taskswarm (Python)

Self-hosted event server that pushes a notification the instant a parallel coding-agent session blocks, needs review, fails, or finishes.

Why this exists

Running three or four coding-agent sessions in parallel turns into a tab-switching problem: nothing pushes state to you, so you find out a session has been sitting on a permission prompt for ten minutes only by polling terminals by eye. TaskSwarm closes that gap: every agent session reports its state to a small local HTTP server, and the instant one transitions to blocked, needs-review, failed, or done, TaskSwarm fires a local OS notification and updates a live status page over Server-Sent Events. No polling, no account, no cloud dependency by default.

This package is the Python distribution of TaskSwarm -- a genuine, independent port of the server, CLI, event schema, and notification logic, not a wrapper around the Node binary. It has zero runtime dependencies: no HTTP framework, no schema-validation library, nothing beyond the Python standard library.

Install

Live on PyPI as taskswarm-cli (renamed from the original taskswarm package, which has stopped receiving updates and points here):

pip install taskswarm-cli

or with uv:

uv add taskswarm-cli

The complementary JS/TS distribution is already live on npm as taskswarm-cli (npm install -g taskswarm-cli, or npx taskswarm-cli start with no install step) -- see the project README for that package. Both are first-class and maintained together; neither is a replacement for the other.

Quickstart

# Terminal 1: start the server
taskswarm start
# TaskSwarm server listening on http://127.0.0.1:4173
# Live status page: http://127.0.0.1:4173/?token=<your-token>

# Terminal 2: report a session's status as it works
taskswarm agent report-status --task my-fix --repo ./api --state running
taskswarm agent report-status --task my-fix --repo ./api --state done

Open the live status page URL printed by start. The row for my-fix updates the instant each report-status call lands, no refresh.

Or call the library directly, in-process, without a subprocess:

from taskswarm import start_server
from taskswarm.adapters import GenericAdapter
from taskswarm.client.api_client import post_event

running = start_server()
adapter = GenericAdapter()
event_input = adapter.to_event_input(
    {"session_id": "my-fix", "repo": "./api", "status": "done", "agent_type": "generic"}
)
event = post_event(running.config.to_dict(), event_input)
print(event["session_id"], "->", event["status"])
running.close()

What it does

• Event server, not a board you have to keep open. taskswarm start boots an HTTP+SSE server (stdlib http.server, no framework dependency) that accepts POST /events, serves GET /events for current session state, and streams new events over GET /live (Server-Sent Events) to the bundled live status page.
• Push notification on the four states that mean "look at this now." The moment a session's status transitions to blocked, needs-review, failed, or done, TaskSwarm fires a native OS notification (osascript on macOS, a terminal-bell console fallback elsewhere). Notification dedup keys on the (status, blocked_reason) pair, so a second, different permission prompt still notifies even though the status (needs-review) didn't change.
• A wrapper-script adapter for any agent. taskswarm agent report-status --task <id> --repo <path> --state <state> is the one primitive every integration is built on -- callable from any script wrapping any CLI agent (Codex, Cursor, or anything else).
• A Claude Code hooks adapter, ported with the same behavior as the npm package's: taskswarm hooks install claude-code writes Stop and Notification hook entries into .claude/settings.json, pointed at the exact, already-installed console script on disk (never a floating PATH lookup re-resolved on every hook fire).
• A bearer-token-gated local API, bound to loopback by default. POST /events and the live page both require the token TaskSwarm generates on first run (~/.taskswarm/config.json, written 0600). Rotate it with taskswarm token rotate.
• Agent-native by design. Every subcommand ships a --json flag with a stable schema, including error output.
• ntfy.sh is opt-in, never default. The only notification channel that leaves your machine, and it's off unless you configure it in ~/.taskswarm/config.json.

CLI reference

taskswarm --version prints the installed package's own version number, read live from pyproject.toml (so it always matches whatever release you actually have installed, rather than a fixed number that would go stale here). The taskswarm-cli console script installed by this package is an identical alias, matching both bin entries the npm package ships.

How it works

agent session / hook / wrapper script
        |
        v
POST /events (bearer token required)  -> event schema validation
        |
        v
EventStore (in-memory + append-only JSONL log at ~/.taskswarm/events.jsonl)
        |
        +--> notify() -- fires on a transition into blocked/needs-review/failed/done
        |         |
        |         +--> local OS notification (always on)
        |         +--> ntfy.sh (opt-in only)
        |
        +--> GET /live (Server-Sent Events) -> live status page

Full data model and the exact notification-dedup rule are in docs/concepts.md.

Security

The local API is gated by a bearer token generated on first run and stored 0600 at ~/.taskswarm/config.json; the server binds to 127.0.0.1 by default. Token comparison uses hmac.compare_digest (constant-time, avoiding timing side-channels), the same property the TypeScript version gets from crypto.timingSafeEqual. See SECURITY.md for the full posture, including what this server does not protect against (it is a local developer tool, not designed to be exposed on a shared or public network). Honest note: this project does not currently publish SLSA provenance, Sigstore signatures, or an SBOM, and has no OpenSSF Scorecard badge -- none of that infrastructure exists yet for either distribution, so it isn't claimed here.

Contributing

See CONTRIBUTING.md for the full guide, covering both the TypeScript and Python codebases.

cd python
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

License

MIT, see LICENSE.