action-platform 0.2.0

Your platform team, as a CLI.

Every new service costs the same week: scaffold, lint config, CI, versioning, deploy pipeline, IAM. Then the next one drifts from the last. Action Platform turns that week into one command and keeps every project on the same rails — without a control plane to host, a UI to learn, or a vendor to trust with your cloud.

pipx install action-platform

action-platform init web python fastapi --name "orders" --cloud aws/lambda

Thirty seconds later you have a FastAPI service with tests, lint, CI wired, a SAM template, a least-privilege IAM policy, and a GitHub repo already pushed (--no-push to keep it local).

Why teams pick it

What you get

action-platform init --list

Commands

action-platform init                              # interactive: type → stack → template → name → ci
action-platform init web go gin --ci gitlab       # direct
action-platform init web python --cloud docker    # project + deploy overlay
action-platform init ... --no-push                # skip creating the remote repo
action-platform install [--dry-run]               # existing repo: platform.toml, hooks, code quality, CI — never overwrites

action-platform cloud set aws/lambda              # add or switch the deploy target
action-platform service add postgres --provider aws-rds

action-platform branch feature 42 login           # develop (or main) → pull → feature/42-login → push
action-platform branch hotfix PROJ-7              # from main/master
action-platform gitflow                           # audit current branch + commits; --install-hooks
action-platform pr [--draft] [--dry-run]          # PR for the current branch: target from git-flow, body from commits
action-platform release patch                     # bump, changelog, tag, GitHub release; off main/master → X.Y.Z-rc.N pre-release
action-platform deploy --stage prod
action-platform rollback
action-platform diagnose
action-platform destroy

Git-flow, enforced

Every project follows the same flow. Git hooks refuse the wrong move before it exists; CI refuses it on the pull request; the CLI and the MCP tools guide the right one.

gitGraph
    commit id: "chore: bootstrap"
    branch develop
    checkout develop
    commit id: "chore(release): 0.3.2-rc.1" tag: "v0.3.2-rc.1"
    branch feature/42-login
    checkout feature/42-login
    commit id: "feat(login): form"
    commit id: "test(login): cover form"
    checkout develop
    merge feature/42-login id: "PR #1 → develop"
    branch release/0.3.2
    checkout release/0.3.2
    commit id: "chore(release): 0.3.2-rc.2" tag: "v0.3.2-rc.2"
    checkout main
    merge release/0.3.2 id: "PR #2 → main"
    commit id: "chore(release): 0.3.2" tag: "v0.3.2"
    checkout develop
    merge main id: "back-merge"
    checkout main
    branch hotfix/PROJ-7
    checkout hotfix/PROJ-7
    commit id: "fix(auth): expiry"
    checkout main
    merge hotfix/PROJ-7 id: "PR #3 → main"
    commit id: "chore(release): 0.3.3" tag: "v0.3.3"
    checkout develop
    merge main id: "back-merge hotfix"

Rules the hooks and CI apply: branch names are <kind>/<code>[-slug]; commits are Conventional Commits; no direct commits on main, master or develop except chore(release):, chore(platform): and the bootstrap commit; a pull request may only target what the table allows.

action-platform branch feature 42 login     # develop → pull → feature/42-login → push
action-platform gitflow                     # audit branch + commits
action-platform pr                          # target and body from the rules and the commits
action-platform release patch               # rc off main, stable on main

Everything a project needs is declared in one file:

[project]
name = "orders"
type = "web"
stack = "python"
template = "fastapi"
ci = "github"
language = "python"

[source_host]
kind = "github"
repo = "acme/orders"

[deploy]
target = "aws/lambda"

[services]
postgres = "aws-rds"

Use it from an AI client

The platform ships as an MCP server. Claude Code, Codex, Cursor — anything that speaks MCP — gets 17 tools (list_matrix, init_project, install_platform, start_branch, gitflow_audit, propose_pull_request, release, deploy, diagnose, …), 6 prompts that put them in the right order (new_service, ship_feature, cut_release, deploy_project, adopt_repository, fix_gitflow) and 12 skills that make the agent preview and ask before anything leaves the machine.

pip install "action-platform[mcp]"
action-platform mcp                 # stdio
action-platform mcp --http          # http://127.0.0.1:8765/mcp

Claude Code:

/plugin marketplace add actionplatform/action-platform
/plugin install action-platform@action-platform

Any client — add to .mcp.json:

{ "mcpServers": { "action-platform": { "command": "uvx", "args": ["--from", "action-platform[mcp]", "action-platform-mcp"] } } }

Extend it

Deploy targets are plugins. Implement the DeployTarget contract — preflight, create, deploy, switch_traffic, rollback, diagnose, delete — publish it under the action_platform.deploy_target entry-point group, and action-platform deploy finds it by name.

from action_platform import ActionPlatform, Config
from action_platform.providers import SourceGithub

tool = ActionPlatform(config=Config(source_host=SourceGithub(repo="acme/orders")))
tool.release("minor")

Templates are plain cookiecutters in actionplatform/templates. Add a stack or a cloud with a pull request — the CLI reads index.toml, nothing to redeploy. Point ACTION_PLATFORM_TEMPLATES at a local checkout while you work on them.

Ecosystem

License

Apache 2.0.