Files
HermenixandClaude Opus 5 0f2810afaa docs(sessions)!: sessions-v4, the per-session live stream
`docs/CONTRACT-sessions.md` goes to sessions-v4 with a change table at the top,
for the reason mcp-v2 got one: everything in it is an addition, and a reader who
learned v3 should be told what moved rather than diff it.

The new §3.5 is the interface that needed freezing most, because it is the only
one in this build that two *processes* have to agree on: `/run/hhstream/live.sock`,
`LiveStreams.attach/detach/list/viewers`, the four JSON ops on the wire, and the
`Attachment` the broker gets back. Its shape is §3.4's deliberately -- the broker
holds the decision (which session streams, under which phrase), the daemon holds
the capture -- so a reader who knows the pointer driver already knows this.

Four rules in §3.5 are written out because each one is a thing a reasonable
implementation would get wrong:

* the daemon holds no session state and no idle clock; `viewers` is a reading,
  and what the broker does with it is the broker's;
* `fps` is optional and omitting it is normal -- the default is the daemon's,
  from its own unit's argv, for the same reason the port is. A broker that sent
  its own default on every attach would turn `sessions.stream.fps` into a
  setting that changed nothing, which is the failure `--xkb-layout` exists to
  avoid;
* `attach` is idempotent on the phrase, so a re-`enable` after a broker restart
  does not double the `grim` rate on one output;
* the phrase is never logged by either side. It is the stream's only credential.

§4 gains four verbs, the three `start` arguments, and the one genuinely new rule
in this version: **a connected viewer is activity**. That is the first exception
to "`touch` is the only mark of activity" since that rule was written, and it is
there for one failure -- an operator watching a long job on his phone, saying
nothing to the broker, and losing the session at three hours *because* watching
was all he was doing. It cannot hold a session open by itself: an enabled stream
with no viewer runs no `grim` and stamps nothing.

No new exit numbers. The three error names map onto codes the frozen table
already has (`stream_unavailable` -> 10, `stream_busy` -> 8, `stream_budget` ->
6), because no new *kind* of failure appeared: no container output is a
compositor-side failure, a phrase collision is a busy, a viewer cap is a budget.

§6 records schema 3 -> 4 (the record gains `stream`, older records migrate to
`null`), and §5 names the socket, the unit and port 9988 beside the other frozen
paths.

What v4 deliberately withholds from the broker is the port, the bind address and
the capture command. The broker learns the URL prefix *from* the daemon, in the
`attach` reply -- a URL it assembled itself from a port it was told about once
is a URL that lies the first time an operator moves the listener.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011ZeH65zd6pFTgBSu52W9PE
2026-10-11 22:46:54 +00:00

37 KiB

CONTRACT — sessions build (sessions-v4)

This file is the interface for the session work: the broker, containers, compositor IPC and pointer and keyboard input. It is the sibling of docs/CONTRACT.md, which freezes the frame-socket wire format; that one is untouched by this build.

Immutable without a version bump. Every slice codes against this and nothing else. If something here is wrong or missing, stop and report it in your final message — the parent edits this file, not you. Do not widen your slice to work around a gap.

Version: sessions-v4 (2026-10-11). v1 was the freeze the four slices built against; v2 recorded the decisions their reports asked for; v3 adds the keyboard; v4 adds the per-session live stream. Every change in all of them has been an addition or a clarification — no v1, v2 or v3 signature changed meaning — so earlier code stays valid except where noted.

v4 changes, all of them the stream (docs/plans/2026-10-04-session-stream-webapp.md, Stage 2):

§ Change Why
§3.5 New: hhstreamd/live.py, the broker's end of /run/hhstream/live.sock, and that socket's wire format. The stream is captured by hhstream-live, a third executor beside the pointer driver. grim blocks tens to hundreds of milliseconds, so it must not run in the broker, and a LAN-facing HTTP parser must not run in the frame server whose one-socket contract cannot grow one. The broker therefore tells a process what to publish, exactly as it tells the driver where to warp.
§4 Verbs stream.enable, stream.disable, stream.list, stream.status. enable and disable are in the mutation set; list and status are not. enable mints or rotates a phrase and opens a capture — a retried enable that minted a second phrase would invalidate the URL somebody is already holding.
§4 start gains stream?, stream_phrase? and stream_fps?. A stream needs the session's container output, which start creates; sequencing "start, then enable" in a caller means every caller repeats the ordering and the failure handling. One request, one answer, and the phrase comes back with the session.
§4 A connected viewer is activity. The broker asks the daemon for viewer counts about once a minute and stamps touch when any is non-zero. The one new way the idle clock moves, and the only exception to "touch is the only mark of activity" since that rule was written. A session an operator is watching on his phone must not be reaped at three hours because watching is all he is doing. An enabled stream with no viewer runs no grim and keeps nothing alive, so this cannot hold a session open by itself.
§4 Error names stream_unavailable (exit 10), stream_busy (exit 8), stream_budget (exit 6). No new exit numbers. The failures are the ones the table already has: no container is a compositor-side failure, a phrase collision is a busy, a viewer cap is a policy budget.
§5 /run/hhstream/live.sock, the unit hhstream-live, and the stream port 9988. Named here with the other frozen paths, because two processes have to agree on them.
§6 Registry schema 3 → 4: the record gains stream. A phrase has to survive a broker restart and a re-enable, so it lives in the record. Older records migrate to stream: null and are adopted, not refused.

v4 deliberately does not give the broker the port, the bind address or the capture command. Those are the daemon's, from its own unit's argv, for the reason the keyboard layout is the driver's: the broker's TOML is read by hhstreamd, and a setting an operator could change in it while the listener stayed where it was would be worse than no setting at all. The broker learns the URL prefix from the daemon, in the attach reply.

v3 changes, all of them keyboard (docs/plans/2026-10-03-keyboard-input.md as amended by docs/plans/2026-10-04-keyboard-verification.md):

§ Change Why
§3.3 PointerLease gains press_key / unpress_key / keys_pressed; LeaseOutcome gains keys_lifted. expire() still returns LeaseInfo and nothing else. One wl_seat means one keyboard focus, so the keyboard shares the pointer's lease. §6.7's "the lease gains a keyboard flag" is the wrong shape: a boolean cannot say to lift Control_L and not a, and order matters for the same reason it does for buttons.
§3.4 Driver ops key and type; pressed gains keys[]; release_all gains lifted_keys[]; the frozen command line gains four --xkb-* flags. A rejected request may carry a kind. There is no keyboard frame and no release_all_keys: the protocol has no frame, and expiry has one call site. The flags are the only path a layout has to the uploaded keymap — the broker's TOML is read by hhstreamd, and the driver is a different unit that never opens it.
§4 Verbs keyboard.key and keyboard.type, both in the mutation set; pointer.state gains keys[]; pointer.release gains keys_lifted[]. A retried type must not type twice, for the reason a retried click must not click twice.
§4 pointer.acquire takes keyboard focus as well as the cursor, and is two warps on the driver socket rather than one. Measured (tests/README-keyboard-probe.md): one warp onto an idle cursor delivers wl_pointer.enter and no wl_keyboard.enter, so a one-warp acquire aims the cursor at the session and leaves the keys with whatever was focused before.

v2 changes:

§ Change Why
§3.1 Monitor gains id: int clients -j reports a client's monitor as a numeric index, Workspace.monitor as a name, so without an id there is no way to join a client to its output. Slice D's aiming is output-local.
§3.1 Client.stable_id defined as "<initial_class>:<pid>" (fallback address); absent pid becomes -1; Event.data is a per-event field table plus always-present raw/args; request_timeout_s is an overridable class attribute (2 s); version() returns the compositor's banner verbatim What slice B had to decide, now written down instead of re-derived.
§3.2 quarantine_argv() replaces quarantine_plan(), returning list[list[str]] | None A quarantine rule is not a container: no output, no numbered workspace, no mode, and container_plan refuses a zero size. Integration updates containers.py and its tests to this shape.
§3.3 mark() added; expire() precedence documented §3.3 gave no way to record a position, so LeaseOutcome.bookmark could only ever be None and pointer.acquire … restore? had nothing to restore.
§3.4 Vendored protocol is permitted, and the XML path is corrected The v1 path was wrong (share/wlr-protocols/unstable/wlr-virtual-pointer-unstable-v1.xml).
§5 The broker state dir is stated as an option default with this machine's override; the frame socket is named §5's value was the deployed one, not the default, and the frame socket was missing while docs/CONTRACT.md still documents the old one.

1. What is being built, in one paragraph

A headless Hyprland desktop is driven by an agent. A broker (hhstreamd) starts allowlisted applications as sessions and tracks them. This build gives each session its own output and workspace (a spatial container), teaches the broker to talk to the compositor (IPC), and adds serialised pointer input aimed at one session, under a lease whose bookmark is stored with the session.

Read docs/PLAN-workspaces-and-pointer.md only if your brief tells you to. It describes the whole system; you are building one part.


2. File ownership — the rule that makes four sessions possible

Slice Owns (only these)
A — stage0-rename nix/module.nix, src/hhstream/cli.py (the two socket-path defaults only), examples/agent-eyes
B — ipc-client src/hhstreamd/ipc.py, tests/ipc/**
C — containers src/hhstreamd/containers.py, tests/containers/**
D — pointer src/hhstreamd/pointer.py, src/hhstream_pointer/**, tests/pointer/**, pyproject.toml, nix/package.nix
parent (integration) src/hhstreamd/{protocol,registry,manager,launcher,compositor,procmon}.py, checks/**, docs/**, src/hhstream/cli.py (everything but the path defaults), nix/module.nix (after A merges)

Two slices never edit the same file. A shared-file change you believe you need is a report, not an edit.

Never touch: /etc/nixos (outside this repo), anything under /var/lib/hermes/nixos, any systemd unit, any running service. Never run nixos-rebuild, nixosTest, nix build .#checks.*, or launch a GUI application. The parent serialises all of that on a 2-core machine.


3. Python seams (exact signatures — code to these)

3.1 hhstreamd/ipc.py (B)

class CompositorUnavailable(Exception):
    """No request socket, or the compositor refused. Callers degrade, never guess."""

@dataclass(frozen=True)
class Monitor:
    id: int; name: str; x: int; y: int; width: int; height: int; scale: float
    active_workspace_id: int | None

@dataclass(frozen=True)
class Workspace:
    id: int; name: str; monitor: str | None; windows: int; is_special: bool

@dataclass(frozen=True)
class Client:
    address: str; pid: int; workspace_id: int | None; monitor: str | None
    class_: str; initial_class: str; title: str; stable_id: str; visible: bool

@dataclass(frozen=True)
class CursorPos:
    x: int; y: int

@dataclass(frozen=True)
class Event:
    name: str                 # "openwindow", "workspace", "monitorremoved", ...
    data: dict                # parsed payload; keys are whatever the event carried

class HyprlandIPC:
    @classmethod
    def discover(cls, runtime_dir: Path, signature: str | None = None) -> "HyprlandIPC": ...
    #   raises CompositorUnavailable. Signature from the argument, else
    #   $HYPRLAND_INSTANCE_SIGNATURE, else the single directory under
    #   <runtime_dir>/hypr/. More than one and no signature = CompositorUnavailable.

    def __init__(self, request_socket: Path, event_socket: Path | None = None) -> None: ...

    def version(self) -> str: ...
    def monitors(self) -> list[Monitor]: ...
    def workspaces(self) -> list[Workspace]: ...
    def clients(self) -> list[Client]: ...
    def cursorpos(self) -> CursorPos: ...
    def query(self, command: str) -> str: ...          # raw request socket, for anything not wrapped
    def events(self) -> Iterator[Event]: ...           # blocking; raises CompositorUnavailable on EOF
    def close(self) -> None: ...

def parse_event(line: str) -> Event: ...               # "name>>a,b,c" -> Event

Rules: every method raises CompositorUnavailable rather than returning a guess. clients() parses clients -j; monitors() parses monitors -j; workspaces() parses workspaces -j. Stdlib only.

Decisions, so they are not re-derived per caller:

  • A client's monitor is an index, an output's name is a name. Client.monitor is the numeric index rendered as text (or None when the reply said nothing); join it to a Monitor through Monitor.id. Workspace.monitor is a name.
  • Client.stable_id is "<initial_class>:<pid>", falling back to address. Hyprland publishes no such key; this is the identity that survives a remap, and the launcher sets the class per session.
  • An absent pid becomes -1, not 0 — no process has pid -1.
  • Events are positional on the wire (name>>a,b,c). Event.data is built from a per-event field table, with the last named field absorbing the remaining commas (titles contain them); raw and args are always present so an unknown event is still usable. An event that carries fewer fields than the table names only what arrived — never invented empties.
  • request_timeout_s is a class attribute (2 s) on HyprlandIPC because __init__ is frozen.
  • version() returns the compositor's banner verbatim, not a parsed number.

3.2 hhstreamd/containers.py (C)

Pure logic. No I/O, no subprocess, no socket, no import of ipc.py. It computes command lines and parses nothing. Seams are primitives so this module stands alone.

@dataclass(frozen=True)
class ContainerPlan:
    output: str                        # hh-<session-id>
    workspace: int
    mode: tuple[int, int]              # width, height
    create: list[list[str]]            # full argv lists, in order, each ["hyprctl", ...]
    teardown: list[list[str]]

def output_name(session_id: str) -> str: ...
    # "hh-" + session_id. Ids are already [a-z0-9-]; assert and raise ValueError otherwise.

def contains_output(name: str) -> bool: ...
    # True for names this broker creates (prefix "hh-"), so a start-up scan can
    # find orphans.

def container_plan(session_id: str, workspace: int, width: int, height: int,
                   app_id: str) -> ContainerPlan: ...
    # create:   output create headless <name>          ["hyprctl","output","create","headless",name]
    #           workspace rule                        ["hyprctl","keyword","workspace",
    #                                                  f"{workspace}, monitor:{name}, default:true, persistent:true"]
    #           window rule                           ["hyprctl","keyword","windowrule",
    #                                                  f"match:initial_class ^{app_id}$, workspace {workspace}, monitor {name}"]
    #           output mode                           ["hyprctl","keyword","monitor", f"{name},{width}x{height},auto,1"]
    # teardown: output remove <name>                  ["hyprctl","output","remove",name]

def app_id_for(session_id: str) -> str: ...
    # the bare "hh-<session-id>" the launcher sets on the app; "^hh-<session-id>$"
    # is built in container_plan.

def quarantine_argv(quarantine_workspace: str) -> list[list[str]] | None: ...
    # the backstop: every window matching no container goes to
    # special:quarantine. Returns the keyword argv, or None when the target is
    # blank/whitespace (which is how "disabled" is spelled).
    #
    # Deliberately NOT a ContainerPlan: a quarantine rule has no output, no
    # numbered workspace and no mode. The argv shape mirrors the per-session
    # rule, with a catch-all class:
    #   ["hyprctl","keyword","windowrule",
    #    f"match:initial_class .*, workspace {target}"]
    # It must be installed BEFORE any per-session rule, or the catch-all wins.
    # That ordering is the caller's, and this module cannot enforce it.

class WorkspaceAllocator:
    def __init__(self, base: int, count: int) -> None: ...
    def allocate(self) -> int: ...       # lowest free slot in [base, base+count); raises NoSlots
    def release(self, workspace: int) -> None: ...      # unknown id -> KeyError (a bug, not a no-op)
    def in_use(self) -> list[int]: ...

class NoSlots(Exception): ...

3.3 hhstreamd/pointer.py (D)

Lease state machine + coordinate maths. No Wayland, no socket, no I/O. The clock is injected so expiry is testable without sleeping.

class LeaseBusy(Exception):        # another session holds it
    def __init__(self, holder: str) -> None: ...
class LeaseInvalid(Exception):     # not held, or held by someone else / stale lease_id

@dataclass(frozen=True)
class Bookmark:
    x: float; y: float             # 0..1, output-local, normalised
    output: str
    at: str                        # ISO-8601 UTC

@dataclass(frozen=True)
class LeaseInfo:
    session_id: str; lease_id: str; acquired_at: float; expires_at: float
    hard_deadline: float

@dataclass(frozen=True)
class LeaseOutcome:
    bookmark: Bookmark | None; reason: str; buttons_lifted: list[str]

class PointerLease:
    def __init__(self, ttl_s: float = 30.0, max_s: float = 300.0,
                 clock: Callable[[], float] = time.monotonic) -> None: ...
    def acquire(self, session_id: str) -> str: ...        # -> lease_id; raises LeaseBusy
    def refresh(self, session_id: str, lease_id: str) -> None: ...   # raises LeaseInvalid
    def release(self, session_id: str, lease_id: str, reason: str) -> LeaseOutcome: ...
    def mark(self, session_id: str, lease_id: str, bookmark: Bookmark) -> None: ...
        # records where the cursor was left, called on every move and before
        # release. Added in v2: without it LeaseOutcome.bookmark could only ever
        # be None, so `pointer.acquire --restore` had nothing to restore.
        # raises LeaseInvalid for a non-holder.
    def state(self) -> dict: ...    # {"holder": str|None, "since": float|None,
                                    #  "lease_id": str|None, "expires_at": float|None}
    def expire(self) -> LeaseInfo | None: ...
        # force-release if past ttl or hard cap; returns what was released.
        # The caller (broker) lifts buttons on the driver side.
    def press(self, session_id: str, button: str) -> None: ...
    def unpress(self, session_id: str, button: str) -> list[str]: ...
    def pressed(self) -> list[str]: ...
    # v3: the same three for keysym names.  A `keys` dict beside `buttons` on
    # the held record, not a `keyboard` boolean: `release_all` has to know
    # *which* keys went down and in what order.
    def press_key(self, session_id: str, key: str) -> None: ...
    def unpress_key(self, session_id: str, key: str) -> list[str]: ...
    def keys_pressed(self) -> list[str]: ...

**`expire()` carries no bookmark, no lifted buttons and no lifted keys** — it
returns the lease that was released, nothing more. The broker therefore has two
duties on expiry: read `pressed()` **and `keys_pressed()`** *before* calling
`expire()` if it wants the lease's view of what was down, and lift them on the
driver (one `release_all`), because the lease cannot lift anything itself. The
bookmark is not `expire()`'s to keep: the broker already stores it with the
session (§1), which is where it belongs.

v3 deliberately did **not** widen this return. The ordering it forces — read,
then expire — is the only thing stopping a key from being forgotten before the
driver has been told to lift it, and on this compositor a forgotten key is
worse than a forgotten button: `InputManager::onKeyboardKey` drops the *next*
press of a code it still believes is down, so the key is dead rather than
merely stuck.

def window_to_absolute(point: tuple[int, int], window: tuple[int, int, int, int],
                       monitor: tuple[int, int, int, int]) -> tuple[float, float]:
    # point is window-local pixels; window is (x, y, w, h) global; monitor is
    # (x, y, w, h) global. Returns normalised 0..1 in the MONITOR's frame,
    # clamped to [0, 1]. Raises ValueError if the point is outside the window.

3.4 src/hhstream_pointer/** (D) — the driver process

A separate process, because a Wayland client needs its own connection and a blocked driver must not stall the broker.

  • Entry point hhstream-pointer (add to pyproject.toml: hhstream-pointer = "hhstream_pointer.__main__:main").

  • Command line the module will pass, exactly (amended in v3: the four --xkb-* flags, whose defaults are the values below):

    hhstream-pointer --socket /run/hhstream/pointer.sock \
                     --compositor-runtime-dir /run/hypr-stream \
                     --xkb-layout us --xkb-model pc105 \
                     --xkb-variant '' --xkb-options ''
    

    These flags are the only path a layout has to the uploaded keymap. There is deliberately no [keyboard] table in the broker's TOML: that file is read by hhstreamd, and this is a different unit that never opens it, so such a table would be a setting an operator could change while every injected key stayed on us.

    A layout xkb_keymap_new_from_names rejects — and a comma-separated one, and an empty one — is exit 2, raised before the compositor is contacted. A keymap the compositor refuses is exit 10, and the socket is never listened on: key before keymap is the protocol error no_keymap, which kills the connection rather than returning anything reportable.

  • Uses pywayland (packaged as python3Packages.pywayland) and the wlr virtual pointer XML from pkgs.wlr-protocols, which lives at share/wlr-protocols/unstable/wlr-virtual-pointer-unstable-v1.xml (the v1 contract had this path wrong: different middle directory, and no z prefix on the filename). Vendoring the generated module is explicitly permitted (v2): generating at build time puts a code generator between a checkout and a runnable driver, and the dev shell has neither the XML nor a scanner. State which you did. pyproject.toml and nix/package.nix are yours precisely so this is possible.

  • Binds zwlr_virtual_pointer_manager_v1 version 2 and uses create_virtual_pointer_with_output(seat, output, id) so coordinates are output-local. If only version 1 is offered, report it — never silently fall back to a global pointer, which would break aiming.

  • Private socket protocol with the broker, newline-delimited JSON:

{"op": "warp",   "output": "hh-x", "x": 0.42, "y": 0.61}   -> {"ok": true}
{"op": "button", "button": "left", "state": "down"}          -> {"ok": true}
{"op": "axis",   "axis": "vertical", "value": 3.0}           -> {"ok": true}
{"op": "frame"}                                              -> {"ok": true}
{"op": "pressed"}                                            -> {"ok": true, "buttons": ["left"], "keys": ["Control_L"]}
{"op": "release_all"}                                        -> {"ok": true, "lifted": ["left"], "lifted_keys": ["Control_L"]}
{"op": "key",    "key": "Return", "state": "down"}           -> {"ok": true, "keys": ["Return"], "lifted": []}
{"op": "type",   "text": "Hello"}                            -> {"ok": true, "keys": []}

Amended in v3 — key and type, for the keyboard. Names on this socket, evdev codes only on the Wayland request: the driver owns the keymap, so it is the only thing that can say which keycode z is on the uploaded layout. Four rules that are not interchangeable:

  • No output and no frame. A virtual keyboard is seat-global, so key and type take no output and must not require a prior warp. And zwp_virtual_keyboard_v1 has no frame request — a key queued until one would never arrive — so each request is flushed the way button already is.
  • pressed and release_all grow key fields rather than gaining a parallel release_all_keys. Expiry has one call site, and a second op would be a second thing to forget in the path that runs when nobody is watching. release_all lifts buttons, then keys in press order, then sends modifiers(0, 0, 0, 0) if a mask was ever set — a virtual key-up clears no modifier state, so without that a lease that ended under Shift hands the next holder a keyboard where every letter is a capital.
  • type is one request for the whole string. The driver resolves every character before emitting anything, so an unmappable one types no prefix. Capped at 4096 Unicode scalars (not bytes) in the driver and in the broker.
  • A rejection may carry kind. {"ok": false, "error": "…", "kind": "invalid_argument"} is the caller's mistake — a keysym the layout does not have, an unmappable character, a bad state, a string over the cap — and the broker maps it to exit 2. Any other value, and an older driver's absent field, stay driver_refused (exit 10), which is what the button and axis paths already do.

Amended after the freeze — axis, for scrolling. Added to the op set above and to the broker's verbs as pointer.scroll. Positive vertical is down and positive horizontal is right: the protocol's direction, not a choice of ours, and the reason there is no up/down vocabulary layered over it. The output is whatever the last warp aimed at, exactly like button, so scrolling somewhere else is a warp first — which is what keeps the whole gesture atomic under the lease. The magnitude is bounded at 100 wheel clicks per request, in both the driver and the broker, because a decimal-point slip should be a refusal rather than a session scrolled somewhere unrecoverable. Unit tests cover validation and the lease; a real-compositor case is still missing — see docs/TODO/scroll-has-no-compositor-proof.md.

{"ok": false, "error": "<text>"} for anything rejected. The driver holds no lease state — the broker owns the lease; the driver is an executor that must never be trusted to serialise anything.

3.5 hhstreamd/live.py (v4) — the broker's end of the stream daemon

A third executor beside the pointer driver, and the same shape as §3.4: the broker holds the decision (which session streams, under which phrase) and the daemon holds the capture. hhstream-live owns the port, the bind address and the capture command, from its own unit's argv; the broker never passes them and never reads them from its own TOML.

class LiveUnavailable(Exception):
    """The stream daemon could not be reached, or did not answer intelligibly."""

class LiveRejected(Exception):
    """The daemon refused.  `kind` maps to an exit code, as the driver's does."""
    def __init__(self, message: str, *, kind: str | None = None) -> None: ...

@dataclass(frozen=True)
class Attachment:
    phrase: str; output: str; fps: int; session: str | None
    url: str                  # the daemon's own prefix, e.g. http://127.0.0.1:9988/<phrase>/
    viewers: int

class LiveStreams:
    def __init__(self, socket_path: Path | str, *, timeout_s: float = 2.0) -> None: ...
    def attach(self, *, phrase: str, output: str, fps: int,
               session: str | None = None, label: str = "") -> Attachment: ...
    def detach(self, phrase: str) -> bool: ...        # False if it was not attached
    def list(self) -> list[Attachment]: ...
    def viewers(self) -> dict[str, int]: ...          # phrase -> connected viewers
    def close(self) -> None: ...

Private socket protocol, newline-delimited JSON, the driver's conventions exactly — {"ok": true, …} or {"ok": false, "error": "…", "kind": "…"}:

{"op": "attach", "phrase": "amber-otter-lantern-harbor-7k2qp", "output": "hh-x",
 "session": "x", "label": "chromium", "fps": 5}        // fps is OPTIONAL
                      -> {"ok": true, "stream": {"phrase": "…", "output": "hh-x",
                          "fps": 5, "session": "x", "url": "http://…/", "viewers": 0}}
{"op": "detach",  "phrase": "…"}     -> {"ok": true, "detached": true}
{"op": "list"}                       -> {"ok": true, "streams": [ {…}, … ]}
{"op": "viewers"}                    -> {"ok": true, "viewers": {"<phrase>": 2}}

Four rules that are not interchangeable:

  • The daemon holds no session state and no idea of an idle clock. It publishes what it was told to publish. viewers is a reading, and what the broker does with it (§4) is the broker's.
  • fps is optional, and omitting it is the normal case. The default rate is the daemon's, from its own unit's argv (sessions.stream.fps), for the same reason the port is. The broker sends one only when a caller named one, or when re-attaching a stream that already had one. A broker that sent its own default on every attach would make the operator's option a setting that changed nothing — the same failure --xkb-layout (§3.4) exists to avoid.
  • attach is idempotent on the phrase. Re-attaching the same phrase to the same output returns the existing stream rather than starting a second capture, because a re-enable after a broker restart must not double the grim rate on one output. The same phrase aimed at a different output is {"kind": "invalid_argument"}.
  • The phrase is never logged by either side. It is the stream's only credential. The daemon logs the output and the viewer count; the broker's audit log records the session and the output (§4), and neither records the phrase.

url comes back from the daemon rather than being built by the broker, for the reason the port is the daemon's: a URL the broker assembled from a port it was told about once is a URL that silently lies the first time an operator moves the listener.



4. Wire format (broker control socket) — B, C and D do not implement this

Newline-delimited JSON over /run/hhstream/control.sock. Existing verbs are unchanged. New verbs (integration implements; listed so slices know what their code will be called with):

Verb Request (beyond verb, request_id) Response
container.list — containers[], orphans[]
container.reconcile id? container, diff
pointer.acquire id, ttl_s?, restore? lease_id, bookmark
pointer.move id, lease_id, x, y | dx, dy bookmark
pointer.button id, lease_id, button, state buttons[]
pointer.click id, lease_id, button?, count? bookmark
pointer.release id, lease_id bookmark, released_by, buttons_lifted[], keys_lifted[]
pointer.state — holder, since, lease_id, buttons[], keys[]
keyboard.key (v3) id, lease_id, key, state keys[]
keyboard.type (v3) id, lease_id, text chars, keys[]
stream.enable (v4) id, phrase?, fps? stream
stream.disable (v4) id stream, detached
stream.list (v4) — streams[], available
stream.status (v4) id? stream | streams[]

v4, the stream. stream.enable and stream.disable are in the request_id mutation set; stream.list and stream.status are not. start additionally accepts stream? (bool), stream_phrase? and stream_fps?, and when stream is true its reply's session carries the same stream object stream.enable would have returned.

The stream object is the record's, with the daemon's url and viewers merged in when it could be asked:

{"enabled": true, "phrase": "amber-otter-lantern-harbor-7k2qp",
 "output": "hh-x", "fps": 5, "port": 9988,
 "url": "http://127.0.0.1:9988/amber-otter-lantern-harbor-7k2qp/",
 "viewers": 0, "enabled_at": "2026-10-11T…Z", "attached": true}

attached is its own field and is not folded into enabled: a record may say a session is enabled while the daemon is down, and "the operator asked for this stream" and "a capture is running right now" are two different facts. A caller that wants the second one reads attached; url and viewers are absent when it is false, rather than stale.

stream.enable requires the session's container output. Without one it is stream_unavailable (exit 10). A stream never falls back to the main output: that output carries every other session, and putting it on a phone is the one mistake this feature must not make. Re-enabling an enabled session returns the same phrase; a phrase that differs from the stored one is stream_busy (exit 8). disable then enable rotates it.

Teardown order is load-bearing. kill detaches the stream before hyprctl output remove, because a capture aimed at a removed output fails every time and the daemon's own failure budget would close the viewers with "unknown output" instead of "the session ended".

v3, the keyboard. keyboard.* verbs take the pointer lease: there is no keyboard.acquire and no second lease id. One wl_seat means one keyboard focus, and the shipped input:follow_mouse = 1 means another session's warp retargets the keys, so two leases would let session B move the pointer while session A believed it was typing into its own window. Both verbs are in the request_id mutation set, for the reason pointer.click is: a retried type that typed twice is the one failure an agent cannot undo.

v3, acquire takes the keyboard too. pointer.acquire sends two warps to the driver — a neighbouring pixel, then the real target — because one warp onto an idle cursor moves the cursor and not keyboard focus. Measured; see tests/README-keyboard-probe.md. The reply gains focus, naming which mechanism took it (warp, focuswindow, or none). This depends on input:follow_mouse_threshold being 0, which nix/hyprland-conf.nix now sets explicitly rather than inheriting.

Error codes (extend ExitCode, do not renumber)

| 0 OK | 2 USAGE | 3 NO_SUCH_SESSION | 4 BAD_STATE | 5 DAEMON_UNREACHABLE | | 6 POLICY | 7 TIMEOUT | 8 BUSY | 9 NO_LEASE (new) | 10 COMPOSITOR (new) |

v4 adds error names and no numbers. stream_unavailable → 10 (no container, or the daemon is unreachable — the session itself is fine either way), stream_busy → 8 (a phrase that is not this session's), stream_budget → 6 (the daemon's viewer or stream cap). A bad fps is invalid_argument → 2, as every other bad argument is.

Activity accounting is load-bearing and has one exception as of v4: touch is the only mark of activity a caller can make; list, status, logs, gc, every pointer.* and keyboard.* verb, and every stream.* verb do not reset the idle clock. A session being typed into is still a session nobody has said they want kept.

The exception is a connected viewer. About once a minute the broker asks the daemon for viewer counts and stamps touch on every session whose stream has at least one; the audit entry names the session, the output and the count, never the phrase. This is not a caller's verb and cannot be invoked — it is the broker observing that a person is looking, which is the one form of "still wanted" that does not come down the control socket. It cannot hold a session open on its own: an enabled stream with no viewer runs no grim and stamps nothing.

The lease clock is a different one, and v3 widens it: every pointer.* and keyboard.* gesture verb refreshes the lease's idle deadline, because they all go through the same gate. pointer acquire's own stderr says so.


5. Names and paths (frozen)

Thing Value
Compositor runtime dir /run/hypr-stream — must be ≤ 25 chars (socket paths must stay under sun_path's 108-byte limit)
Compositor request socket <runtime dir>/hypr/<signature>/.socket.sock
Compositor event socket <runtime dir>/hypr/<signature>/.socket2.sock
Frame socket (published) <runtime dir>/stream.sock — default /run/hypr-stream/stream.sock. Also frozen in docs/CONTRACT.md, which v1 of this file left disagreeing with it.
Broker control socket /run/hhstream/control.sock
Broker↔driver socket /run/hhstream/pointer.sock
Broker↔stream socket /run/hhstream/live.sock (v4)
Stream port option default 9988 (sessions.stream.port), fixed and single — no range, no fallback. The daemon owns it; the broker learns the URL from the daemon (§3.5).
Stream bind option default 127.0.0.1 (sessions.stream.bind)
Broker state dir option default /var/lib/hhstream; this machine sets /var/lib/hermes/hhstream-sessions (sessions.stateDir). v1 listed the deployed value, which read as a contradiction of the module.
Session runtime dir /run/hhstream/sessions/<session-id>
Container output name hh-<session-id>
Container workspace band 1000 + slot, allocated from the registry's free list
Container mode 1280x720
Units hyprland-headless, hyprland-headless-stream, hhstream-sessions, hhstream-pointer, hhstream-live (v4)
Keyboard layout option default us (sessions.keyboard.layout), reaching the driver as --xkb-layout. One group: a comma is refused. Hyprland's input:kb_layout is not read.
Binaries hhstream, hhstreamd, hhstream-server, hhstream-client, hhstream-pointer, hhstream-live
Registry schema 4 (v4; 3 → 4 adds stream, migrating to null)
Service account hermes, supplementary group hhstream

6. Rules of engagement

  1. Only your files (§2). A shared-file change you need is a report.
  2. Unit tests only. nix develop -c python -m pytest tests/<yours> -q. Nothing that builds a VM, rebuilds the system, or touches a live service.
  3. Stdlib only, except slice D which may add pywayland.
  4. Machine style: module docstrings that say why, not what; comments carry reasoning; no TODO placeholders; no speculative generality.
  5. Definition of done: a commit on your branch; your tests pass; your final message states (a) what you built, (b) the exact test command and its result, (c) what you could not verify, (d) any contract gap you hit. A slice that "finishes" without (b) is not done.
  6. Do not open by reading the whole repo. Your brief plus this contract is your context. Read the two or three files you actually touch.
  7. If the contract is wrong, say so and stop. If your slice turns out to need another slice's file, say so and stop. Both are successful outcomes.