@agentteams/cli

A CLI for working with the AgentTeams API from your terminal. It supports convention sync and plan/comment/report management.

Installation

npm install -g @agentteams/cli

Short alias (agt)

agentteams is also installed as agt. The canonical name keeps working — the alias is an addition, not a replacement.

agt --help    # same as: agentteams --help

• The alias symlink is created at install time. If you installed an earlier version, rerun npm install -g @agentteams/cli to get it.
• Use the alias only when you type commands yourself. Scripts, MCP server registration, and agent guides should keep using agentteams, which exists in every environment.
• Help output echoes the alias only where npm installs bins as symlinks (POSIX). Windows .cmd/.ps1 shims and pnpm global installs are shell wrappers, so agt --help still prints Usage: agentteams ... there. Running the alias itself works everywhere.

Quick Start

1. Initialize

agentteams init

The init command:

• Opens a browser for OAuth authentication
• Creates .agentteams/config.json
• Stores teamId, projectId, apiKey, and a non-default apiUrl when needed
• Saves the convention template to .agentteams/convention.md
• Syncs convention files into .agentteams/<category>/*.md
• Adds .agentteams to .gitignore — it may contain API keys and must not be committed
• Creates the agent entry point file for each AI client detected in the folder, leaving any file that already exists untouched

Choosing a login method

The browser callback is always the default. Nothing is auto-detected — you pick a method by what you type.

| Situation | What to run | How it authenticates | | ------------------------------------------------------------ | ----------------------------------------------------------------------- | ----------------------------------------------------------------------------- | | Local machine with a browser (default) | agentteams init / agentteams auth login | The browser calls back to a temporary port on this machine | | Remote or headless shell (SSH, container, WSL, no DISPLAY) | agentteams init --device-auth / agentteams auth login --device-auth | A short code you approve in a browser on any other device — no local port | | CI and unattended automation | AGENTTEAMS_API_KEY, or a project service token | No human approval step at all |

The default path opens a local port and waits for the browser to call back. Over SSH the browser runs on your laptop, so the callback reaches your laptop's localhost and never the remote shell — that is the case --device-auth exists for. Device authorization is not a CI mechanism: it always requires a person to approve the request.

--device-auth

# Log in only
agentteams auth login --device-auth

# Initialize a project (team/project/agent are chosen on the approval screen)
agentteams init --device-auth

The terminal prints a verification URL and an 8-character code. Open the URL on any device, sign in, check that the code on screen matches the one in your terminal, and approve. The CLI stores the login and continues.

To avoid typing the flag on every command on a remote machine, declare it once for that machine:

# Persist it in ~/.agentteams/config.json (never in a project config)
agentteams auth login --device-auth --set-default

# Or per shell / per service unit
export AGENTTEAMS_DEVICE_AUTH=1

agentteams auth status reports whether this machine defaults to device authorization. To turn it back off, remove "deviceAuth" from ~/.agentteams/config.json and unset AGENTTEAMS_DEVICE_AUTH.

Where the login is stored

The personal login is a refresh token, and it has to survive the command that created it — otherwise the next command is signed out again.

The CLI always prefers the OS credential store: macOS Keychain, Windows Credential Manager, or libsecret (Secret Service) on Linux. When that store cannot keep the token, the CLI falls back to a permission-protected file under ~/.agentteams/credentials, one file per API server:

• POSIX: directory 0700, file 0600, and the CLI refuses to read or write the file if it is a symlink, is owned by another user, or is reachable by group or other.
• Windows: icacls removes inheritance and grants only the current account. If that cannot be applied or verified, the file backend is not used at all.

The fallback is triggered by the OS store failing, not by which platform you are on, because the three platforms fail at different moments. On Linux the failure is visible up front; on macOS and Windows the availability check passes and only the write fails, which is why an SSH login there used to be revoked after you had already approved it on another device.

A protected file is weaker than an OS keyring: it is not encrypted, so anyone who can read your files or gain root on that machine can read the token. agentteams auth status always names the backend that actually holds the token and repeats that warning. To forbid the fallback — the login then fails before asking you to approve anything, exactly as it did before:

export AGENTTEAMS_DISABLE_FILE_CREDENTIALS=1

This applies to init --device-auth and auth login --device-auth on Linux, macOS and Windows alike, including machines you only ever reach over SSH.

The variable stops the CLI writing a credential file. A file written before you set it stays readable and removable, so agentteams auth logout can still revoke that token and delete it — otherwise setting the variable would strand a live login on disk with no command left that could reach it.

auth status names the reason the OS store was skipped when this process is the one that discovered it — always on Linux, where the failure shows up before the login. On macOS and Windows the failure is only visible at the moment of a write, so a later auth status reports the backend without repeating why; re-run the login (or auth logout and back in) to see the OS store's own error again.

Troubleshooting

| Message | Cause | What to do | | -------------------------------------------------------------------- | ----------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | OAuth callback timed out after 60 seconds. | The default flow was used from a shell the browser cannot call back to | Re-run with --device-auth | | does not support device authorization | The server is an older AgentTeams API | Re-run without --device-auth, or upgrade the server | | device authorization turned off because the server has no APP_URL | Server configuration | Ask the operator to set APP_URL; use the default flow meanwhile | | The device code expired before it was approved | The code is valid for 15 minutes | Run the command again for a new code | | The sign-in request was denied in the browser | Deny was clicked | Run the command again if that was a mistake | | Cannot start device authorization: ... credential store | Neither the OS credential store nor the file fallback can keep a login | Read the reason in brackets — it names both halves. Unset AGENTTEAMS_DISABLE_FILE_CREDENTIALS if you set it, or fix the OS store (install/unlock libsecret, unlock the macOS login keychain, sign in interactively on Windows) | | Store: protected file (~/.agentteams/credentials) in auth status | The OS credential store was unavailable, so the file fallback is in use | Nothing is broken. Fix the OS store if you want the stronger backend; on Linux the message after it names the reason, on macOS and Windows only the run that first fell back can show it |

Service URLs (Defaults and Overrides)

The CLI talks to two services:

• Web app (OAuth flow): defaults to https://agentteams.run
• API: defaults to https://api.agentteams.run; init only writes apiUrl when the authorized API URL is different

2. Use Conventions

agentteams convention list
agentteams convention show
agentteams convention download
agentteams convention create --file .agentteams/rules/new-rule.md
agentteams convention update --file .agentteams/rules/context.md
agentteams convention update --file .agentteams/rules/context.md --apply
agentteams convention delete --file .agentteams/rules/context.md
agentteams convention delete --file .agentteams/rules/context.md --apply

Breaking Changes

Every resource command is now <resource> <action> with its own subcommand, so each action's --help lists only the options it actually reads. The following public arguments were removed in that move. Passing one fails at argument parsing with unknown option / unknown command, and the CLI prints a hint: line with the replacement.

| Removed | Replacement | | ---------------------------------------------------- | ----------------------------------------------------------------------------------- | | --format | Nothing — output is already JSON. Still available on init/auth/sync/doctor. | | --api-key | --api-key-file <path> or AGENTTEAMS_API_KEY | | --team-id | AGENTTEAMS_TEAM_ID or the teamId config field | | --limit (report/postmortem/coaction/document list) | --page-size (search --limit is unchanged) | | plan show, code-review show, task show | get | | plan issue | plan link-issue (issue stays as an alias for one release) |

If an agent in your project still passes --format json, its convention copy is stale — run agentteams convention download to refresh it.

Core Commands

init

Initialize the CLI via OAuth and download conventions.

agentteams init

convention

Manage project conventions.

agentteams convention list
agentteams convention show
agentteams convention download
agentteams convention create --file .agentteams/rules/new-rule.md
agentteams convention update --file .agentteams/rules/context.md
agentteams convention update --file .agentteams/rules/context.md --apply
agentteams convention delete --file .agentteams/rules/context.md
agentteams convention delete --file .agentteams/rules/context.md --apply

convention download saves files by category in .agentteams/<category>/. If file names collide within the same category, suffixes like -2, -3 are added.

convention create

Create a new convention.

• The input file must be under .agentteams/<category>/*.md, and <category> is inferred from the path.
• Frontmatter is optional. Supported fields: trigger, description, title (optional). Use # AGENT_RULES in the document body instead of the deprecated agentInstruction frontmatter field.
• After creation, the CLI immediately updates .agentteams/conventions.manifest.json, so you can update/delete the same file right away.
• Run agentteams convention download if you want to refresh convention.md and the server-normalized (downloadable) markdown.

Examples:

agentteams convention create --file .agentteams/rules/new-rule.md

convention update / convention delete

• By default, update and delete run in dry-run mode. They print a diff/plan and do not modify the server.
• Use --apply to actually update/delete the server resource.
• Only files produced by agentteams convention download are allowed. The CLI uses .agentteams/conventions.manifest.json to map local files to server conventions.

Examples:

# Preview changes (dry-run)
agentteams convention update --file .agentteams/rules/context.md

# Apply update to server
agentteams convention update --file .agentteams/rules/context.md --apply

# Preview deletion (dry-run)
agentteams convention delete --file .agentteams/rules/context.md

# Apply deletion to server
agentteams convention delete --file .agentteams/rules/context.md --apply

Common errors:

• 403 Forbidden: the server rejected the operation due to missing write permissions.
• 409 Conflict: optimistic-lock conflict (someone else updated the convention). Download again and retry.

sync

컨벤션과 스킬 패키지를 동기화합니다. 로컬에서 수정·추가·삭제한 스킬 파일이 있으면 해당 패키지를 보존하고 충돌 경로를 안내합니다. session sync에도 같은 보호가 적용됩니다.

agentteams sync

skill download

스킬 패키지를 내려받고 도구별 미러를 갱신합니다. 원본 또는 미러에 로컬 변경이 있으면 해당 패키지의 다운로드와 정리 삭제를 적용하지 않습니다. 서버에서 삭제되거나 이름이 바뀐 스킬, 미러 대상 축소에도 같은 규칙이 적용됩니다. 충돌하지 않는 다른 패키지는 계속 동기화합니다.

agentteams skill download
agentteams skill download --id <skill-id>
agentteams skill download --skill-targets agents,claude
agentteams skill download --skill-targets none

충돌 출력에는 적용되지 않은 패키지, 파일 경로와 사유가 표시됩니다. modified는 수정, added는 로컬 추가, deleted는 로컬 삭제, unknown은 배포 기준 없음, unsafe는 심볼릭 링크 등 자동 처리 불가를 뜻합니다. 파일 내용은 출력하지 않습니다. 구버전에서 받은 패키지는 원본 상태를 확인할 기준이 없어 내용이 같아도 충돌로 보존될 수 있습니다.

로컬 변경을 유지하려면 파일을 백업하거나 필요한 내용을 서버 패키지에 반영한 뒤 다시 다운로드하세요. 서버 상태로 교체하기로 결정한 경우에는 패키지 ID를 지정해 강제 실행합니다.

agentteams skill download --id <skill-id> --force

이 명령은 선택한 패키지의 원본·미러와 이전 이름 경로에 있는 로컬 변경을 버립니다. 서버에서 삭제된 패키지를 선택하면 로컬 사본도 제거합니다. 다른 패키지는 변경하지 않으며, --id 또는 --all 없는 --force는 거부합니다. 미러 범위를 제한하려면 --skill-targets도 함께 지정하세요. 심볼릭 링크는 강제 실행으로 처리하지 않으므로 먼저 직접 정리해야 합니다.

스킬 미러 루트 전체가 삭제되면 선택된 미러를 다시 만들며, 링크로 바뀐 미러는 해당 경로만 건너뛰고 같은 이름의 원본 갱신을 계속합니다. 미러 안의 개별 파일 편집·삭제·추가는 보호합니다. 건너뛴 미러는 배포 완료로 기록하지 않습니다.

모든 구버전 패키지를 한 번에 이행하려면 로컬 편집을 백업한 뒤 agentteams skill download --force --all을 실행하세요. 이 명령은 모든 패키지와 미러의 로컬 변경을 서버 상태로 교체하고 서버에서 삭제된 패키지를 제거합니다. --all은 --force가 필요하며 --id와 함께 쓸 수 없습니다. 자동 동기화에서는 이 옵션을 전달하지 않습니다.

구버전 패키지를 강제로 내려받은 뒤에는 새 배포 기준이 기록되어 다음 실행부터 변경 여부를 판별합니다. 일반적인 다운로드·적용 오류가 나면 기존 파일과 manifest를 보존하거나 복원하므로 원인을 해결한 뒤 같은 명령을 다시 실행할 수 있습니다. Skill rollback failed가 표시되면 안내된 복구 디렉터리를 보존하고 백업과 현재 파일을 확인하세요. skills.sync.lock 오류가 나면 다른 다운로드가 실행 중인지 확인하고, 강제 종료된 실행임과 남은 복구 디렉터리를 확인한 뒤 agentteams skill download --release-lock으로 잠금만 해제하고 다운로드를 재시도하세요. 잠금에는 PID와 획득 시각이 기록되며, 해제 명령은 실행 중인 PID의 잠금을 거부합니다. 다운로드 옵션과 해제 옵션은 함께 사용할 수 없습니다.

session

Session-start sync. Checks both surfaces, downloads only what changed, and reports which always-on files must be re-read (reread) or no longer apply (invalidated). Exits cleanly on failure so a session start is never blocked.

agentteams session sync

guide

Read the platform guide that governs a record type. Same loader as the agentteams_guide_get MCP tool, which exists only in the full tool profile — this is the path for sessions without it.

agentteams guide list
agentteams guide get --record-kind completion-report

plan

Manage plans.

Note: Plans are always created as BACKLOG. Even if you pass --status to plan create, the server will ignore it. Use plan update to change status after creation.

agentteams plan list
agentteams plan get --id <plan-id>
agentteams plan get --id <plan-id> --include-deps
agentteams plan status --id <plan-id>
agentteams plan set-status --id <plan-id> --status <status>

agentteams plan create \
  --title "Implement feature" \
  --content "Detailed content" \
  --type FEATURE \
  --complexity FULL \
  --priority HIGH

# optional checklist template for create
agentteams plan create \
  --title "Refactor module" \
  --template "refactor-minimal"

  # repository linkage
  # - `plan create` sends the current git origin URL by default.
  # - Use `--repository-remote-url <url>` to override it.

# quick log: record already-done work (creates a plan + report in one shot)
agentteams plan quick --title "Quick task" --content "Implemented X and verified with tests" --type CHORE
agentteams plan update --id <plan-id> --status TODO
agentteams plan update --id <plan-id> --status IN_PROGRESS
agentteams plan download --id <plan-id>
agentteams plan cleanup --id <plan-id>
agentteams plan delete --id <plan-id>

Status values: BACKLOG, TODO, ASSIGNED, IN_PROGRESS, BLOCKED, DONE, CANCELLED

Types: FEATURE, BUG_FIX, ISSUE, REFACTOR, CHORE

Complexity: MINIMAL, STANDARD, FULL — required for plan create (--complexity). MINIMAL = 1 task / 1–2 files / single domain; STANDARD = 2–3 tasks / known scope; FULL = 4+ tasks, multi-wave, or risk signals (schema/auth/billing/quota/deployment, cross-workspace, large diff, unfamiliar domain). Changing complexity on update (--complexity with optional --complexity-reason) records a MODIFICATION comment.

Priorities: LOW, MEDIUM, HIGH

Plan template values (create): refactor-minimal, quick-minimal

plan quick (quick log) behavior — the path for recording work you already finished without a pre-existing plan:

• Creates a plan with --content as the plan body (--content or --file is required)
• Uses LOW as the default priority (override with --priority)
• Defaults to MINIMAL complexity (override with --complexity)
• Starts and finishes the plan in one flow
• Does not attach a completion report; use the full plan workflow for detailed reporting

--include-deps (get/show): also fetches dependency data and appends a Dependencies section in text output; in json output it adds data.dependencies with blocking and dependents arrays.

comment

Manage plan comments.

agentteams comment list --plan-id <plan-id>

agentteams comment create \
  --plan-id <plan-id> \
  --type GENERAL \
  --content "Work completed"

agentteams comment update --id <comment-id> --content "Updated content"
agentteams comment delete --id <comment-id>

Types: RISK, MODIFICATION, GENERAL

dependency

Manage plan dependencies.

agentteams dependency list --plan-id <plan-id>
agentteams dependency create --plan-id <plan-id> --blocking-plan-id <blocking-plan-id>
agentteams dependency delete --plan-id <plan-id> --dep-id <dependency-id>

agent-config

View or delete agent configurations.

agentteams agent-config list
agentteams agent-config get --id <config-id>
agentteams agent-config delete --id <config-id>

report

Manage completion reports.

A completion report is always tied to a plan, so report create requires --plan-id — there is no standalone (plan-less) report. To record work you already finished without a pre-existing plan, use a quick log (agentteams plan quick) instead.

Tip: Include reproducible verification evidence (commands + outcomes), but keep outcomes short: pass/fail + 1–3 lines of summary. Do not paste long raw logs into the report body.

agentteams report list

agentteams report create \
  --plan-id <plan-id> \
  --title "AgentTeams completion report" \
  --file ./report.md \
  --status COMPLETED

# repository linkage
# - `report create` sends the current git origin URL by default.
# - Use `--repository-remote-url <url>` to override it.

# with metrics (auto + manual)
agentteams report create \
  --plan-id <plan-id> \
  --title "CLI metrics report" \
  --file ./report.md \
  --files-modified 5 \
  --lines-added 120 \
  --lines-deleted 30 \
  --quality-score 95

# disable git auto collection
agentteams report create \
  --plan-id <plan-id> \
  --title "Manual metrics report" \
  --file ./report.md \
  --no-git

Status values: COMPLETED, FAILED, PARTIAL

Metrics behavior:

• Auto-collected on report create (unless --no-git): commitHash, branchName, filesModified, linesAdded, linesDeleted
• Manual only: durationSeconds, commitStart, commitEnd, pullRequestId
• Manual options always override auto-collected values

postmortem

Manage post mortems.

Tip: If you have platform guides downloaded under .agentteams/platform/guides/, prefer the template in post-mortem-guide.md.

agentteams postmortem list

agentteams postmortem create \
  --title "Deployment incident analysis" \
  --content "## Root cause\n- Missing configuration" \
  --action-items "Automate rollback,Pre-release checklist" \
  --status RESOLVED

# repository linkage
# - `postmortem create` sends the current git origin URL by default.
# - Use `--repository-remote-url <url>` to override it.

Status values: OPEN, IN_PROGRESS, RESOLVED

search

Search across all entity types in a project.

# Basic search
agentteams search --query "login feature"

# Filter by entity types
agentteams search --query "auth" --types PLAN,CO_ACTION

# Limit results and token budget (useful for agents)
agentteams search --query "deployment" --limit 5 --max-tokens 4000

# Output is JSON by default, so it pipes straight into jq
agentteams search --query "refactor"

Searchable entity types: PLAN, CO_ACTION, COMPLETION_REPORT, POST_MORTEM, CONVENTION

resolve

Resolve one entity reference. Pass the reference token exactly as it appears — resolve works out the type, strips the id prefix, and fetches the entity for you, so you do not have to pick the right per-type command yourself.

agentteams resolve "plan:agentteams_pln_f62762fc-730a-4201-8586-e2541505ed1b"
agentteams resolve "agentteams_doc_f62762fc-730a-4201-8586-e2541505ed1b"
agentteams resolve "convention:<id>:.agentteams/rules/context.md"
agentteams resolve "codeReview:<reviewId>:<findingId>"
agentteams resolve "GITHUB_ISSUE:owner/repo#12"

# A whole markdown link works too
agentteams resolve "[Safari pull-to-refresh](plan:agentteams_pln_<uuid>)"

Accepted forms: type:id, type:id:path, a bare prefixed id, an external marker (LINEAR_ISSUE, GITHUB_ISSUE, GITHUB_PR, GITLAB_ISSUE, GITLAB_MERGE_REQUEST, BITBUCKET_ISSUE, BITBUCKET_PR), or any of those wrapped in a [label](...) link. Three-part references are read by their parent type: convention:id:path carries a local path, while codeReview:reviewId:findingId and plan:planId:taskId address a child entity.

The response kind tells you what to do next:

| kind | Types | What to do | | ----------- | ---------------------------------------------------------------- | ----------------------------------------------------------- | | file | plan, completionReport, postMortem, coAction, document | Read filePath — the body was downloaded as a local .md. | | record | codeReview, codeReviewFinding, planTask, LINEAR_ISSUE | Use record, the structured payload returned inline. | | localFile | convention with a path | Read filePath (no network call was made). | | external | GITHUB_*, GITLAB_*, BITBUCKET_* | Open url or run suggestedCommand (gh / glab). |

filePath is always absolute, so it opens the same way from any working directory.

References are user-authored text, so resolve validates them before acting: an AgentTeams id (and the parent id of a three-part reference) must be a bare UUID, otherwise the reference is rejected instead of being sent to the API. The path in a convention:id:path reference must resolve inside the project's .agentteams/ directory; anything else falls back to the server record rather than nominating a local file to read.

External references are never fetched for you: resolve derives the URL and the gh/glab command from the locator and stops there. GitLab references return a command but no URL, because a self-managed GitLab host cannot be derived from the project path.

Every response also carries fallbackCommand, the equivalent per-type command (report download --id, code-review get --finding-id, task get --task-id, linear issue get --issue-id, …) for older CLI versions that do not ship resolve.

Agents connected over MCP get the same dispatch as the agentteams_resolve tool (full profile only), so they can resolve a reference without shelling out. One difference: there is no file kind — a plan or report body comes back inline as a record rather than downloaded, since an MCP server has no working-directory contract. A localFile result still carries an absolute filePath, plus the reference's project-root-relative path; when the MCP session is not bound to a local checkout only path comes back, and — as with the command — a path that does not exist under .agentteams/ degrades to the server record.

config

agentteams config whoami

config whoami prints current environment variable values for AGENTTEAMS_API_KEY and AGENTTEAMS_API_URL.

Configuration

Configuration is merged in this priority order (highest first):

1. CLI options
2. Environment variables (AGENTTEAMS_*)
3. Project config (.agentteams/config.json)
4. Global config (~/.agentteams/config.json)

Config File Example

{
  "teamId": "team_xxx",
  "projectId": "proj_xxx",
  "apiKey": "key_xxx"
}

apiUrl is omitted for the default API. When AGENTTEAMS_API_URL points somewhere else — a local API or the dev environment — init persists that value so later commands keep using the same API. The API destination comes from the CLI's own environment only; the browser callback never supplies it.

Environment Variable Example

export AGENTTEAMS_API_KEY="key_your_api_key_here"
export AGENTTEAMS_API_URL="https://api.agentteams.run"
export AGENTTEAMS_TEAM_ID="team_xxx"
export AGENTTEAMS_PROJECT_ID="proj_xxx"

Output Format

리소스·자동화 명령은 기본적으로 전체 구조화 결과를 stdout에 JSON으로 출력합니다. 따라서 별도의 포맷 옵션 없이도 기본 출력을 JSON.parse, jq 등의 파이프라인 소비자에 안전하게 전달할 수 있습니다.

출력 계약은 다음과 같습니다.

• 기본: 리소스·자동화 명령의 전체 JSON 결과를 stdout에 출력합니다.
• 사람용 예외: init, auth, sync, doctor는 사람이 읽기 쉬운 기본 출력을 유지합니다. 기계가 읽을 수 있는 결과가 필요하면 이 네 명령에 한해 --format json을 전달합니다. 다른 명령에는 --format이 없습니다(기본 출력이 이미 JSON입니다).
• --output-file <path>: 전체 결과를 파일에 저장한 뒤, 저장 경로와 간결한 요약을 stdout에 출력합니다.
• --verbose: 전체 raw 결과를 stdout에 출력합니다. --output-file과 함께 사용하면 저장 경로 요약 뒤에 전체 결과도 이어서 출력합니다.

Prefixed entity IDs

Entity references copied from the AgentTeams web UI carry a type prefix (e.g. agentteams_pln_<uuid>). Any --id/--plan-id/--completion-report-id/etc. value is normalized to its bare id automatically, so you can paste a prefixed id directly:

agentteams plan get --id agentteams_pln_f62762fc-730a-4201-8586-e2541505ed1b
# resolves to plan f62762fc-730a-4201-8586-e2541505ed1b

This normalization only removes the prefix — you still choose the command that matches the entity type. resolve sits one level above it: give it the whole reference token and it works out the type as well, then dispatches to the right lookup.

agentteams plan list
agentteams plan update --id <plan-id> --status IN_PROGRESS

Note: --format exists only on init, auth, sync, and doctor. Every other command already prints JSON.

Error Guide

The API may include an optional machine-readable errorCode in error responses:

{
  "statusCode": 403,
  "error": "Forbidden",
  "message": "Cross-project access denied",
  "errorCode": "CROSS_PROJECT_ACCESS_DENIED"
}

The CLI prioritizes errorCode when present and falls back to message/status-based handling for older API responses.

| Error | Meaning | Resolution | | ------------------ | ------------------------------ | -------------------------------------- | | 401 Unauthorized | Invalid API key | Check apiKey or AGENTTEAMS_API_KEY | | 403 Forbidden | No access to project | Verify projectId | | 404 Not Found | Resource does not exist | Verify ID or create the resource | | Network error | Cannot reach server | Check apiUrl and server status | | Missing config | Config file/env vars not found | Run agentteams init |

License

Apache-2.0

This license applies to the CLI code distributed in this package. Use of the AgentTeams service/API may require credentials and is governed by separate service terms/policies.