obsbot-mcp 0.4.1 → 0.6.0

package/dist/geometry/aim.d.ts

@@ -0,0 +1,259 @@
+import type { FovType } from "../codec/commands.js";
+export interface Frame {
+    width: number;
+    height: number;
+}
+export interface Pose {
+    yaw: number;
+    pitch: number;
+}
+export interface Optics {
+    /**
+     * Total linear magnification relative to the WIDE field, which is 1.0.
+     *
+     * One number, deliberately. The camera has a single magnification scale: the
+     * discrete FOV modes are points on it (FOV_MAGNIFICATION) and a continuous
+     * zoom writes to it directly (magnificationFromZoomRatio). Setting a zoom
+     * ratio REPLACES the mode's magnification rather than multiplying it —
+     * `narrow` plus ratio 1.5 measures 2.509, the same as `wide` plus 1.5, not
+     * 1.47 x 2.5. An earlier `{ fov, zoom }` shape made that double-counting easy
+     * to write and had to warn against it; this shape makes it unrepresentable.
+     */
+    magnification: number;
+    /**
+     * Whether the capture path horizontally flips the preview. This inverts the
+     * yaw correction, so it is an explicit input rather than a baked-in
+     * assumption. Defaults to false.
+     */
+    mirrored?: boolean;
+}
+/**
+ * Horizontal field of view of the CAPTURE STREAM for each FOV setting, in
+ * degrees — DERIVED, not three separate measurements.
+ *
+ * Earlier revisions carried wide/medium/narrow as three independently measured
+ * absolutes at +/-3 degrees each. That threw away the most precise thing known
+ * about them: the RATIOS between the modes are measured to ~0.05%, roughly 60x
+ * better than any of the absolutes, so three free values let the modes drift out
+ * of proportion with each other for no reason.
+ *
+ * One anchor plus measured magnifications instead. The anchor's uncertainty
+ * still propagates, but it now moves all three together, which is the honest
+ * representation of what is known.
+ *
+ * MEASURED 2026-07-25 by solving the camera intrinsics from pure gimbal
+ * rotations. A camera that only rotates induces an exact homography
+ * H = K R K^-1 between views, with no dependence on scene depth, so frames at
+ * known gimbal angles determine the focal lengths outright. No distance is
+ * measured anywhere — the gimbal angle is the ruler. That is what makes this
+ * tighter than the tape-measured letter sheet behind the old +/-3 degrees.
+ *
+ * Six rotations (pitch +/-10, +/-20; yaw +/-10) over a static scene, 313-1243
+ * inliers each, gave fx = 1452-1455 px on a 1920-wide frame across every subset
+ * — a spread of 0.2% — which is HFOV 66.84-66.90. Rounded to 67. This also
+ * independently reproduces an earlier pan-and-track measurement of 66.4.
+ *
+ * The per-mode magnifications come from fitting a similarity transform between
+ * frames at each setting on a fixed scene (275 and 683 inliers, 0.48 and 0.44 px
+ * residual). They are pure ratios, so they are unaffected by which capture
+ * format the measurement went through — which matters, because the 1080p pixel
+ * formats do NOT share a field of view (see the note on the capture path below).
+ *
+ * Verified head-to-head on hardware, same feature and same start pose with only
+ * the constant differing: a target at u = +0.91 left a yaw residual of -0.823
+ * degrees under the old 68 and -0.274 under 67.
+ *
+ * CAPTURE FORMAT WARNING: at 1920x1080 this camera has two different windows
+ * onto the sensor, and FRAME RATE selects between them — not the codec.
+ * MEASURED 2026-07-25 at one pose and one zoom: MJPEG@30 vs YUYV@30 came out at
+ * scale 1.00001, t (0.2, 0.0) over 2382 inliers at 0.12 px, i.e. the same field
+ * to within a fifth of a pixel; MJPEG@60 is a 1.214x crop of BOTH (1.21422 and
+ * 1.21404). An earlier revision recorded this as "MJPEG is a 1.201x crop of
+ * YUYV" — that comparison was MJPEG@60 against YUYV@30 and charged the codec
+ * for what the frame rate did.
+ *
+ * These constants describe the WIDE (30fps) field. That is what
+ * `obsbot_capture_snapshot` delivers: its graph negotiates MJPG 1920x1080@30,
+ * which the reply now states outright in `sourceFormat`.
+ * `obsbot_capture_preview` pins 60fps to buy smooth motion and therefore shows
+ * ~21% less — so preview pixels are NOT interchangeable with snapshot pixels
+ * for aiming.
+ *
+ * Any future measurement must state pixel format AND frame rate; neither a
+ * resolution nor a codec alone identifies the field.
+ *
+ * SCOPE: 16:9 capture at any resolution. A 4:3 path would need re-measuring.
+ */
+export declare const WIDE_HFOV_DEG = 67;
+/**
+ * Linear magnification of each FOV setting relative to the wide field.
+ *
+ * MEASURED 2026-07-25. These are the precise part of the pair: the ratios are
+ * good to ~0.05% where the anchor above is good to perhaps 0.5%.
+ *
+ * The continuous zoom writes to this same scale rather than multiplying on top
+ * of it — `narrow` plus zoom ratio 1.5 measures 2.509, the same as `wide` plus
+ * 1.5 (2.501), not 1.47 x 2.5. The discrete modes and the zoom control are two
+ * ways of writing to one magnification scale, which is why setting zoom ratio
+ * 1.0 does not reliably clear `custom`: it is the same optical state as `wide`.
+ */
+export declare const FOV_MAGNIFICATION: Record<FovType, number>;
+export declare const HORIZONTAL_FOV_DEG: Record<FovType, number>;
+/**
+ * Empirical correction applied on top of the aspect-derived vertical half-angle.
+ * MEASURED, not geometric.
+ *
+ * Square-pixel geometry says tan(V) = tan(H) * (height/width) — 0.5625 at 16:9.
+ * Hardware says the vertical field is shorter than that; this factor carries the
+ * difference.
+ *
+ * MEASURED 2026-07-25 from the same intrinsics solve as WIDE_HFOV_DEG above.
+ * fy came out 1502-1520 px across every subset of six rotations, implying this
+ * factor at 0.957-0.967. Rounded to 0.957.
+ *
+ * This REPLACES an earlier value of 0.898, which was 7% low. That figure came
+ * from a measurement this project's own history recorded as inconclusive: the
+ * constant itself was off by about 6.2%, versus the ~4.3% effect it was trying
+ * to capture — roughly 1.4x the effect itself, not an order of magnitude off.
+ *
+ * The up/down asymmetry is ~1%, not the ~5% once believed: solving from the
+ * up-tilt alone gives 0.957 and from the down-tilt alone 0.967. One constant
+ * captures that comfortably. Do NOT reintroduce a two-branch vertical constant
+ * on the strength of the old figure. Both candidate explanations for a genuine
+ * asymmetry were tested and eliminated — the principal point is centred (cx, cy
+ * within a few px of frame centre in every solve) and radial distortion is
+ * negligible (k1 ~= -0.02).
+ *
+ * Verified head-to-head on hardware, same feature and same start pose with only
+ * this constant differing: a target at v = -0.83 left a pitch residual of -0.597
+ * degrees under 0.898 and +0.072 under 0.957, an 8x improvement that lands
+ * inside the noise.
+ *
+ * Known limit: the intrinsics fit carries 2.4-2.8 px rms, not sub-pixel, so
+ * something is unmodelled — most likely the entrance pupil sitting off the
+ * gimbal's rotation axes, which translates the lens as it turns and is
+ * depth-dependent. Sampling was symmetric so it should not bias fx or fy, but do
+ * not claim more precision than that.
+ *
+ * SCOPE: measured on the 16:9 capture path. A 4:3 path needs its own measurement.
+ */
+export declare const VERTICAL_TANGENT_CORRECTION = 0.957;
+/** Magnification of the wide field, and of the whole scale, at its extremes. */
+export declare const MIN_MAGNIFICATION = 1;
+export declare const MAX_MAGNIFICATION = 4;
+/**
+ * Linear magnification for a UVC zoom ratio. MEASURED 2026-07-25: magnification
+ * is linear in the ratio, `m = 3r - 2`, holding to better than 0.05% at ratios
+ * 1.25, 1.5 and 2.0. So ratio 2.0 is 4x linear — carried for a long time as an
+ * unsourced note, now measured to four figures. See the spec's section 1.1.
+ */
+export declare const magnificationFromZoomRatio: (ratio: number) => number;
+/** Inverse of {@link magnificationFromZoomRatio}: the ratio that yields `m`. */
+export declare const zoomRatioFromMagnification: (m: number) => number;
+/** Effective half-angles of the visible field, in degrees, after zoom and aspect. */
+export declare function halfAngles(optics: Optics, frame: Frame): {
+    h: number;
+    v: number;
+};
+export interface Offset {
+    dYaw: number;
+    dPitch: number;
+}
+/**
+ * Convert a pixel in a captured frame to the yaw/pitch delta that would bring
+ * that point to the center of frame, treating each axis independently.
+ *
+ * PER-AXIS ONLY — this does NOT account for axis coupling. The gimbal's yaw
+ * axis is world-vertical, so yawing while pitched sweeps a cone; the two axes
+ * commute only at zero pitch or zero horizontal offset (see `aimAtPixel`,
+ * which composes a rotation instead and is what every tool actually points
+ * the camera with). Adding this function's dYaw/dPitch to a pose that is not
+ * level reproduces the additive bug `aimAtPixel` was fixed to remove. It is
+ * used only by tests now, to pin the per-axis tangent mapping in isolation
+ * from the coupling — do not reach for it as a general "point the camera"
+ * answer.
+ *
+ * A rectilinear lens maps angle through a tangent: tan(theta) = u * tan(hfov/2),
+ * where u is the normalized offset from center. The linear approximation is
+ * exact at the center and again at the edge, and wrong in between — always low,
+ * peaking near 1.58 degrees at u ~= 0.55 on the wide setting (WIDE_HFOV_DEG =
+ * 67). That error is the difference between landing on target and visibly
+ * hunting, so the tangent form is not optional.
+ *
+ * Signs: +yaw pans camera-LEFT and image x grows rightward, so the yaw term is
+ * negated. +pitch tilts DOWN and image y grows downward, so the pitch term is
+ * not. Both conventions are hardware-verified.
+ */
+export declare function pixelToOffset(x: number, y: number, frame: Frame, optics: Optics): Offset;
+/**
+ * Mechanical limits of the Tiny 2's gimbal, in degrees. Hardware-verified.
+ *
+ * These live here rather than in the tool layer so there is exactly one
+ * definition: `obsbot_gimbal_move` imports them for its own clamping. Two copies
+ * of a bound that must agree is a defect waiting to happen.
+ *
+ * ±150 is the mechanical yaw range on EVERY platform, and these limits are not
+ * platform-conditional. Measured 2026-07-25: commanded 145 reads back 145, and
+ * 150 reads back 149. Position feedback comes from the camera's physical
+ * encoder, which is a property of the hardware and does not vary by OS.
+ *
+ * Do not "fix" this to 130. `transport/linux.ts` and `transport/macos.ts` record
+ * a `CT_PANTILT_ABSOLUTE` range of ±468000 arcsec = ±130°, which reads like a
+ * conflict and was raised as one during review. It is not: that is the range the
+ * UVC control *advertises* — a descriptor value that under-reports the mechanism
+ * it describes — not where the gimbal stops. Nothing clamps to it either;
+ * `LinuxTransport.gimbalSet` writes `yawDeg * ARCSEC_PER_DEG` unclamped, so the
+ * arcsec figure lives only in comments. See the spec's §8.
+ */
+export declare const GIMBAL_YAW_LIMIT_DEG = 150;
+export declare const GIMBAL_PITCH_LIMIT_DEG = 90;
+export interface Aim {
+    target: Pose;
+    /**
+     * The rotation REQUESTED to reach the raw, unclamped target —
+     * `rawTarget - current` — not the rotation actually applied. When `clamped`
+     * is true, `target` stops short of `current + offset`; `offset` still
+     * reports what was asked for, not what the gimbal was told to do.
+     */
+    offset: Offset;
+    /** True if either axis saturated, meaning the target was not reachable. */
+    clamped: boolean;
+    /**
+     * True when the target ray's horizontal direction points opposite the
+     * camera's current heading — an "over the top" solution. The pixel lies
+     * past vertical from the current pose and is not reachable by any rotation
+     * that keeps the image upright; clamping it to the nearest yaw would slew
+     * the camera toward the far side of the room, not toward the target.
+     */
+    overTheTop: boolean;
+}
+/**
+ * Absolute pose that brings the given pixel to the centre of frame.
+ *
+ * Composes a rotation rather than adding two scalars. The gimbal's yaw axis is
+ * world-vertical, so yawing while pitched sweeps a CONE: the two rotations do
+ * not commute, and `target = current + offset` is exact only at zero pitch or
+ * zero horizontal offset. Adding them cost 0.98 degrees at pitch 7.6 / yaw 31 on
+ * hardware, against `pitch * (1 - cos yaw)` = 1.09 predicted.
+ *
+ * The ray to the target in camera coordinates (x right, y down, z forward) is
+ * rotated into world coordinates by the current orientation, and the target pose
+ * is read back off that direction. The two axes reduce to the old additive sum
+ * under DIFFERENT conditions, not the same one: yaw reduces exactly to the old
+ * sum whenever pitch = 0, for any pixel, because yaw and pitch commute at zero
+ * tilt. Pitch reduces to the old sum only when u = 0 (the pixel sits on the
+ * vertical centre line) — composed pitch is asin(dy/n) where n includes dx, so
+ * away from u = 0 it differs from the old atan(dy) even when pitch is 0. That is
+ * what the invariant tests pin.
+ *
+ * Saturation is REPORTED, not silent: if the target lies outside the gimbal's
+ * range the caller has to know it landed short rather than assume the aim
+ * succeeded, since a silent clamp presents as "the camera aimed and missed".
+ *
+ * `current` must be where the camera actually was when the frame was captured.
+ * The module cannot verify that — see the spec's section 5 for what breaks it.
+ * Note that on Windows the pose the caller reads is FLOORED to whole degrees
+ * (spec section 4.2), which costs up to another degree; that is a separate
+ * defect in the pose source, not in this function.
+ */
+export declare function aimAtPixel(x: number, y: number, frame: Frame, optics: Optics, current: Pose): Aim;

package/dist/geometry/aim.js

@@ -0,0 +1,317 @@
+// Pure geometry for aiming the gimbal at a point seen in a snapshot. No I/O, no
+// transport, no device access — everything here is a function of its arguments,
+// which is what makes it testable with no camera attached.
+//
+// Degrees in the public API (matching every other angle in this codebase),
+// radians internal only.
+const toRad = (deg) => (deg * Math.PI) / 180;
+const toDeg = (rad) => (rad * 180) / Math.PI;
+/**
+ * Horizontal field of view of the CAPTURE STREAM for each FOV setting, in
+ * degrees — DERIVED, not three separate measurements.
+ *
+ * Earlier revisions carried wide/medium/narrow as three independently measured
+ * absolutes at +/-3 degrees each. That threw away the most precise thing known
+ * about them: the RATIOS between the modes are measured to ~0.05%, roughly 60x
+ * better than any of the absolutes, so three free values let the modes drift out
+ * of proportion with each other for no reason.
+ *
+ * One anchor plus measured magnifications instead. The anchor's uncertainty
+ * still propagates, but it now moves all three together, which is the honest
+ * representation of what is known.
+ *
+ * MEASURED 2026-07-25 by solving the camera intrinsics from pure gimbal
+ * rotations. A camera that only rotates induces an exact homography
+ * H = K R K^-1 between views, with no dependence on scene depth, so frames at
+ * known gimbal angles determine the focal lengths outright. No distance is
+ * measured anywhere — the gimbal angle is the ruler. That is what makes this
+ * tighter than the tape-measured letter sheet behind the old +/-3 degrees.
+ *
+ * Six rotations (pitch +/-10, +/-20; yaw +/-10) over a static scene, 313-1243
+ * inliers each, gave fx = 1452-1455 px on a 1920-wide frame across every subset
+ * — a spread of 0.2% — which is HFOV 66.84-66.90. Rounded to 67. This also
+ * independently reproduces an earlier pan-and-track measurement of 66.4.
+ *
+ * The per-mode magnifications come from fitting a similarity transform between
+ * frames at each setting on a fixed scene (275 and 683 inliers, 0.48 and 0.44 px
+ * residual). They are pure ratios, so they are unaffected by which capture
+ * format the measurement went through — which matters, because the 1080p pixel
+ * formats do NOT share a field of view (see the note on the capture path below).
+ *
+ * Verified head-to-head on hardware, same feature and same start pose with only
+ * the constant differing: a target at u = +0.91 left a yaw residual of -0.823
+ * degrees under the old 68 and -0.274 under 67.
+ *
+ * CAPTURE FORMAT WARNING: at 1920x1080 this camera has two different windows
+ * onto the sensor, and FRAME RATE selects between them — not the codec.
+ * MEASURED 2026-07-25 at one pose and one zoom: MJPEG@30 vs YUYV@30 came out at
+ * scale 1.00001, t (0.2, 0.0) over 2382 inliers at 0.12 px, i.e. the same field
+ * to within a fifth of a pixel; MJPEG@60 is a 1.214x crop of BOTH (1.21422 and
+ * 1.21404). An earlier revision recorded this as "MJPEG is a 1.201x crop of
+ * YUYV" — that comparison was MJPEG@60 against YUYV@30 and charged the codec
+ * for what the frame rate did.
+ *
+ * These constants describe the WIDE (30fps) field. That is what
+ * `obsbot_capture_snapshot` delivers: its graph negotiates MJPG 1920x1080@30,
+ * which the reply now states outright in `sourceFormat`.
+ * `obsbot_capture_preview` pins 60fps to buy smooth motion and therefore shows
+ * ~21% less — so preview pixels are NOT interchangeable with snapshot pixels
+ * for aiming.
+ *
+ * Any future measurement must state pixel format AND frame rate; neither a
+ * resolution nor a codec alone identifies the field.
+ *
+ * SCOPE: 16:9 capture at any resolution. A 4:3 path would need re-measuring.
+ */
+export const WIDE_HFOV_DEG = 67;
+/**
+ * Linear magnification of each FOV setting relative to the wide field.
+ *
+ * MEASURED 2026-07-25. These are the precise part of the pair: the ratios are
+ * good to ~0.05% where the anchor above is good to perhaps 0.5%.
+ *
+ * The continuous zoom writes to this same scale rather than multiplying on top
+ * of it — `narrow` plus zoom ratio 1.5 measures 2.509, the same as `wide` plus
+ * 1.5 (2.501), not 1.47 x 2.5. The discrete modes and the zoom control are two
+ * ways of writing to one magnification scale, which is why setting zoom ratio
+ * 1.0 does not reliably clear `custom`: it is the same optical state as `wide`.
+ */
+export const FOV_MAGNIFICATION = {
+    wide: 1,
+    medium: 1.15060,
+    narrow: 1.47073,
+};
+export const HORIZONTAL_FOV_DEG = {
+    wide: 2 * toDeg(Math.atan(Math.tan(toRad(WIDE_HFOV_DEG / 2)) / FOV_MAGNIFICATION.wide)),
+    medium: 2 * toDeg(Math.atan(Math.tan(toRad(WIDE_HFOV_DEG / 2)) / FOV_MAGNIFICATION.medium)),
+    narrow: 2 * toDeg(Math.atan(Math.tan(toRad(WIDE_HFOV_DEG / 2)) / FOV_MAGNIFICATION.narrow)),
+};
+/**
+ * Empirical correction applied on top of the aspect-derived vertical half-angle.
+ * MEASURED, not geometric.
+ *
+ * Square-pixel geometry says tan(V) = tan(H) * (height/width) — 0.5625 at 16:9.
+ * Hardware says the vertical field is shorter than that; this factor carries the
+ * difference.
+ *
+ * MEASURED 2026-07-25 from the same intrinsics solve as WIDE_HFOV_DEG above.
+ * fy came out 1502-1520 px across every subset of six rotations, implying this
+ * factor at 0.957-0.967. Rounded to 0.957.
+ *
+ * This REPLACES an earlier value of 0.898, which was 7% low. That figure came
+ * from a measurement this project's own history recorded as inconclusive: the
+ * constant itself was off by about 6.2%, versus the ~4.3% effect it was trying
+ * to capture — roughly 1.4x the effect itself, not an order of magnitude off.
+ *
+ * The up/down asymmetry is ~1%, not the ~5% once believed: solving from the
+ * up-tilt alone gives 0.957 and from the down-tilt alone 0.967. One constant
+ * captures that comfortably. Do NOT reintroduce a two-branch vertical constant
+ * on the strength of the old figure. Both candidate explanations for a genuine
+ * asymmetry were tested and eliminated — the principal point is centred (cx, cy
+ * within a few px of frame centre in every solve) and radial distortion is
+ * negligible (k1 ~= -0.02).
+ *
+ * Verified head-to-head on hardware, same feature and same start pose with only
+ * this constant differing: a target at v = -0.83 left a pitch residual of -0.597
+ * degrees under 0.898 and +0.072 under 0.957, an 8x improvement that lands
+ * inside the noise.
+ *
+ * Known limit: the intrinsics fit carries 2.4-2.8 px rms, not sub-pixel, so
+ * something is unmodelled — most likely the entrance pupil sitting off the
+ * gimbal's rotation axes, which translates the lens as it turns and is
+ * depth-dependent. Sampling was symmetric so it should not bias fx or fy, but do
+ * not claim more precision than that.
+ *
+ * SCOPE: measured on the 16:9 capture path. A 4:3 path needs its own measurement.
+ */
+export const VERTICAL_TANGENT_CORRECTION = 0.957;
+/** Magnification of the wide field, and of the whole scale, at its extremes. */
+export const MIN_MAGNIFICATION = 1;
+export const MAX_MAGNIFICATION = 4;
+/**
+ * Linear magnification for a UVC zoom ratio. MEASURED 2026-07-25: magnification
+ * is linear in the ratio, `m = 3r - 2`, holding to better than 0.05% at ratios
+ * 1.25, 1.5 and 2.0. So ratio 2.0 is 4x linear — carried for a long time as an
+ * unsourced note, now measured to four figures. See the spec's section 1.1.
+ */
+export const magnificationFromZoomRatio = (ratio) => 3 * ratio - 2;
+/** Inverse of {@link magnificationFromZoomRatio}: the ratio that yields `m`. */
+export const zoomRatioFromMagnification = (m) => (m + 2) / 3;
+// The tangents are the useful form for every downstream calculation, so they are
+// computed once here and the degree-valued halfAngles() is a thin wrapper. Going
+// through degrees would mean an atan followed immediately by a tan.
+//
+// This is also the ONE place every public entry point (halfAngles, pixelToOffset,
+// aimAtPixel) funnels through, which is why the magnification bound is enforced
+// here rather than trusted from the caller. A magnification of 0 divides out to
+// tanH = Infinity, which atan silently resolves to a 90-degree half-angle; a
+// negative value flips the half-angle's sign with no error; NaN propagates all
+// the way through aimAtPixel's output. Before this guard MIN_MAGNIFICATION/
+// MAX_MAGNIFICATION were exported but enforced nowhere — decorative constants
+// that a malformed device reading (this module's callers now derive magnification
+// from a reported zoomPercent) could silently violate. Callers that already
+// validate their own input (resolveMagnification in src/mcp/tools.ts) should
+// never actually trip this; it exists as the backstop for whichever caller
+// doesn't.
+const halfAngleTangents = (optics, frame) => {
+    if (!Number.isFinite(optics.magnification) ||
+        optics.magnification < MIN_MAGNIFICATION ||
+        optics.magnification > MAX_MAGNIFICATION) {
+        throw new RangeError(`optics.magnification must be finite and within [${MIN_MAGNIFICATION}, ${MAX_MAGNIFICATION}], got ${optics.magnification}`);
+    }
+    const tanH = Math.tan(toRad(WIDE_HFOV_DEG / 2)) / optics.magnification;
+    // Vertical starts from the horizontal half-angle scaled by the frame aspect —
+    // what square-pixel geometry predicts — then takes the measured correction,
+    // because hardware says the real vertical field is ~4.3% shorter than that.
+    // See VERTICAL_TANGENT_CORRECTION for the intrinsics solve over six gimbal
+    // rotations behind it.
+    return { tanH, tanV: tanH * (frame.height / frame.width) * VERTICAL_TANGENT_CORRECTION };
+};
+/** Effective half-angles of the visible field, in degrees, after zoom and aspect. */
+export function halfAngles(optics, frame) {
+    const { tanH, tanV } = halfAngleTangents(optics, frame);
+    return { h: toDeg(Math.atan(tanH)), v: toDeg(Math.atan(tanV)) };
+}
+/**
+ * Convert a pixel in a captured frame to the yaw/pitch delta that would bring
+ * that point to the center of frame, treating each axis independently.
+ *
+ * PER-AXIS ONLY — this does NOT account for axis coupling. The gimbal's yaw
+ * axis is world-vertical, so yawing while pitched sweeps a cone; the two axes
+ * commute only at zero pitch or zero horizontal offset (see `aimAtPixel`,
+ * which composes a rotation instead and is what every tool actually points
+ * the camera with). Adding this function's dYaw/dPitch to a pose that is not
+ * level reproduces the additive bug `aimAtPixel` was fixed to remove. It is
+ * used only by tests now, to pin the per-axis tangent mapping in isolation
+ * from the coupling — do not reach for it as a general "point the camera"
+ * answer.
+ *
+ * A rectilinear lens maps angle through a tangent: tan(theta) = u * tan(hfov/2),
+ * where u is the normalized offset from center. The linear approximation is
+ * exact at the center and again at the edge, and wrong in between — always low,
+ * peaking near 1.58 degrees at u ~= 0.55 on the wide setting (WIDE_HFOV_DEG =
+ * 67). That error is the difference between landing on target and visibly
+ * hunting, so the tangent form is not optional.
+ *
+ * Signs: +yaw pans camera-LEFT and image x grows rightward, so the yaw term is
+ * negated. +pitch tilts DOWN and image y grows downward, so the pitch term is
+ * not. Both conventions are hardware-verified.
+ */
+export function pixelToOffset(x, y, frame, optics) {
+    const { tanH, tanV } = halfAngleTangents(optics, frame);
+    const u = (2 * x) / frame.width - 1;
+    const v = (2 * y) / frame.height - 1;
+    const uEff = optics.mirrored ? -u : u;
+    return {
+        dYaw: -toDeg(Math.atan(uEff * tanH)),
+        dPitch: toDeg(Math.atan(v * tanV)),
+    };
+}
+/**
+ * Mechanical limits of the Tiny 2's gimbal, in degrees. Hardware-verified.
+ *
+ * These live here rather than in the tool layer so there is exactly one
+ * definition: `obsbot_gimbal_move` imports them for its own clamping. Two copies
+ * of a bound that must agree is a defect waiting to happen.
+ *
+ * ±150 is the mechanical yaw range on EVERY platform, and these limits are not
+ * platform-conditional. Measured 2026-07-25: commanded 145 reads back 145, and
+ * 150 reads back 149. Position feedback comes from the camera's physical
+ * encoder, which is a property of the hardware and does not vary by OS.
+ *
+ * Do not "fix" this to 130. `transport/linux.ts` and `transport/macos.ts` record
+ * a `CT_PANTILT_ABSOLUTE` range of ±468000 arcsec = ±130°, which reads like a
+ * conflict and was raised as one during review. It is not: that is the range the
+ * UVC control *advertises* — a descriptor value that under-reports the mechanism
+ * it describes — not where the gimbal stops. Nothing clamps to it either;
+ * `LinuxTransport.gimbalSet` writes `yawDeg * ARCSEC_PER_DEG` unclamped, so the
+ * arcsec figure lives only in comments. See the spec's §8.
+ */
+export const GIMBAL_YAW_LIMIT_DEG = 150;
+export const GIMBAL_PITCH_LIMIT_DEG = 90;
+const clampTo = (value, limit) => Math.min(limit, Math.max(-limit, value));
+/**
+ * Absolute pose that brings the given pixel to the centre of frame.
+ *
+ * Composes a rotation rather than adding two scalars. The gimbal's yaw axis is
+ * world-vertical, so yawing while pitched sweeps a CONE: the two rotations do
+ * not commute, and `target = current + offset` is exact only at zero pitch or
+ * zero horizontal offset. Adding them cost 0.98 degrees at pitch 7.6 / yaw 31 on
+ * hardware, against `pitch * (1 - cos yaw)` = 1.09 predicted.
+ *
+ * The ray to the target in camera coordinates (x right, y down, z forward) is
+ * rotated into world coordinates by the current orientation, and the target pose
+ * is read back off that direction. The two axes reduce to the old additive sum
+ * under DIFFERENT conditions, not the same one: yaw reduces exactly to the old
+ * sum whenever pitch = 0, for any pixel, because yaw and pitch commute at zero
+ * tilt. Pitch reduces to the old sum only when u = 0 (the pixel sits on the
+ * vertical centre line) — composed pitch is asin(dy/n) where n includes dx, so
+ * away from u = 0 it differs from the old atan(dy) even when pitch is 0. That is
+ * what the invariant tests pin.
+ *
+ * Saturation is REPORTED, not silent: if the target lies outside the gimbal's
+ * range the caller has to know it landed short rather than assume the aim
+ * succeeded, since a silent clamp presents as "the camera aimed and missed".
+ *
+ * `current` must be where the camera actually was when the frame was captured.
+ * The module cannot verify that — see the spec's section 5 for what breaks it.
+ * Note that on Windows the pose the caller reads is FLOORED to whole degrees
+ * (spec section 4.2), which costs up to another degree; that is a separate
+ * defect in the pose source, not in this function.
+ */
+export function aimAtPixel(x, y, frame, optics, current) {
+    const { tanH, tanV } = halfAngleTangents(optics, frame);
+    const u = (2 * x) / frame.width - 1;
+    const v = (2 * y) / frame.height - 1;
+    const uEff = optics.mirrored ? -u : u;
+    // Ray to the target, in camera coordinates: x right, y down, z forward.
+    const dx = uEff * tanH;
+    const dy = v * tanV;
+    const dz = 1;
+    const n = Math.sqrt(dx * dx + dy * dy + dz * dz);
+    const cy = Math.cos(toRad(current.yaw));
+    const sy = Math.sin(toRad(current.yaw));
+    const cp = Math.cos(toRad(current.pitch));
+    const sp = Math.sin(toRad(current.pitch));
+    // Pitch first, about the camera's own x-axis (+pitch tilts DOWN)...
+    const px = dx / n;
+    const py = (dy * cp + dz * sp) / n;
+    const pz = (-dy * sp + dz * cp) / n;
+    // ...then yaw, about the world vertical (+yaw pans camera-LEFT).
+    const wx = px * cy - pz * sy;
+    const wy = py;
+    const wz = px * sy + pz * cy;
+    const rawPitch = toDeg(Math.asin(Math.max(-1, Math.min(1, wy))));
+    // atan2 returns [-180, 180] — -180 IS included, and is exactly the case the
+    // following paragraph handles. Pick the representative nearest the current
+    // yaw so a target past +150 reads as +183 rather than -177. Without this a
+    // saturating aim clamps to the WRONG END of the range.
+    //
+    // This is a modulo wrap rather than `Math.round((current.yaw - rawYaw) /
+    // 360)`: at an exact 180-degree difference — dead behind the camera, which
+    // happens for real at u = 0 aiming past vertical — that rounds ties toward
+    // +Infinity regardless of sign (JS's Math.round(0.5) is 1, not 0), which
+    // silently flips -180 to +180 and clamps on the wrong end. The wrap below
+    // always resolves an exact tie to the lower edge of the window instead.
+    const rawYawAtan2 = toDeg(Math.atan2(-wx, wz));
+    const rawYaw = current.yaw + (((rawYawAtan2 - current.yaw + 180) % 360 + 360) % 360 - 180);
+    const yaw = clampTo(rawYaw, GIMBAL_YAW_LIMIT_DEG);
+    const pitch = clampTo(rawPitch, GIMBAL_PITCH_LIMIT_DEG);
+    // Over the top: the target ray's horizontal direction points opposite the
+    // camera's current heading. (hx, hz) is that heading; (wx, wz) is the target
+    // direction, both already computed above. A negative dot product means the
+    // only way to face the target while keeping the image upright is to go past
+    // vertical — clamping the resulting yaw to the nearest limit would slew the
+    // camera toward the OPPOSITE side of the room, not toward the target, so the
+    // caller must refuse this case rather than clamp-and-move it.
+    const hx = -sy;
+    const hz = cy;
+    const overTheTop = hx * wx + hz * wz < 0;
+    return {
+        target: { yaw, pitch },
+        offset: { dYaw: rawYaw - current.yaw, dPitch: rawPitch - current.pitch },
+        clamped: yaw !== rawYaw || pitch !== rawPitch,
+        overTheTop,
+    };
+}
+//# sourceMappingURL=aim.js.map

package/dist/geometry/aim.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"aim.js","sourceRoot":"","sources":["../../src/geometry/aim.ts"],"names":[],"mappings":"AAAA,gFAAgF;AAChF,gFAAgF;AAChF,2DAA2D;AAC3D,EAAE;AACF,2EAA2E;AAC3E,yBAAyB;AAmCzB,MAAM,KAAK,GAAG,CAAC,GAAW,EAAU,EAAE,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC;AAC7D,MAAM,KAAK,GAAG,CAAC,GAAW,EAAU,EAAE,CAAC,CAAC,GAAG,GAAG,GAAG,CAAC,GAAG,IAAI,CAAC,EAAE,CAAC;AAE7D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAwDG;AACH,MAAM,CAAC,MAAM,aAAa,GAAG,EAAE,CAAC;AAEhC;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAA4B;IACxD,IAAI,EAAE,CAAC;IACP,MAAM,EAAE,OAAO;IACf,MAAM,EAAE,OAAO;CAChB,CAAC;AAEF,MAAM,CAAC,MAAM,kBAAkB,GAA4B;IACzD,IAAI,EAAE,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvF,MAAM,EAAE,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,GAAG,iBAAiB,CAAC,MAAM,CAAC,CAAC;IAC3F,MAAM,EAAE,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,GAAG,iBAAiB,CAAC,MAAM,CAAC,CAAC;CAC5F,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqCG;AACH,MAAM,CAAC,MAAM,2BAA2B,GAAG,KAAK,CAAC;AAEjD,gFAAgF;AAChF,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC;AACnC,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAEnC;;;;;GAKG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAAG,CAAC,KAAa,EAAU,EAAE,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC;AAEnF,gFAAgF;AAChF,MAAM,CAAC,MAAM,0BAA0B,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;AAE7E,iFAAiF;AACjF,iFAAiF;AACjF,oEAAoE;AACpE,EAAE;AACF,kFAAkF;AAClF,gFAAgF;AAChF,gFAAgF;AAChF,6EAA6E;AAC7E,+EAA+E;AAC/E,4EAA4E;AAC5E,8EAA8E;AAC9E,kFAAkF;AAClF,4EAA4E;AAC5E,6EAA6E;AAC7E,2EAA2E;AAC3E,WAAW;AACX,MAAM,iBAAiB,GAAG,CAAC,MAAc,EAAE,KAAY,EAAkC,EAAE;IACzF,IACE,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,aAAa,CAAC;QACtC,MAAM,CAAC,aAAa,GAAG,iBAAiB;QACxC,MAAM,CAAC,aAAa,GAAG,iBAAiB,EACxC,CAAC;QACD,MAAM,IAAI,UAAU,CAClB,mDAAmD,iBAAiB,KAAK,iBAAiB,UAAU,MAAM,CAAC,aAAa,EAAE,CAC3H,CAAC;IACJ,CAAC;IACD,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,aAAa,CAAC;IACvE,8EAA8E;IAC9E,4EAA4E;IAC5E,4EAA4E;IAC5E,2EAA2E;IAC3E,uBAAuB;IACvB,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,GAAG,CAAC,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,2BAA2B,EAAE,CAAC;AAC3F,CAAC,CAAC;AAEF,qFAAqF;AACrF,MAAM,UAAU,UAAU,CAAC,MAAc,EAAE,KAAY;IACrD,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,GAAG,iBAAiB,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IACxD,OAAO,EAAE,CAAC,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;AAClE,CAAC;AAOD;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,MAAM,UAAU,aAAa,CAAC,CAAS,EAAE,CAAS,EAAE,KAAY,EAAE,MAAc;IAC9E,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,GAAG,iBAAiB,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IACxD,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;IACpC,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;IACrC,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACtC,OAAO;QACL,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC;QACpC,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;KACnC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAG,GAAG,CAAC;AACxC,MAAM,CAAC,MAAM,sBAAsB,GAAG,EAAE,CAAC;AAuBzC,MAAM,OAAO,GAAG,CAAC,KAAa,EAAE,KAAa,EAAU,EAAE,CACvD,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC;AAE3C;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,MAAM,UAAU,UAAU,CACxB,CAAS,EACT,CAAS,EACT,KAAY,EACZ,MAAc,EACd,OAAa;IAEb,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,GAAG,iBAAiB,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IACxD,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;IACpC,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;IACrC,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAEtC,wEAAwE;IACxE,MAAM,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;IACvB,MAAM,EAAE,GAAG,CAAC,GAAG,IAAI,CAAC;IACpB,MAAM,EAAE,GAAG,CAAC,CAAC;IACb,MAAM,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC,CAAC;IAEjD,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;IACxC,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;IACxC,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC;IAC1C,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC;IAE1C,oEAAoE;IACpE,MAAM,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;IAClB,MAAM,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC;IACnC,MAAM,EAAE,GAAG,CAAC,CAAC,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC;IACpC,iEAAiE;IACjE,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;IAC7B,MAAM,EAAE,GAAG,EAAE,CAAC;IACd,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;IAE7B,MAAM,QAAQ,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACjE,4EAA4E;IAC5E,2EAA2E;IAC3E,2EAA2E;IAC3E,uDAAuD;IACvD,EAAE;IACF,yEAAyE;IACzE,2EAA2E;IAC3E,2EAA2E;IAC3E,yEAAyE;IACzE,0EAA0E;IAC1E,wEAAwE;IACxE,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;IAC/C,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,WAAW,GAAG,OAAO,CAAC,GAAG,GAAG,GAAG,CAAC,GAAG,GAAG,GAAG,GAAG,CAAC,GAAG,GAAG,GAAG,GAAG,CAAC,CAAC;IAE3F,MAAM,GAAG,GAAG,OAAO,CAAC,MAAM,EAAE,oBAAoB,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,EAAE,sBAAsB,CAAC,CAAC;IAExD,0EAA0E;IAC1E,6EAA6E;IAC7E,2EAA2E;IAC3E,4EAA4E;IAC5E,4EAA4E;IAC5E,6EAA6E;IAC7E,8DAA8D;IAC9D,MAAM,EAAE,GAAG,CAAC,EAAE,CAAC;IACf,MAAM,EAAE,GAAG,EAAE,CAAC;IACd,MAAM,UAAU,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;IAEzC,OAAO;QACL,MAAM,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE;QACtB,MAAM,EAAE,EAAE,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,GAAG,EAAE,MAAM,EAAE,QAAQ,GAAG,OAAO,CAAC,KAAK,EAAE;QACxE,OAAO,EAAE,GAAG,KAAK,MAAM,IAAI,KAAK,KAAK,QAAQ;QAC7C,UAAU;KACX,CAAC;AACJ,CAAC"}

package/dist/mcp/ready.d.ts

@@ -15,6 +15,7 @@ export type ReadyResult = {
     ok: true;
     transport: ObsbotTransport;
     reconnected: boolean;
+    woke: boolean;
 } | {
     ok: false;
     reason: "unreachable" | "wake-timeout";

package/dist/mcp/ready.js

@@ -48,7 +48,9 @@ export async function ensureReady(getTransport, reconnect, opts = {}) {
             return { ok: false, reason: "unreachable", error: `camera not reachable: ${msg(e2)}` };
         }
     }
+    let woke = false;
     if (!awake) {
+        woke = true;
         await t.sendVendor(encodeSetRunStatus("run").buildFrame(t.nextSeq()));
         let waited = 0;
         while (waited < wakeTimeoutMs) {
@@ -69,6 +71,6 @@ export async function ensureReady(getTransport, reconnect, opts = {}) {
         }
         await sleep(settleMs); // let the gimbal finish rising before we drive it
     }
-    return { ok: true, transport: t, reconnected: reconnect?.takeReconnected() ?? false };
+    return { ok: true, transport: t, reconnected: reconnect?.takeReconnected() ?? false, woke };
 }
 //# sourceMappingURL=ready.js.map

package/dist/mcp/ready.js.map

@@ -1 +1 @@
-{"version":3,"file":"ready.js","sourceRoot":"","sources":["../../src/mcp/ready.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AACpD,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AA0B1D,MAAM,QAAQ,GAAwB,EAAE,MAAM,EAAE,GAAG,EAAE,aAAa,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC;AAE1F,MAAM,KAAK,GAAG,CAAC,EAAU,EAAiB,EAAE,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;AACnF,gFAAgF;AAChF,gFAAgF;AAChF,gDAAgD;AAChD,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC,CAAU,EAAU,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AAExF,MAAM,SAAS,GAAG,KAAK,EAAE,CAAkB,EAAoB,EAAE,CAC/D,YAAY,CAAC,MAAM,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC;AAE3C;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,YAA4C,EAC5C,SAAwB,EACxB,OAAkB,EAAE;IAEpB,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,QAAQ,EAAE,GAAG,EAAE,GAAG,QAAQ,EAAE,GAAG,IAAI,EAAE,CAAC;IAErE,IAAI,CAAkB,CAAC;IACvB,IAAI,CAAC;QACH,CAAC,GAAG,MAAM,YAAY,EAAE,CAAC;IAC3B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,qBAAqB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;IACpF,CAAC;IAED,IAAI,KAAc,CAAC;IACnB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC;IAC7B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,oEAAoE;QACpE,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,yBAAyB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACxF,CAAC;QACD,MAAM,SAAS,CAAC,UAAU,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,CAAC,GAAG,MAAM,YAAY,EAAE,CAAC;YACzB,KAAK,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC;QAC7B,CAAC;QAAC,OAAO,EAAE,EAAE,CAAC;YACZ,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,yBAAyB,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACzF,CAAC;IACH,CAAC;IAED,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,CAAC,CAAC,UAAU,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtE,IAAI,MAAM,GAAG,CAAC,CAAC;QACf,OAAO,MAAM,GAAG,aAAa,EAAE,CAAC;YAC9B,MAAM,KAAK,CAAC,MAAM,CAAC,CAAC;YACpB,MAAM,IAAI,MAAM,CAAC;YACjB,IAAI,CAAC;gBACH,IAAI,MAAM,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC;oBACvB,KAAK,GAAG,IAAI,CAAC;oBACb,MAAM;gBACR,CAAC;YACH,CAAC;YAAC,MAAM,CAAC;gBACP,+DAA+D;YACjE,CAAC;QACH,CAAC;QACD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,cAAc,EAAE,KAAK,EAAE,oCAAoC,EAAE,CAAC;QAC5F,CAAC;QACD,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,kDAAkD;IAC3E,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,EAAE,WAAW,EAAE,SAAS,EAAE,eAAe,EAAE,IAAI,KAAK,EAAE,CAAC;AACxF,CAAC"}
+{"version":3,"file":"ready.js","sourceRoot":"","sources":["../../src/mcp/ready.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AACpD,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AA0B1D,MAAM,QAAQ,GAAwB,EAAE,MAAM,EAAE,GAAG,EAAE,aAAa,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC;AAE1F,MAAM,KAAK,GAAG,CAAC,EAAU,EAAiB,EAAE,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;AACnF,gFAAgF;AAChF,gFAAgF;AAChF,gDAAgD;AAChD,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC,CAAU,EAAU,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AAExF,MAAM,SAAS,GAAG,KAAK,EAAE,CAAkB,EAAoB,EAAE,CAC/D,YAAY,CAAC,MAAM,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC;AAE3C;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,YAA4C,EAC5C,SAAwB,EACxB,OAAkB,EAAE;IAEpB,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,QAAQ,EAAE,GAAG,EAAE,GAAG,QAAQ,EAAE,GAAG,IAAI,EAAE,CAAC;IAErE,IAAI,CAAkB,CAAC;IACvB,IAAI,CAAC;QACH,CAAC,GAAG,MAAM,YAAY,EAAE,CAAC;IAC3B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,qBAAqB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;IACpF,CAAC;IAED,IAAI,KAAc,CAAC;IACnB,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC;IAC7B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,oEAAoE;QACpE,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,yBAAyB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACxF,CAAC;QACD,MAAM,SAAS,CAAC,UAAU,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,CAAC,GAAG,MAAM,YAAY,EAAE,CAAC;YACzB,KAAK,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC;QAC7B,CAAC;QAAC,OAAO,EAAE,EAAE,CAAC;YACZ,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,yBAAyB,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACzF,CAAC;IACH,CAAC;IAED,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,IAAI,GAAG,IAAI,CAAC;QACZ,MAAM,CAAC,CAAC,UAAU,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtE,IAAI,MAAM,GAAG,CAAC,CAAC;QACf,OAAO,MAAM,GAAG,aAAa,EAAE,CAAC;YAC9B,MAAM,KAAK,CAAC,MAAM,CAAC,CAAC;YACpB,MAAM,IAAI,MAAM,CAAC;YACjB,IAAI,CAAC;gBACH,IAAI,MAAM,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC;oBACvB,KAAK,GAAG,IAAI,CAAC;oBACb,MAAM;gBACR,CAAC;YACH,CAAC;YAAC,MAAM,CAAC;gBACP,+DAA+D;YACjE,CAAC;QACH,CAAC;QACD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,cAAc,EAAE,KAAK,EAAE,oCAAoC,EAAE,CAAC;QAC5F,CAAC;QACD,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,kDAAkD;IAC3E,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,EAAE,WAAW,EAAE,SAAS,EAAE,eAAe,EAAE,IAAI,KAAK,EAAE,IAAI,EAAE,CAAC;AAC9F,CAAC"}

package/dist/mcp/server.js

@@ -71,7 +71,7 @@ export async function startServer(opts = {}) {
     process.on("SIGTERM", () => { shutdown(); process.exit(0); });
     const server = new Server(
     // Must match package.json's version; test/version-sync.test.ts enforces it.
-    { name: "obsbot-mcp", version: "0.4.1" }, { capabilities: { tools: {} } });
+    { name: "obsbot-mcp", version: "0.6.0" }, { capabilities: { tools: {} } });
     server.setRequestHandler(ListToolsRequestSchema, async () => ({
         tools: tools.map((tool) => ({
             name: tool.name,