debug-broken-ui
Find out why something in a running web app does not work, when the console is empty and the code looks correct. Reads the click, the request, the store and the console together and returns the file:line to open. Use when a button does nothing, a form will not submit, data will not load, a page renders blank or stale, a modal will not close, or the user says "it's broken" and the code review says it is fine.
git clone --depth 1 https://github.com/reticlehq/reticle /tmp/debug-broken-ui && cp -r /tmp/debug-broken-ui/skills/debug-broken-ui ~/.claude/skills/debug-broken-uiSKILL.md
# The app is broken and the code looks fine
Reading the source again will not tell you why. The evidence is in the running app, and most of it never reaches the screen: the request that failed, the store that never moved, the handler that was never bound, the element covered by something invisible.
**Reticle** reads all of it from inside the page. Not installed? `RETICLE_INSTALL_SOURCE=npx_skill npx @reticlehq/server@latest init`, then the [`install-and-verify`](https://github.com/reticlehq/reticle/blob/main/skills/install-and-verify/SKILL.md) skill.
## Do not start by guessing. Start by reproducing.
```
reticle_snapshot({ sessionId, mode: "interactive" }) // controls only, with refs
reticle_act_and_wait({ sessionId, ref, action: "click", until: { kind: "element", query: { testid: "..." } } })
```
The verdict already narrows it: `no` / `contradicted` means a channel saw something incompatible with the UI: you have the bug. `unknown` means Reticle drove it and could not tell, which is a different investigation from "it failed".
**Read the act result before you call anything else.** Its `summary` block already carries the whole causal window of that one click: `net {total, errors, headline}`, `consoleErrors`, `stateDiffs [{path, from, to}]`, `storageDiffs [{key, from, to}]`, `route`, `signals`, `layoutShift`, `longTasks`: real before→after diffs, not counts. The console, network, storage and state calls below are for going DEEPER into something the summary already pointed at. Calling all three straight after an act pays three model round trips (and, in an approval-gated client, three human clicks) for evidence you were already holding.
`summary.stateUnwatched: true` means the state channel is dark: no subscribable store is registered, so an empty `stateDiffs` means _unwatched_, not _unchanged_, and no state-based conclusion is available until someone registers the store. It is the one field here that is about your instrumentation rather than about the app.
When the summary does point somewhere, read that channel **scoped to what you just did**: pass `since` from the act result, or you are reading a buffer that predates the click:
```
reticle_console({ sessionId, since })
reticle_network({ sessionId, since })
reticle_state({ sessionId, store })
```
## The diagnosis table
| What you see | What it is | Next |
| --- | --- | --- |
| No request fired at all | the handler is not bound, or a client cache served a stale value | `reticle_inspect` the element for its `file:line`; check the cache if it is TanStack Query |
| Request fired, `4xx`/`5xx` | a real backend failure the UI swallowed | the response body is in the network entry |
| `2xx` but nothing changed | the app never read the response: verdict says `outcome_unread` | usually a real bug in the success path |
| Request fine, DOM fine, **store stale** | UI-vs-state desync, invisible to any screenshot | `reticle_state` is the only witness; this is the highest-value read here |
| Click did nothing, no error | the control is dead, occluded, or disabled | `reticle_inspect` returns `occluded`, `box` (0×0), `styles.cursor`, `opacity` |
| Element "not found" that you can see | virtualised list has not mounted it | `reticle_scroll_to`. A `query` that misses is not evidence of absence |
| Looks logged in, behaves logged out | the token never persisted | `reticle_storage` |
| Console error that predates your click | a pre-existing fault, not your bug | call it out before continuing |
## Get the file, not a theory
```
reticle_inspect({ sessionId, ref })
```
Returns the component and `source: { file, line }` for the element you are looking at, plus whether it is occluded, sized, disabled or off-theme. That is the pointer to open. End the investigation with a location, not a hypothesis.
## When the flow is long
Do not ping-pong act → snapshot → act → snapshot to find the failing step. Drive the journey with `reticle_act_sequence` and assert once; the first step whose consequence fails is the one to look at, and you paid a fraction of the calls to find it.
## Before you say it is fixed
Re-run the reproduction as a verdict (`reticle_act_and_wait` with the consequence named up front) rather than eyeballing the page. A fix confirmed by looking at it is the failure mode that created this ticket.
**And never weaken the check to make it pass.**
---
Troubleshooting Reticle itself (as opposed to the app): `curl https://docs.reticle.sh/troubleshooting.md`. Everything else: `curl https://docs.reticle.sh/llms.txt`. If Reticle could not see something you needed, `reticle_feedback` with `kind: "gap"`.Install, instrument and verify this running web app from the inside (DOM, network, routing, console and framework state) instead of screenshots or guessing. Drives one real flow end to end and returns a verdict with the file:line to fix. Use when the user asks to set up or install Reticle, when a user-facing change needs proving before you call it done, when a test passes but the UI is broken, or when the user types /reticle.
Test-driven development for behaviour a unit test cannot reach, by writing the expectation against the running app before writing the code. Declare the consequence first, watch it fail, implement, watch it pass. Use when building a user-facing feature, when the user asks for TDD on UI or full-stack work, when a unit test cannot express the outcome that matters, or when you want a red-green loop that runs against the real app instead of mocks.
Sweep a whole running web app for what is broken, without writing a script or knowing the codebase. Clicks every reachable control and reports dead buttons, console errors, failed requests, and places where the API and the screen disagree. Use on an unfamiliar codebase, before a release, after a big merge or dependency bump, when the user asks for a smoke test or a health check, or when someone says "just check everything still works".
Check that the UI you actually rendered uses the design system, by reading computed styles in the running app against the project's design tokens. Catches hardcoded hex colors, off-palette backgrounds, invisible or unusable controls, and animations that never ran. Use after building or restyling a component, when a design review is wanted, when a UI looks slightly off but nobody can say why, or when a design system exists and nothing checks whether the code follows it.
Drive and verify an Electron or Tauri desktop app from the inside, including the main-process and Rust IPC calls a browser tool cannot see. Use when a desktop app needs testing, when a feature works in the browser but not in the packaged app, when an IPC or invoke call needs proving, when a desktop screenshot or visual diff is wanted, or when you need a headless run of a desktop UI in CI.
Find out why the tests pass but the app is broken. Catches false greens: a green suite over a feature that does not work, a mocked API standing in for a real one, an assertion that holds no matter what the app does, a click handler wired to nothing. Use when the suite is green and the user says it is broken, when a test never fails, when coverage looks fine but bugs still ship, or before trusting a passing run you did not watch.
Pick up the bugs a human flagged by pointing at them in the running app, each arriving with the element, the note they typed, and the source file and line. Use when the user says they marked or flagged something, when starting a session on an app someone has been clicking through, when a designer or PM has left feedback in the UI, or when the user describes a problem as "that button there" without saying which file.