress	Read .coder/STATE.md and .coder/ROADMAP.md
coder next	Suggest the next workflow command from project state
coder milestone	Audit, complete, archive, or advance a phase

---

What Is Roadmap Only

These commands are frequently referenced in older docs but are not currently wired in the binary:

chat
review
debug
plan
qa
workflow
new-project
map-codebase
discuss-phase
plan-phase
execute-phase
ship
todo
stats
health
note
do

When explaining them:

• describe them as roadmap or historical design only
• do not imply they are callable today
• prefer the current memory, skill, session, progress, next, and milestone commands instead

---

Runtime Architecture

CLI side

• cmd/coder/main.go: top-level dispatch
• cmd/coder/config.go: loads ~/.coder/config.json
• cmd/coder/cmd_skill.go: skill command handlers
• cmd/coder/cmd_memory.go: memory command handlers
• cmd/coder/cmd_session.go: local session management
• cmd/coder/cmd_progress.go: .coder/ state readers
• cmd/coder/cmd_milestone.go: phase lifecycle helpers

Server side

• cmd/coder-node/main.go: wires dependencies and starts transports
• internal/usecase/skill: ingest and search logic
• internal/usecase/memory: store, search, lifecycle, compact
• internal/usecase/auth: registration, token, activity
• internal/infra/postgres: storage for memory, skill, auth
• internal/transport/grpc and internal/transport/http: remote APIs

---

Important Flows

Skill flow

coder skill search does not parse local files itself. It calls coder-node, which searches already ingested skill chunks.

coder skill ingest:

1. reads skills from local FS, embedded FS, or GitHub
2. sends SKILL.md plus rule files to the server
3. server parses sections, optionally embeds them, and stores chunks

Memory flow

coder memory store/search/recall/verify/supersede/audit are lifecycle-aware and operate through the server unless local Postgres mode is configured. coder memory recall now uses a shared recall contract at the usecase and transport layers rather than a CLI-only decision loop.

coder memory search and coder memory recall also refresh the local .coder/active-memory.json snapshot so the latest recalled context can be inspected with coder memory active. Both commands also update .coder/context-state.json so active skills and memory can be recovered together.

Session flow

coder session is local-only. It writes to:

.coder/session.md
.coder/sessions/*.md

Project-state flow

coder progress, coder next, and coder milestone read or mutate:

.coder/STATE.md
.coder/ROADMAP.md
.coder/phases/
.coder/archive/

coder next may recommend roadmap commands that are not implemented yet. It reflects intended workflow state, not guaranteed command availability.

---

Guidance For Agents

When asked about coder commands:

1. check cmd/coder/main.go first
2. prefer implemented command behavior over old docs
3. call out explicitly when a document is describing roadmap rather than current behavior
4. do not tell users that coder itself runs LLM chat flows unless the code actually wires those commands