feat(agent): add a hardware-free mock agent for GUI development - #568
Open
AprilNEA wants to merge 7 commits into
Open
feat(agent): add a hardware-free mock agent for GUI development#568AprilNEA wants to merge 7 commits into
AprilNEA wants to merge 7 commits into
Conversation
openlogi-agent-mock serves the real Agent tarpc service on the real IPC socket from a scripted in-memory inventory: an online mouse (DPI + SmartShift state that persists and reads back, battery draining ~1%/minute so poll-driven repaints are visible), an offline mouse, a lighting-capable keyboard, and one directly-attached device — covering every capability-gated panel and both route kinds without hardware. start_pairing runs a scripted Bolt flow (discovery, passkey, paired) and the paired device joins the inventory. The mock holds agent.lock, so real agents spawned meanwhile (GUI auto-spawn, launchd) exit as duplicates; it answers the current PROTOCOL_VERSION so a release agent's takeover handshake stands down. Never bundled: xtask copies openlogi-agent by exact path.
Greptile SummaryAdds a development-only mock agent that implements the real IPC contract using scripted devices and pairing behavior, allowing GUI development without Logitech hardware.
Confidence Score: 5/5The PR appears safe to merge. No blocking failure remains; the previously reported stale pairing tasks now verify their originating session before discovering or completing a device.
|
| Filename | Overview |
|---|---|
| crates/openlogi-agent/src/bin/mock_agent.rs | Implements the mock IPC agent, scripted inventories and settings, standalone lighting, and session-scoped pairing lifecycle. |
| crates/openlogi-agent/Cargo.toml | Registers the mock as a separate development binary without changing the production agent target. |
| docs/DEVELOPMENT.md | Documents how to run the mock with the development GUI and explains profile and lock behavior. |
| AGENTS.md | Adds a concise pointer to the hardware-free GUI development workflow. |
Sequence Diagram
sequenceDiagram
participant GUI
participant Mock as Mock Agent
participant State as Pairing State
GUI->>Mock: start_pairing()
Mock->>State: begin_pairing(session id)
Mock-->>GUI: Searching
Mock-->>GUI: DeviceFound
GUI->>Mock: pair_device(address)
Mock->>State: verify active session id
Mock-->>GUI: Passkey
Mock->>State: end matching session
Mock->>State: append paired device
Mock-->>GUI: Paired(slot)
Reviews (4): Last reviewed commit: "fix(agent): drop the mock's passkey prom..." | Re-trigger Greptile
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Adds the run recipe and its constraints (real agent must be stopped; they share agent.lock) to the developer handbook, and a pointer in the agent guide's build/run section.
next_pairing awaited the receiver while holding its lock, so a start_pairing arriving mid-hold waited out the poll; it now polls at a 100ms tick and holds the lock only for the try_recv. Rebuilds the scripted inventory per poll (the mouse battery is the one field that moves) instead of patching a stored template, which drops the by-uid lookup and the slot-0 fallback pairing had when that lookup could miss.
Brings the mock up to the current contract: the two standalone-light RPCs, the snapshot's standalone list and camera_active, and the thumbwheel capability. Adds a scripted Litra Glow (capabilities the light panel gates on, writes accepted, unknown routes rejected) and a camera_active flag that flips on a timer so camera-linked rendering has an input to follow. The mock now claims the openlogi-dev profile unless OPENLOGI_PROFILE says otherwise. Dev bundles moved to their own socket, so a production-profile mock would never meet the dev GUI — and would sit on the installed app's paths instead. Its light keeps a Some(registry_model_id): #571 makes a None truncate the bincode stream and the whole snapshot undecodable.
The end-of-file-fixer hook rewrites both on every push, which blocks the push until they are committed. Byte-for-byte what the hook produces.
AprilNEA
force-pushed
the
feat/mock-agent
branch
from
August 11, 2026 09:54
deb8d32 to
3c7253d
Compare
A pairing session's spawned tasks sleep across a cancel, so a session started right after one could be corrupted by its predecessor: the stale discovery task handed the replacement a device it never searched for, and the stale pair task consumed the replacement outright — adding the cancelled device and sending Paired through its own dropped channel, so the GUI saw neither the pairing nor a live session. Sessions now carry an id and both tasks act only while it is still the live one. Cancelling with nothing active also returns Ok, matching PairingManager::cancel — the mock must not hand the GUI a contract the real agent does not have. Verified against a running mock: cancel-then-restart inside the discovery delay yields exactly one DeviceFound (the new session's own), and inside the passkey wait yields no Paired, an unchanged inventory, and a session that is still cancellable.
There was a problem hiding this comment.
Pull request overview
Adds a dev-only openlogi-agent-mock binary in crates/openlogi-agent that serves the real tarpc Agent IPC contract from a scripted, in-memory inventory so the GUI can be developed and exercised without any attached Logitech hardware.
Changes:
- Introduces
openlogi-agent-mock(new bin target) implementing theAgentservice with scripted inventory + pairing flow. - Documents how to run the GUI against the mock agent in
docs/DEVELOPMENT.mdand adds a pointer inAGENTS.md. - Removes trailing blank lines in workspace
Cargo.tomlandthird_party/tarpc/Cargo.tomlto satisfy formatting hooks.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| crates/openlogi-agent/src/bin/mock_agent.rs | New mock agent binary implementing the IPC contract with scripted inventory/pairing. |
| crates/openlogi-agent/Cargo.toml | Registers the new openlogi-agent-mock binary target. |
| docs/DEVELOPMENT.md | Adds a “no hardware” GUI development run recipe using the mock agent. |
| AGENTS.md | Adds a short pointer to the mock-agent workflow for GUI development. |
| Cargo.toml | Trims a trailing blank line to satisfy formatting hooks. |
| third_party/tarpc/Cargo.toml | Trims a trailing blank line to satisfy formatting hooks. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Comment on lines
+781
to
+797
| let state = Arc::clone(&self.state); | ||
| tokio::spawn(async move { | ||
| tokio::time::sleep(PASSKEY_DELAY).await; | ||
| let _ = tx.send(PairingUpdate::Passkey(PasskeyMethod::Keyboard( | ||
| "482913".to_string(), | ||
| ))); | ||
| tokio::time::sleep(PASSKEY_TYPING_DELAY).await; | ||
| let mut state = state.lock().await; | ||
| // No session of ours left = cancelled while the "user" was typing. | ||
| // Ending it by id also means a session started in the meantime is | ||
| // neither consumed here nor completed with our (dropped) channel. | ||
| if state.end_pairing(id).is_none() { | ||
| return; | ||
| } | ||
| let slot = state.pair_scripted(&name); | ||
| let _ = tx.send(PairingUpdate::Paired { slot }); | ||
| }); |
Owner
Author
There was a problem hiding this comment.
Right, and it mattered more than it looks: only start_pairing swaps the long-poll receiver, so after a plain cancel the GUI is still polling that same channel and would have seen a passkey prompt arrive behind Failed(Cancelled). Fixed in 3dfe1f4 — the task now checks it is still the live session before sending.
Verified against a running mock: cancelling 400ms into the 800ms delay yields only Failed(Cancelled), with no prompt behind it.
Comment on lines
+161
to
+167
| #[expect( | ||
| unsafe_code, | ||
| reason = "the profile must be chosen before openlogi_core::paths caches it, and only a process-wide env var selects it" | ||
| )] | ||
| unsafe { | ||
| std::env::set_var("OPENLOGI_PROFILE", "dev"); | ||
| } |
Owner
Author
There was a problem hiding this comment.
std::env::set_var is not safe here — it became unsafe in Rust 2024, which this workspace uses. Without the block it does not compile, and if the block were redundant the unused_unsafe warning would fail the build under -D warnings.
The env var is also the only mechanism available: openlogi_core::paths selects the profile from OPENLOGI_PROFILE (or a .dev bundle id, which a bare cargo run binary does not have) and caches the answer in a OnceLock. Setting it as the first statement of main — before any runtime, subscriber or thread exists — is exactly the condition that makes the call sound, which is what the SAFETY comment records.
A plain cancel leaves the GUI polling the same channel (only start_pairing swaps the receiver), so the stale task's Passkey arrived after Failed(Cancelled). The task now checks it is still the live session first. Verified: cancelling 400ms into the 800ms passkey delay yields only Failed(Cancelled), no prompt behind it, and an unchanged inventory.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds
openlogi-agent-mock, a dev-only binary that serves the realAgenttarpccontract from a scripted in-memory inventory. The GUI runs unmodified against it —
no device, no receiver, no Accessibility grant — which unblocks UI work on machines
with no hardware attached and makes states that are awkward to reproduce on real
devices (offline device, unsupported feature, a full pairing flow, a standalone
light) reachable on demand.
It defaults itself to the
openlogi-devprofile — the one the dev app bundlealready uses — so it meets the dev GUI on the dev socket and the installed
production app keeps running untouched.
The mock is a separate binary rather than a flag on the real agent: no production
code path changes, and
xtaskbundlesopenlogi-agentby exact path, so it cannever ship.
Changes
openlogi-agent: newopenlogi-agent-mockbin (src/bin/mock_agent.rs).thumbwheel), an offline mouse (no capabilities — exercises the
presumed_from_kindfallback) and a lighting-capable keyboard, plus adirectly-attached mouse and a standalone Litra Glow: both
DeviceRoutekinds,every capability-gated panel, and the standalone-light surface.
feature answer
FeatureUnsupportedand unknown routesDeviceNotFound, so theGUI's permanent-vs-transient error handling is exercised.
camera_activeflips every30s, making poll-driven repaints and camera-linked light rendering observable.
start_pairingruns a scripted Bolt session (Searching → DeviceFound →Passkey → Paired) and the paired device joins the inventory.
agent.lockand answers the currentPROTOCOL_VERSION,so a real agent spawned meanwhile exits as a duplicate and a release agent's
takeover handshake stands down.
agent_versioncarries a-mocksuffix so amock session is identifiable in the UI.
docs/DEVELOPMENT.md;pointer in
AGENTS.md's build/run section.Cargo.tomlandthird_party/tarpc/Cargo.toml— theend-of-file-fixerhook rewrites both onevery push, so pushes are blocked until they are committed. Unrelated to the
mock; happy to split it out.
No changes to the IPC contract —
PROTOCOL_VERSIONis untouched.Testing
devenv tasks run openlogi:check(fmt + clippy-D warnings+ workspace tests)— passes.
decodes snapshots, and resolves a real asset depot for every scripted device
(
mx_master_3s,mx_keys,mx_vertical,litra_glow). The installedproduction agent kept running throughout, untouched.
handshake at v13, snapshot,
set_dpi(3000)→read_dpireads back 3000,read_smartshift,FeatureUnsupportedon the keyboard,DeviceNotFoundon anunknown route,
set_light/set_light_manual_poweraccepted for the light andrejected for a wrong route, and the pairing flow through
Paired { slot: 4 }with the device appearing in the next snapshot.
Device-card visuals were not screenshotted (this dev host has no screen-recording
permission); the panels are driven by the scripted
Capabilitiesverified above.Note
Building this surfaced #571:
StandaloneDevice::registry_model_idisskip_serializing_if, which truncates the bincode stream when it isNoneandmakes the entire snapshot undecodable by the GUI. The mock therefore always sends
a
Some(...). Not fixed here — it needs aPROTOCOL_VERSIONbump and goldenregeneration of its own.