recovery-probe 0.5.0-preview.7 → 0.6.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (231) hide show
  1. package/CHANGELOG.md +200 -0
  2. package/CONTRIBUTING.md +24 -7
  3. package/README.md +81 -65
  4. package/bin/recovery-probe.mjs +2 -187
  5. package/dist/cli/action.d.mts +20 -0
  6. package/dist/cli/action.mjs +57 -0
  7. package/dist/cli/android.d.mts +1 -0
  8. package/dist/cli/android.mjs +107 -0
  9. package/dist/cli/args.d.mts +18 -0
  10. package/dist/cli/args.mjs +63 -0
  11. package/dist/cli/check.d.mts +41 -0
  12. package/dist/cli/check.mjs +185 -0
  13. package/dist/cli/compare.d.mts +1 -0
  14. package/dist/cli/compare.mjs +86 -0
  15. package/dist/cli/core-demo.d.mts +1 -0
  16. package/dist/cli/core-demo.mjs +9 -0
  17. package/dist/cli/demo.d.mts +11 -0
  18. package/dist/cli/demo.mjs +15 -0
  19. package/dist/cli/discover.d.mts +29 -0
  20. package/dist/cli/discover.mjs +92 -0
  21. package/dist/cli/doctor.d.mts +1 -0
  22. package/dist/cli/doctor.mjs +41 -0
  23. package/dist/cli/init.d.mts +1 -0
  24. package/dist/cli/init.mjs +95 -0
  25. package/dist/cli/legacy.d.mts +16 -0
  26. package/dist/cli/legacy.mjs +117 -0
  27. package/dist/cli/login.d.mts +11 -0
  28. package/dist/cli/login.mjs +40 -0
  29. package/dist/cli/main.d.mts +1 -0
  30. package/dist/cli/main.mjs +181 -0
  31. package/dist/cli/mcp.d.mts +1 -0
  32. package/dist/cli/mcp.mjs +29 -0
  33. package/dist/cli/net.d.mts +1 -0
  34. package/dist/cli/net.mjs +123 -0
  35. package/dist/cli/observe.d.mts +1 -0
  36. package/dist/cli/observe.mjs +238 -0
  37. package/dist/cli/reads.d.mts +29 -0
  38. package/dist/cli/reads.mjs +95 -0
  39. package/dist/cli/refresh.d.mts +1 -0
  40. package/dist/cli/refresh.mjs +130 -0
  41. package/dist/cli/repair.d.mts +1 -0
  42. package/dist/cli/repair.mjs +87 -0
  43. package/dist/cli/runner.d.mts +24 -0
  44. package/dist/cli/runner.mjs +105 -0
  45. package/dist/cli/start.d.mts +1 -0
  46. package/dist/cli/start.mjs +26 -0
  47. package/dist/cli/suspects.d.mts +1 -0
  48. package/dist/cli/suspects.mjs +59 -0
  49. package/dist/cli/terminal.d.mts +5 -0
  50. package/dist/cli/terminal.mjs +19 -0
  51. package/dist/cli/until-stopped.d.mts +3 -0
  52. package/dist/cli/until-stopped.mjs +17 -0
  53. package/dist/cli/verify.d.mts +1 -0
  54. package/dist/cli/verify.mjs +58 -0
  55. package/dist/cli/watch.d.mts +1 -0
  56. package/dist/cli/watch.mjs +129 -0
  57. package/dist/cli/windows.d.mts +11 -0
  58. package/dist/cli/windows.mjs +35 -0
  59. package/dist/core/faults.d.mts +9 -0
  60. package/dist/core/faults.mjs +19 -0
  61. package/dist/core/package.d.mts +3 -0
  62. package/dist/core/package.mjs +6 -0
  63. package/dist/{ipc-observation.d.mts → core/screen.d.mts} +5 -2
  64. package/dist/{ipc-observation.mjs → core/screen.mjs} +5 -1
  65. package/dist/core/urls.d.mts +1 -0
  66. package/dist/core/urls.mjs +8 -0
  67. package/dist/desktop.d.mts +4 -67
  68. package/dist/desktop.mjs +4 -433
  69. package/dist/fetch-fault.d.mts +2 -21
  70. package/dist/fetch-fault.mjs +1 -84
  71. package/dist/index.d.mts +6 -0
  72. package/dist/index.mjs +4 -256
  73. package/dist/ipc-inspector.d.mts +1 -19
  74. package/dist/ipc-inspector.mjs +1 -144
  75. package/dist/ipc-probe.cjs +26 -4
  76. package/dist/ipc-probe.d.cts +12 -0
  77. package/dist/net-fault.d.mts +1 -0
  78. package/dist/net-fault.mjs +1 -0
  79. package/dist/reports/action.d.mts +2 -0
  80. package/dist/reports/action.mjs +25 -0
  81. package/dist/{desktop-report.d.mts → reports/check.d.mts} +1 -1
  82. package/dist/{desktop-report.mjs → reports/check.mjs} +3 -2
  83. package/dist/reports/discovery.d.mts +23 -0
  84. package/dist/{discovery-report.mjs → reports/discovery.mjs} +12 -2
  85. package/dist/reports/evidence.d.mts +48 -0
  86. package/dist/reports/evidence.mjs +169 -0
  87. package/dist/reports/folder.d.mts +8 -0
  88. package/dist/reports/folder.mjs +22 -0
  89. package/dist/{report-template.d.mts → reports/template.d.mts} +1 -1
  90. package/dist/{report-template.mjs → reports/template.mjs} +2 -2
  91. package/dist/reports/templates/action.html +51 -0
  92. package/dist/{report-templates → reports/templates}/discovery.html +8 -0
  93. package/dist/targets/android.d.mts +55 -0
  94. package/dist/targets/android.mjs +172 -0
  95. package/dist/targets/browser.d.mts +13 -0
  96. package/dist/targets/browser.mjs +25 -0
  97. package/dist/targets/cdp-bridge.d.mts +31 -0
  98. package/dist/targets/cdp-bridge.mjs +237 -0
  99. package/dist/targets/cdp.d.mts +7 -0
  100. package/dist/targets/cdp.mjs +45 -0
  101. package/dist/targets/ipc-window.d.mts +11 -0
  102. package/dist/targets/ipc-window.mjs +17 -0
  103. package/dist/{visibility.d.mts → targets/visibility.d.mts} +4 -1
  104. package/dist/{visibility.mjs → targets/visibility.mjs} +39 -22
  105. package/dist/tools/compare.d.mts +76 -0
  106. package/dist/tools/compare.mjs +255 -0
  107. package/dist/tools/doctor.d.mts +33 -0
  108. package/dist/tools/doctor.mjs +183 -0
  109. package/dist/tools/guided-start.d.mts +47 -0
  110. package/dist/tools/guided-start.mjs +279 -0
  111. package/dist/tools/init.d.mts +38 -0
  112. package/dist/tools/init.mjs +24 -0
  113. package/dist/tools/mcp/server.d.mts +30 -0
  114. package/dist/tools/mcp/server.mjs +131 -0
  115. package/dist/tools/mcp/tools.d.mts +5 -0
  116. package/dist/tools/mcp/tools.mjs +540 -0
  117. package/dist/tools/pick-marker.d.mts +10 -0
  118. package/dist/tools/pick-marker.mjs +99 -0
  119. package/dist/tools/repair/process.d.mts +17 -0
  120. package/dist/tools/repair/process.mjs +99 -0
  121. package/dist/tools/repair/provider.d.mts +53 -0
  122. package/dist/tools/repair/provider.mjs +72 -0
  123. package/dist/tools/repair/run.d.mts +72 -0
  124. package/dist/tools/repair/run.mjs +444 -0
  125. package/dist/tools/suspects.d.mts +54 -0
  126. package/dist/tools/suspects.mjs +444 -0
  127. package/dist/tools/verify.d.mts +63 -0
  128. package/dist/tools/verify.mjs +185 -0
  129. package/dist/tools/watch.d.mts +109 -0
  130. package/dist/tools/watch.mjs +328 -0
  131. package/dist/transports/fetch.d.mts +21 -0
  132. package/dist/transports/fetch.mjs +80 -0
  133. package/dist/{http-observer.d.mts → transports/http/observer.d.mts} +4 -2
  134. package/dist/{http-observer.mjs → transports/http/observer.mjs} +10 -13
  135. package/dist/transports/http/reads.d.mts +2 -0
  136. package/dist/transports/http/reads.mjs +13 -0
  137. package/dist/transports/http/route.d.mts +24 -0
  138. package/dist/transports/http/route.mjs +117 -0
  139. package/dist/transports/ipc/control.d.mts +19 -0
  140. package/dist/transports/ipc/control.mjs +64 -0
  141. package/dist/transports/ipc/inspector.d.mts +19 -0
  142. package/dist/transports/ipc/inspector.mjs +144 -0
  143. package/dist/transports/net.d.mts +80 -0
  144. package/dist/transports/net.mjs +307 -0
  145. package/dist/transports/session.d.mts +52 -0
  146. package/dist/{refresh-adapters.mjs → transports/session.mjs} +20 -10
  147. package/dist/web.d.mts +2 -24
  148. package/dist/web.mjs +2 -52
  149. package/dist/workflows/action.d.mts +74 -0
  150. package/dist/workflows/action.mjs +310 -0
  151. package/dist/workflows/check/browser.d.mts +21 -0
  152. package/dist/workflows/check/browser.mjs +104 -0
  153. package/dist/workflows/check/http.d.mts +51 -0
  154. package/dist/workflows/check/http.mjs +270 -0
  155. package/dist/{ipc-check.d.mts → workflows/check/ipc.d.mts} +3 -3
  156. package/dist/{ipc-check.mjs → workflows/check/ipc.mjs} +8 -31
  157. package/dist/workflows/discover/experiment.d.mts +59 -0
  158. package/dist/{discovery-control.mjs → workflows/discover/experiment.mjs} +30 -1
  159. package/dist/workflows/discover/http.d.mts +41 -0
  160. package/dist/workflows/discover/http.mjs +113 -0
  161. package/dist/workflows/discover/ipc.d.mts +41 -0
  162. package/dist/workflows/discover/ipc.mjs +125 -0
  163. package/dist/workflows/discover/run.d.mts +153 -0
  164. package/dist/workflows/discover/run.mjs +291 -0
  165. package/dist/workflows/discover/screenshots.d.mts +10 -0
  166. package/dist/workflows/discover/screenshots.mjs +20 -0
  167. package/dist/workflows/observe/files.d.mts +49 -0
  168. package/dist/workflows/observe/files.mjs +112 -0
  169. package/dist/{observation-kit.mjs → workflows/observe/kit.mjs} +18 -0
  170. package/dist/workflows/observe/network.d.mts +13 -0
  171. package/dist/workflows/observe/network.mjs +57 -0
  172. package/dist/{observation-report.d.mts → workflows/observe/report.d.mts} +33 -0
  173. package/dist/{observation-report.mjs → workflows/observe/report.mjs} +27 -6
  174. package/dist/{observe.d.mts → workflows/observe/run.d.mts} +15 -9
  175. package/dist/{observe.mjs → workflows/observe/run.mjs} +119 -17
  176. package/dist/workflows/observe/steps.d.mts +58 -0
  177. package/dist/{observe-steps.mjs → workflows/observe/steps.mjs} +63 -15
  178. package/dist/workflows/reads.d.mts +16 -0
  179. package/dist/workflows/reads.mjs +46 -0
  180. package/dist/workflows/refresh.d.mts +156 -0
  181. package/dist/workflows/refresh.mjs +398 -0
  182. package/dist/workflows/web.d.mts +12 -0
  183. package/dist/workflows/web.mjs +23 -0
  184. package/docs/android.md +67 -0
  185. package/docs/api.md +9 -1
  186. package/docs/ci.md +1 -1
  187. package/docs/compare.md +134 -0
  188. package/docs/desktop.md +18 -62
  189. package/docs/discovery.md +101 -0
  190. package/docs/ipc.md +36 -56
  191. package/docs/mcp.md +87 -0
  192. package/docs/net.md +83 -0
  193. package/docs/observe.md +61 -16
  194. package/docs/refresh.md +3 -3
  195. package/docs/repair.md +5 -5
  196. package/docs/runner.md +5 -2
  197. package/docs/start.md +5 -13
  198. package/docs/suspects.md +100 -0
  199. package/docs/verify.md +60 -0
  200. package/docs/watch.md +59 -0
  201. package/package.json +34 -35
  202. package/bin/attach.mjs +0 -79
  203. package/bin/core-demo.mjs +0 -11
  204. package/bin/desktop.mjs +0 -157
  205. package/bin/ipc.mjs +0 -108
  206. package/bin/observe.mjs +0 -255
  207. package/bin/refresh.mjs +0 -162
  208. package/bin/repair.mjs +0 -112
  209. package/bin/start.mjs +0 -39
  210. package/bin/web.mjs +0 -120
  211. package/dist/discovery-control.d.mts +0 -16
  212. package/dist/discovery-report.d.mts +0 -40
  213. package/dist/guided-start.mjs +0 -339
  214. package/dist/http-discover.mjs +0 -356
  215. package/dist/index.d.ts +0 -54
  216. package/dist/ipc-control.d.mts +0 -12
  217. package/dist/ipc-control.mjs +0 -27
  218. package/dist/ipc-discover.d.mts +0 -119
  219. package/dist/ipc-discover.mjs +0 -386
  220. package/dist/observe-steps.d.mts +0 -36
  221. package/dist/pick-marker.mjs +0 -116
  222. package/dist/refresh-adapters.d.mts +0 -40
  223. package/dist/refresh.mjs +0 -423
  224. package/dist/repair-process.mjs +0 -108
  225. package/dist/repair-provider.mjs +0 -94
  226. package/dist/repair.mjs +0 -550
  227. /package/dist/{report-templates → reports/templates}/desktop.html +0 -0
  228. /package/dist/{report-templates → reports/templates}/ipc.html +0 -0
  229. /package/dist/{report-templates → reports/templates}/observation.html +0 -0
  230. /package/dist/{report-templates → reports/templates}/refresh.html +0 -0
  231. /package/dist/{observation-kit.d.mts → workflows/observe/kit.d.mts} +0 -0
package/CHANGELOG.md CHANGED
@@ -4,6 +4,206 @@ All notable changes to Recovery Probe are documented here. The project follows [
4
4
 
5
5
  Version 0.2.0 is the first npm release. Earlier 0.1.x revisions were repository-only prototypes.
6
6
 
7
+ ## 0.6.0 (2026-10-23)
8
+
9
+ Promotes previews 1–13 to the stable npm `latest` channel without additional runtime changes.
10
+
11
+ - Add `init` for guided first checks and `doctor` for setup diagnostics, including IPC hook channels.
12
+ - Unify CLI commands around `check`, `discover`, `action`, `windows`, `reads`, `login` and `demo`, with consistent option parsing and JSON output where supported.
13
+ - Add `verify` and the MCP `verify_change` tool to compare checks before and after a change and scan changed lines for suspects.
14
+ - Add row-state markers and `UNEXPECTED_ROW_STATE` observations, plus discovery experiments that fail later calls separately and optional screenshots.
15
+ - Share discovery, read sessions, HTTP fault handling, target identification and report-folder creation across workflows. Reports expose one evidence shape for comparison, watching and verification, while older reports remain readable.
16
+ - Organize source into tested architectural layers, consolidate documentation and validate documentation links.
17
+
18
+ Upgrading from 0.5.0: public package exports and the IPC hook path are unchanged; restart the development app so its hook matches the CLI. Old CLI spellings remain supported until 1.0 and retain exit code 2 for errors. New commands use 0 for passed, 1 for failed, 2 for inconclusive or needs review, 3 for usage/config errors and 4 when a run could not start. Update scripts that interpret these codes. For `demo` and JSON-runner `check`, `--out` names a new report folder; old spellings retain file output. Unsupported imports into internal `dist/` files may break after the source reorganization. Experimental workflows remain experimental.
19
+
20
+ ## 0.6.0-preview.13 (2026-10-23)
21
+
22
+ - One fault layer. The fault kinds are listed once (`FAULT_KINDS` in `core/faults`), and one function answers a request with a fault for `withFault`, the HTTP observer and the HTTP check. The HTTP check fails its read through the same observer as discovery and observation, instead of its own copy of the route handler.
23
+ - A successful read after a fault counts when its response finishes, not when its headers arrive, as in discovery.
24
+ - The IPC check, the HTTP check and the JSON runner stay separate on purpose: they promise different things. The IPC check needs exact call counts and takes its deadline from the measured read; the HTTP check clicks Retry and waits for a ready marker; the runner opens a fresh browser for each fault.
25
+
26
+ ## 0.6.0-preview.12 (2026-10-23)
27
+
28
+ - Every `report.json` has an `evidence` block in one shape: workflow, transport, target, problems, outcomes, what the run checked and why parts don't count. `compare`, `watch` and `verify` read it; older reports are read from their other fields as before.
29
+ - `compare` matches runs by workflow and transport instead of the report's label, so a check from `check --cdp` compares with one from the old `web --attach`, and an older refresh report compares with a new one.
30
+ - Discovery, action and refresh findings are told apart by experiment: before, `rejection-1`, `rejection-2` and `rejection-1-call2` all counted as `rejection`, so a problem that moved between them looked unchanged.
31
+
32
+ ## 0.6.0-preview.11 (2026-10-23)
33
+
34
+ - One set of commands. `check`, `discover` and `action` read what to do from the config: a `channel` means IPC, an `endpoint` means HTTP, and `check` attaches when given `--cdp` or a config with `cdp` or `page`, otherwise it opens Chromium. `windows`, `reads`, `login` and `demo` replace `--list`, `--discover` without a config, `--login` and `--demo`.
35
+ - The old spellings (`--demo`, `--config`, `desktop`, `web`, `web --attach`, `ipc`) run the new commands with the same options, print one line saying what replaced them, and keep exit code 2 for errors. They work until 1.0.
36
+ - `recovery-probe --help` lists every command with what it does, and `recovery-probe <command> --help` its options. Every command parses its options the same way.
37
+ - Exit codes: 0 passed, 1 a check failed, 2 needs review or inconclusive, 3 wrong usage or config, 4 couldn't run. Before, usage and setup errors exited with 2.
38
+ - `check`, `discover`, `action`, `reads` and `windows` take `--json`. `check` takes `--screenshots` in a browser it opens too.
39
+ - When nothing answers on a debugging port, the error says how to start the app instead of showing Playwright's connection log.
40
+ - `demo --out` and `check --out` with a JSON-runner config name a new folder for report.json, like every other command; the old spellings still write one file.
41
+
42
+ ## 0.6.0-preview.10 (2026-10-23)
43
+
44
+ - One function finds the IPC hook's window and proves the debugging port and the hook reach the same one. It replaces five copies in the IPC check, discovery, the action check, observe and refresh, which had different error messages and, in refresh, removed any window challenge instead of only its own.
45
+ - `ipc`, `observe`, `refresh`, `web` and `web --attach` create their report folder the same way: `--out` must be a new folder, its parent folders are created, and without `--out` it's a new `.recovery-probe/<command>-…` folder.
46
+ - `watch`, `verify` and the MCP server still run each check as its own CLI process, on purpose: a hung or crashing check can be stopped without stopping them.
47
+
48
+ ## 0.6.0-preview.9 (2026-10-23)
49
+
50
+ - IPC and HTTP discovery run on one engine, built on the same read sessions as refresh. The IPC and HTTP modules only set up their transport.
51
+ - Both now behave the same where they had drifted apart:
52
+ - Over HTTP, a healthy load that doesn't settle stops the run with `BASELINE_NOT_HEALTHY`, as over IPC. `BASELINE_UNSTABLE` means healthy loads ended on different screens, over both, without the extra text IPC added.
53
+ - Over IPC, an `UNSTABLE` finding has low confidence, as over HTTP.
54
+ - HTTP findings and runs have the fields IPC ones had: `observationMs`, advice per classification, `verification`, `recoveryDetectedAt`, `timingNote`, `assessment`, and `uncaughtErrors` in the timeline.
55
+ - The read sessions take `skip`, and the HTTP one keeps the requests its observers saw.
56
+
57
+ ## 0.6.0-preview.8 (2026-10-23)
58
+
59
+ - The source is split into layers (core, transports, targets, reports, workflows, tools, cli), and a test keeps each layer from importing the ones above it. `desktop.mts`, which held the HTTP check, CDP attachment and endpoint listing, is now three modules, and `index.mts` no longer holds a check engine. The public imports (`recovery-probe`, `/fetch`, `/net`, `/desktop`, `/web`, `/ipc-inspector`, `/ipc`) and the hook's path are unchanged. Imports of other files inside `dist/` were never supported and now break.
60
+ - Paths to files that ship with the package come from one place, `core/package.mts`.
61
+ - Tests live in `test/unit`, `test/integration`, `test/electron` and `test/package`, and Electron tests share one launcher that stops the whole process tree.
62
+
63
+ ## 0.6.0-preview.7 (2026-10-23)
64
+
65
+ - New `recovery-probe init [URL]`. It finds what to test and runs a first check: the guided web check for a URL, the guided desktop check for an app on the debugging port, or IPC discovery when the app has the hook, after asking which channel loads the screen. Running it again repeats the check; `--fresh` sets it up again.
66
+ - New `recovery-probe doctor`. It checks Node.js, Playwright, Chromium, the app on the debugging port and the IPC hook, and prints the fix for anything missing, including the hook code with the path to load it from. `--json` prints the report.
67
+ - The IPC hook's `identify()` returns `channels`, the channels it may fail. `doctor` names the ones the app never registered, which usually means a typo or a hook loaded after the handlers.
68
+ - Agents can run `doctor` through the MCP tool `run_recovery_probe`, and the server tells them to when a check can't attach.
69
+
70
+ ## 0.6.0-preview.6 (2026-10-23)
71
+
72
+ - A shorter README, and MCP setup for any agent instead of one client.
73
+ - Discovery over IPC and HTTP is described once, in `docs/discovery.md`. The docs describe current behaviour; version history stays in this changelog.
74
+ - The package ships every doc, including `docs/verify.md`, which was missing.
75
+ - A docs test checks that relative links and headings resolve.
76
+
77
+ ## 0.6.0-preview.5 (2026-10-23)
78
+
79
+ - IPC and HTTP discovery count how often each healthy load reads the target. When every load reads it the same number of times, they also fail the 2nd, 3rd and 4th call of a load on its own, as findings such as `rejection-1-call2` with a `call` field. `readiness.callsGap` says when later calls were not failed on their own. IPC discovery needs the hook from this version for these experiments; restart the app after upgrading.
80
+ - `ipc --discover --screenshots` and `web --attach --discover --screenshots` save the screen after the first baseline and each faulted run, and show the screenshots in `report.html`.
81
+ - The IPC hook's `begin` takes `skip`, the calls on the channel that pass before the first fault.
82
+
83
+ ## 0.6.0-preview.4 (2026-10-23)
84
+
85
+ - A click on a `<label>` records its action once. The browser also clicks the label's control, which counted as a second click on the same action.
86
+
87
+ ## 0.6.0-preview.3 (2026-10-23)
88
+
89
+ - New `recovery-probe verify` and MCP tool `verify_change`. A plan lists the checks that prove a change; `--baseline` runs them before it, and `verify` runs them again, compares the runs and scans the lines changed since the base for suspects. The change is VERIFIED only when every check improved or stayed clean with complete runs on both sides and no changed line is a suspect.
90
+ - `scanForSuspects` takes `changedLinesOnly`, which leaves out suspects on lines the change did not touch.
91
+ - File names from `suspects --changed-since` keep the spelling of the working directory, such as a Windows short name, instead of appearing as `../../..` paths.
92
+
93
+ ## 0.6.0-preview.2 (2026-10-23)
94
+
95
+ - `UNEXPECTED_ROW_STATE` ends a stretch when the row leaves the region, so a row that comes back gets a new finding instead of extending the old one.
96
+ - Marker labels cannot be `none` or `other`, the states rows get without a marker or allowlisted text.
97
+
98
+ ## 0.6.0-preview.1 (2026-10-23)
99
+
100
+ - Observation rows can take their state from a marker element, such as an icon, with `rows.markers`. `state` and `states` are optional when markers are set.
101
+ - `rows.unexpected` lists states a region must not show. A row shown in one of them is reported as `UNEXPECTED_ROW_STATE`, once per stretch of samples, with the samples that showed it and the action before it.
102
+
103
+ ## 0.5.0 (2026-10-21)
104
+
105
+ Promotes previews 1–22 to the stable npm `latest` channel. The preview entries below retain the detailed history.
106
+
107
+ - Synchronize the brief-content observation fixture with completed samples so a slow sampler cannot race its restoration timer. Keep coverage of transitions entirely between samples.
108
+ - Complete the TypeScript migration of the library and CLI. Published JavaScript and declarations are built into `dist/`; public package imports and the `recovery-probe` command remain available.
109
+ - Add IPC action checks for simulated success, failure and double-clicks, delayed injected results, and scripted fill/press steps.
110
+ - Add a host-level HTTPS fault proxy and network steps during observation, plus evidence of local files removed by the app.
111
+ - Add MCP tools for coding agents, experimental source scanning with `suspects`, before/after report comparison, and `watch` to rerun checks after saves.
112
+ - Add experimental Android WebView attachment and Meta Quest keep-awake support, with diagnostics for visible pages that are not rendering.
113
+
114
+ Upgrading from 0.4.0: keep using the public package exports and CLI. If `RECOVERY_PROBE_IPC_MODULE` stores the old `src/ipc-probe.cjs` path, resolve it again with `node -p "require.resolve('recovery-probe/ipc')"`; it now points into `dist/`. Restart the development app after upgrading so its loaded IPC hook matches the CLI. The optional TypeScript peer dependency is needed only for source scanning. Experimental workflows remain experimental in this stable package.
115
+
116
+ ## 0.5.0-preview.22 (2026-10-21)
117
+
118
+ - The command line is written in TypeScript under `src/cli` and compiled to `dist/cli`. `bin/recovery-probe.mjs` is now a launcher for it; the `recovery-probe` command, its options and its output are unchanged.
119
+ - `npm run demo` runs `dist/cli/core-demo.mjs`.
120
+
121
+ ## 0.5.0-preview.21 (2026-10-21)
122
+
123
+ - The opt-in repair workflow (`recovery-probe repair`) is written in TypeScript. Its files, prompts and checks are unchanged.
124
+ - All of the library source is now TypeScript, and every published declaration is generated from it.
125
+
126
+ ## 0.5.0-preview.20 (2026-10-21)
127
+
128
+ - HTTP discovery (`web --attach --discover`), the guided setup (`recovery-probe start`) and its content picker are written in TypeScript. Their reports and prompts are unchanged.
129
+
130
+ ## 0.5.0-preview.19 (2026-10-21)
131
+
132
+ - The package root (`withFault`, `probeRecovery`), `recovery-probe/desktop` and `recovery-probe/web` are written in TypeScript. Their reports are unchanged.
133
+ - Their declarations are generated from the source instead of written by hand. They accept everything the old ones did and add `ValidDesktopConfig`, `DesktopDiscovery`, a typed `ProbeError` constructor, and the `baselineTimeoutMs`, `recoveryTimeoutMs` and `times` fields that `DesktopReport` already carried.
134
+ - The root declarations moved from `dist/index.d.ts` to `dist/index.d.mts`. The `types` fields point to the new file.
135
+
136
+ ## 0.5.0-preview.18 (2026-10-21)
137
+
138
+ - Refresh experiments (`recovery-probe refresh`) are written in TypeScript. Their reports are unchanged.
139
+
140
+ ## 0.5.0-preview.17 (2026-10-21)
141
+
142
+ - **`recovery-probe android --keep-awake`** keeps a Meta Quest's display awake and pauses the Guardian boundary until the forward stops, then restores both. Checks can then run on a headset that nobody wears.
143
+ - `android` warns when the display is asleep or the Guardian dialog has focus. The page then draws no frames, and clicks time out.
144
+ - Checks that record the window state now tell a visible page apart from one that only reports itself visible: `visible-not-rendering` (no frames) or `visible-without-viewport` (no size). They also say so on the terminal.
145
+ - Checked on a Quest 2: discovery and a recovery check on a production app's WebView, and clicks with `--keep-awake`. See docs/android.md.
146
+
147
+ ## 0.5.0-preview.16 (2026-10-20)
148
+
149
+ - **`recovery-probe watch [PATH...] [-- CHECK]`** watches source files while you work. After each save it scans the changed files for suspects and reruns the check against the running development app. It compares the run with the first one and the one before, by the rules of `compare`.
150
+ - The check runs once at the start for the starting point. `--settle` waits for the dev server to reload, and each run gets its own folder and log.
151
+ - `summary.json` and `summary.md` (the pull request block for the first run against the latest) are rewritten after every run.
152
+ - Without a check, the watch only scans for suspects. See docs/watch.md.
153
+
154
+ ## 0.5.0-preview.15 (2026-10-20)
155
+
156
+ - **`recovery-probe compare`** compares reports of the same check from before and after a change and says what happened to each problem: gone, new, still there, or not checked to the end. A problem counts as gone only when a complete run after the change checked it, so a run that stopped early never reads as a fix.
157
+ - Each side takes one or more reports, report folders or folders of runs, so repeated runs can be compared.
158
+ - It returns a verdict: `IMPROVED`, `PARTLY_IMPROVED`, `REGRESSED`, `MIXED`, `UNCHANGED`, `INCONCLUSIVE` or `NOT_COMPARABLE`. It exits 1 only for a new suspected defect.
159
+ - `--markdown` prints a block for the pull request. See docs/compare.md.
160
+ - The MCP server adds `compare_reports`, and tells agents to compare reports before they say a change fixed something.
161
+
162
+ ## 0.5.0-preview.14 (2026-10-20)
163
+
164
+ - **`recovery-probe suspects`** (experimental) reads JavaScript and TypeScript sources for error handling that loses data:
165
+ - `UNKNOWN_AS_EMPTY`: a failed or unexpected read becomes an empty collection;
166
+ - `FAILURE_ONLY_LOGGED`: a read stores its data when it succeeds, but its failure is only logged.
167
+ - Each suspect names the read to fail and the check that proves or dismisses it. `--changed-since main` scans only what a branch changed, and `// recovery-probe-ignore` marks a reviewed line. It exits 2 when there are suspects to review. See docs/suspects.md.
168
+ - The MCP server adds `find_suspects`.
169
+ - TypeScript is a new optional peer dependency, needed only by the scanner.
170
+
171
+ ## 0.5.0-preview.13 (2026-10-20)
172
+
173
+ - **`recovery-probe android`** (experimental) makes an Android app's WebView, or Chrome, inspectable like a desktop debugging endpoint. It finds the DevTools socket over adb, forwards a local port, checks that DevTools answers and removes the forward on exit. The printed URL works with `desktop --cdp` and as `cdp` in configs. It is served through a small bridge that hides blank WebViews, which Android pauses and which never answer, so a client does not wait on them. Checked on a Quest 2. See docs/android.md.
174
+
175
+ ## 0.5.0-preview.12 (2026-10-20)
176
+
177
+ - **`recovery-probe mcp`** serves Recovery Probe to AI coding agents as MCP tools over stdio:
178
+ - `find_page_reads` and `check_read_recovery` for a page's reads;
179
+ - four tools that run the network fault proxy;
180
+ - `run_recovery_probe` for observe, ipc, refresh, desktop and web.
181
+ - When an agent connects, the server tells it how to report results: a pass covers only what was checked, and inconclusive codes are not passes.
182
+ - A caught fault returns the report folder with `fix-brief.md` and `reproduce.mjs`.
183
+ - No new dependencies. See docs/mcp.md.
184
+ - The package description now says the checks can run from an AI coding agent.
185
+
186
+ ## 0.5.0-preview.11 (2026-10-20)
187
+
188
+ - **`observe --files DIR`** compares the files under a directory at each interval and reports `LOCAL_FILES_REMOVED` when an interval lost files, with counts and bytes in the report and the paths in a private `files.json`. It catches a failed read that made the app delete installed content, which the screen alone does not show. See docs/observe.md.
189
+
190
+ ## 0.5.0-preview.10 (2026-10-20)
191
+
192
+ - **Network steps** for `observe --steps`: with `--network URL`, a step can take hosts away from the app's own HTTPS connections through a running `recovery-probe net`, and `"restore"` brings them back. The capture always removes the faults at the end. The report records what each step faulted, adds `NETWORK_FAULT_NOT_TRIGGERED` or `NETWORK_FAULTS_NOT_RESTORED` coverage gaps, and labels the mode `observation with network faults`. See docs/observe.md.
193
+ - `recovery-probe net` runs without `--rules` and answers a loopback control; `recovery-probe/net` adds `connectNetworkFaults` and the `control` option.
194
+
195
+ ## 0.5.0-preview.9 (2026-10-20)
196
+
197
+ - **`recovery-probe net`** fails an app's own HTTPS connections, for requests made outside the page by an SDK, a main process or a command-line tool. It runs a loopback proxy for the app's `HTTPS_PROXY` and refuses, resets, stalls or delays connections to the hosts you name, in time windows counted from the app's first connection. It can close open connections so clients have to reconnect. It sees only hosts and ports: HTTPS stays end to end. See docs/net.md.
198
+ - `recovery-probe/net` exports the same proxy as `startNetworkFaultProxy`.
199
+
200
+ ## 0.5.0-preview.8 (2026-10-16)
201
+
202
+ - **`ipc --action`** checks what the user sees after an action that writes. It runs scripted setup steps, clicks the action while the hook answers its write channel, and compares a simulated success, a simulated failure and a double-click. It reports `SILENT_FAILURE`, `FALSE_SUCCESS`, `UNCLEAR_FEEDBACK`, `UNCAUGHT_ERROR` and `DUPLICATE_SUBMIT`. Nothing is written. See docs/ipc.md.
203
+ - Steps gain `fill` and `press`, also available to `observe --steps`. Typed values are never written to reports.
204
+ - The IPC hook can delay injected results with `delayMs` to simulate request time; `identify` reports this as `capabilities.faultDelay`.
205
+ - Discovery waits for the app's pending calls before each run, so its cleanup no longer fails when an app retry is still in flight.
206
+
7
207
  ## 0.5.0-preview.7 (2026-10-16)
8
208
 
9
209
  - The IPC check and IPC discovery (`recovery-probe ipc` and `ipc --discover`) are written in TypeScript. Their reports are unchanged.
package/CONTRIBUTING.md CHANGED
@@ -16,18 +16,35 @@ npm ci
16
16
  npm run format:check
17
17
  npm run lint
18
18
  npm run typecheck
19
- npm test
20
- npm run test:package
19
+ npm test # unit tests, no browser
20
+ npm run test:package # installs the packed package like a user would
21
21
  npx playwright install chromium
22
- npm run test:browser
23
- npm run test:repair # safety, authenticated portal, before/after pipeline
24
- # After installing the optional Electron test runtime:
25
- npm run test:repair:electron
22
+ npm run test:integration # real Chromium
23
+ npm install --no-save --package-lock=false electron@44.5.1
24
+ npm run test:electron # real Electron
26
25
  ```
27
26
 
27
+ Tests live in `test/unit`, `test/integration`, `test/electron` and `test/package`, and each script runs every `*.test.mjs` in its folder. Shared helpers are in `test/helpers` (free ports, launching an Electron fixture, the repair fixture) and fixture apps in `test/fixtures`. Electron tests start Electron through `test/helpers/electron.mjs`, which also clears `ELECTRON_RUN_AS_NODE` (some editors set it) and stops the whole process tree.
28
+
28
29
  Repair integration tests use a **controlled mock model response** but real apps, fault injection, source changes and reruns. They validate orchestration, not live AI diagnosis quality.
29
30
 
30
- Code and JSON are formatted with Prettier (`npm run format`) and linted with ESLint (`npm run lint`); CI fails on either. Functions passed to `page.evaluate` run in the browser: keep them self-contained, and list new in-page modules in `eslint.config.mjs` so they get browser globals. The package is moving from JavaScript to TypeScript, one module at a time; write new modules as `.mts` (or `.cts` for CommonJS). `npm run build` compiles them into `dist` with strict checks and generated declarations, and copies the remaining JavaScript, hand-written declarations and report templates unchanged. `npm ci` builds once, and every script that runs code builds first, so tests never use a stale `dist`. `npm run typecheck` also checks the JavaScript in `src`, `bin` and `examples` with strict mode off; add JSDoc types where inference is not enough. TypeScript is pinned to 6.0 because typescript-eslint, which lints the TypeScript files, does not support 7 yet. Markdown, HTML fixtures, report templates and workflow files are left as written. Report markup lives in `src/report-templates`: `{{slot}}` takes text and escapes it, `{{{slot}}}` takes markup built with `escapeHtml`. `.gitattributes` keeps LF line endings on every platform so local checks match CI. The repository-wide reformat is listed in `.git-blame-ignore-revs`; run `git config blame.ignoreRevsFile .git-blame-ignore-revs` to skip it in local blame.
31
+ Code and JSON are formatted with Prettier (`npm run format`) and linted with ESLint (`npm run lint`); CI fails on either. Functions passed to `page.evaluate` run in the browser: keep them self-contained, and list new in-page modules in `eslint.config.mjs` so they get browser globals. The package is written in TypeScript: modules are `.mts` (or `.cts` for CommonJS), and the command line lives in `src/cli`, with `bin/recovery-probe.mjs` only starting it. `npm run build` compiles them into `dist` with strict checks and generated declarations, and copies the report templates unchanged. `npm ci` builds once, and every script that runs code builds first, so tests never use a stale `dist`. `npm run typecheck` also checks the JavaScript in `bin`, `examples` and `scripts` with strict mode off; add JSDoc types where inference is not enough. TypeScript is pinned to 6.0 because typescript-eslint, which lints the TypeScript files, does not support 7 yet. Markdown, HTML fixtures, report templates and workflow files are left as written. Report markup lives in `src/reports/templates`: `{{slot}}` takes text and escapes it, `{{{slot}}}` takes markup built with `escapeHtml`. `.gitattributes` keeps LF line endings on every platform so local checks match CI. The repository-wide reformat is listed in `.git-blame-ignore-revs`; run `git config blame.ignoreRevsFile .git-blame-ignore-revs` to skip it in local blame.
32
+
33
+ ## Layout
34
+
35
+ `src/` is split into layers, and each layer imports only itself and the layers below it:
36
+
37
+ | Folder | Holds |
38
+ | --- | --- |
39
+ | `core` | Fault kinds, URL identity, reading and comparing the screen, package paths |
40
+ | `transports` | Failing reads: Playwright routes, the fetch wrapper, the HTTPS proxy, the IPC inspector and control |
41
+ | `targets` | Reaching the app: launching Chromium, attaching over CDP, window visibility, Android |
42
+ | `reports` | Report templates and the writers for check, discovery and action reports |
43
+ | `workflows` | The experiments: checks, discovery, refresh, observation, actions |
44
+ | `tools` | What's built on them: init, doctor, guided start, suspects, compare, watch, verify, MCP, repair |
45
+ | `cli` | Argument parsing and output |
46
+
47
+ The files directly in `src/` are the package's public entry points. They only re-export, so their import paths never change; `ipc-probe.cts` is the hook apps load by path. `test/unit/layers.test.mjs` fails when a module imports a layer above it at run time; type imports are free.
31
48
 
32
49
  ## Install an unpublished source preview
33
50
 
package/README.md CHANGED
@@ -1,109 +1,125 @@
1
1
  # Recovery Probe
2
2
 
3
- **Test whether web and desktop apps show the right data.**
3
+ See what your app shows when a data request fails.
4
4
 
5
- Apps can show the wrong data without any error: a list that disappears after one failed read, a cached screen that never picks up a change, or a row whose state lags behind data the app already received. Recovery Probe reproduces these cases on demand. It injects bounded faults into data reads and compares the screen with a healthy run, or it watches reads and the screen without any fault. Each run writes a local report that separates what it measured from what it suspects.
6
-
7
- Installing the package does not add recovery behavior to your application.
5
+ Most apps don't crash when a read fails. The list goes empty, the spinner never stops, or the screen keeps showing old data. There's no error and nothing in the logs, so it ships. Recovery Probe fails those reads on purpose, runs the same screen healthy for comparison, and tells you what changed.
8
6
 
9
7
  [![CI](https://github.com/BojanKovachki/recovery-probe/actions/workflows/validate.yml/badge.svg)](https://github.com/BojanKovachki/recovery-probe/actions/workflows/validate.yml)
10
8
  [![npm](https://img.shields.io/npm/v/recovery-probe.svg)](https://www.npmjs.com/package/recovery-probe)
11
9
  [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
12
10
 
13
- ## See it work
11
+ ![Recovery Probe catches a broken retry flow and passes the fixed one](docs/demo.svg)
14
12
 
15
- The package includes a deliberately broken app and a corrected one. Both pass the normal loading check; only the corrected one recovers from all three faults.
13
+ ## Try it
16
14
 
17
- ![Recovery Probe catches a broken retry flow and verifies the corrected flow](docs/demo.svg)
15
+ You need Node.js 22 or newer.
18
16
 
19
17
  ```bash
20
- mkdir recovery-probe-tools && cd recovery-probe-tools
21
- npm install --save-dev --save-exact recovery-probe@0.4.0 playwright@1.62.1
18
+ npm install --save-dev recovery-probe playwright
22
19
  npx playwright install chromium
23
- npx recovery-probe --demo
20
+ npx recovery-probe demo
24
21
  ```
25
22
 
26
- `--demo` exits successfully only when it observes that contrast: the happy path passes, the broken recovery is caught, and the corrected flow passes. Add `--json` to print the full report or `--out report.json` to save it.
23
+ The demo runs two versions of a small app. Both load fine. Only the fixed one survives a 503, a dropped connection and a broken JSON response.
27
24
 
28
- ## Test your own app
25
+ ## Check your own app
29
26
 
30
- | Goal | Start with | Guide |
31
- | --- | --- | --- |
32
- | Set up a first recovery check interactively | `recovery-probe start web URL` or `start desktop` | [Guided start](docs/start.md) |
33
- | Discover and test the reads behind a screen in a browser or Electron window | `recovery-probe desktop --cdp URL` | [Attachment and HTTP discovery](docs/desktop.md) |
34
- | Compare an Electron screen under IPC faults | `recovery-probe ipc --discover --config FILE` | [IPC setup and discovery](docs/ipc.md) |
35
- | Test a refresh after content has already loaded | `recovery-probe refresh --config FILE --json` | [Refresh experiments](docs/refresh.md) |
36
- | Watch cached content and row states after user actions, without faults | `recovery-probe observe --config FILE [--steps FILE]` | [Observation](docs/observe.md) |
37
- | Run a repeatable recovery check in CI | `npm run test:ci-example` (source checkout) | [CI example](docs/ci.md) |
38
- | Assert recovery inside your own tests | `withFault` or `createFaultFetch` | [API](docs/api.md) |
39
- | Propose and verify a source change | Opt-in repair workflow | [Repair setup and limits](docs/repair.md) |
27
+ **Web app.** Start it, then run:
40
28
 
41
- `desktop` also attaches to an existing Chromium-based web browser.
29
+ ```bash
30
+ npx recovery-probe init http://localhost:3000
31
+ ```
42
32
 
43
- **Where it works:**
33
+ A browser opens. Sign in, go to the screen you care about, and click something that only shows up once its data has loaded. Recovery Probe finds the request behind it, fails it three ways, and tells you whether the data came back.
44
34
 
45
- - **Web apps** in Chromium-based browsers: launch a test browser or attach to an authorized development one.
46
- - **Electron apps that read over HTTP** in the renderer: attach through a loopback debugging endpoint.
47
- - **Electron apps that read over IPC**: add a development-only hook to the main process.
48
- - **JavaScript code that takes a fetch function**: use `recovery-probe/fetch`, without Playwright.
35
+ **Electron app.** Start it with `--remote-debugging-address=127.0.0.1 --remote-debugging-port=9222`, then run `npx recovery-probe init`.
49
36
 
50
- Desktop support means Electron, not arbitrary native applications. IPC faults do not exercise native networking, HTTP status handling or token refresh. An unsupported data path is inconclusive, never a successful recovery check.
37
+ **Electron app that loads data over IPC.** Load the small development-only hook in the main process. `npx recovery-probe doctor` prints the code and the path to load it from. Restart the app with the inspector on, then run `npx recovery-probe init`: it lists the hook's channels, asks which one loads the screen, and fails it in a few ways.
51
38
 
52
- ## Add a recovery check to a Playwright test
39
+ Each run saves an HTML and JSON report. Run the same command again to repeat the check, or add `--fresh` to set it up again. If something's missing, `npx recovery-probe doctor` says what and how to fix it.
53
40
 
54
- Use `withFault` inside an existing Playwright test. The callback contains the real user action and the application-specific recovery assertion.
41
+ ## More
55
42
 
56
- ```js
57
- import { test, expect } from '@playwright/test';
58
- import { withFault } from 'recovery-probe';
59
-
60
- test('profile recovers after a transient 503', async ({ page }) => {
61
- const stats = await withFault(page, {
62
- match: '**/api/profile',
63
- kind: 'http-error',
64
- status: 503,
65
- count: 1,
66
- }, async () => {
67
- await page.goto('http://127.0.0.1:3000/profile');
68
- await page.getByRole('button', { name: 'Retry' }).click();
69
- await expect(page.getByTestId('profile-name')).toHaveText('Test User');
70
- });
71
-
72
- expect(stats.applied).toBe(1);
73
- });
43
+ | To | Run | Docs |
44
+ | --- | --- | --- |
45
+ | Fail the read behind a screen without writing assertions | `discover` | [Discovery](docs/discovery.md) |
46
+ | Catch stale screens and rows stuck in an old state, without faults | `observe` | [Observe](docs/observe.md) |
47
+ | See what users get when Save or Send fails | `action` | [Action check](docs/ipc.md#action-check) |
48
+ | Fail an SDK's or a main process's own HTTPS connections | `net` | [Network faults](docs/net.md) |
49
+ | Find code that turns a failed read into "no data" | `suspects` | [Suspects](docs/suspects.md) |
50
+ | Prove a change fixed it | `compare`, `watch`, `verify` | [Compare](docs/compare.md), [Watch](docs/watch.md), [Verify](docs/verify.md) |
51
+
52
+ ## Use it from a coding agent
53
+
54
+ `recovery-probe mcp` is an MCP server. Any agent that supports MCP can use it to fail a read and look at the screen before it claims the code handles errors. Most agents take this config:
55
+
56
+ ```json
57
+ {
58
+ "mcpServers": {
59
+ "recovery-probe": { "command": "npx", "args": ["recovery-probe", "mcp"] }
60
+ }
61
+ }
74
62
  ```
75
63
 
76
- If no request matches, `withFault` throws `FAULT_NOT_TRIGGERED` instead of letting the test pass silently. Code that takes a fetch function can use `createFaultFetch` instead. See the [API reference](docs/api.md) and the [JSON-configured runner](docs/runner.md).
64
+ Other config formats and the list of tools are in [MCP](docs/mcp.md).
65
+
66
+ ## Use it in your tests
67
+
68
+ `withFault` fails matching requests inside a Playwright test:
69
+
70
+ ```js
71
+ import { test, expect } from '@playwright/test'
72
+ import { withFault } from 'recovery-probe'
73
+
74
+ test('profile recovers after a 503', async ({ page }) => {
75
+ const stats = await withFault(
76
+ page,
77
+ { match: '**/api/profile', kind: 'http-error', status: 503 },
78
+ async () => {
79
+ await page.goto('http://127.0.0.1:3000/profile')
80
+ await page.getByRole('button', { name: 'Retry' }).click()
81
+ await expect(page.getByTestId('profile-name')).toHaveText('Test User')
82
+ },
83
+ )
84
+ expect(stats.applied).toBe(1)
85
+ })
86
+ ```
77
87
 
78
- ## Install
88
+ If nothing matches, `withFault` throws `FAULT_NOT_TRIGGERED` instead of passing. For code that takes a `fetch` function there's `createFaultFetch`. See [API](docs/api.md).
79
89
 
80
- **0.4.0** is the stable release on npm's `latest` channel, and the commands above pin it for reproducible reports. Prereleases use the `next` channel and are opt-in.
90
+ ## How it decides
81
91
 
82
- Use Node.js 22 or newer. A separate tools folder keeps the tester out of your application's production dependencies. Browser and desktop commands need Playwright; attached targets use their existing browser runtime. If a version is not on npm yet, use the [source installation](CONTRIBUTING.md#install-an-unpublished-source-preview): a GitHub merge alone does not publish the package. See [CHANGELOG.md](CHANGELOG.md) for version history.
92
+ - Every faulted run has a healthy run right before it, on the same screen, and the report compares the two.
93
+ - A result is a pass, a failure or inconclusive. A fault that never fired, an unstable baseline or a path it can't reach is inconclusive, never a pass.
94
+ - Some findings, like `SILENT_EMPTY`, are leads. They tell you where to look, not that the code is wrong.
95
+ - Exit codes: 0 passed, 1 a check failed, 2 needs review or inconclusive, 3 wrong usage or config, 4 couldn't run.
83
96
 
84
- ## Reports, agents and privacy
97
+ ## Where it works
85
98
 
86
- Reports separate measured facts, such as fault counts, successful reads, timings and UI changes, from suspected defects. A `CONTENT_LOSS` finding needs review. An unconsumed fault, an unstable control or an unsupported path is inconclusive. Observation findings such as `DELAYED_UI_UPDATE` are leads to check against the application logic, not verdicts.
99
+ - Web apps in Chromium, either launched by Recovery Probe or attached through a local debugging port.
100
+ - Electron apps, over HTTP in the renderer or over IPC with the hook.
101
+ - Native HTTPS clients that use `HTTPS_PROXY`, such as an SDK in the main process. Only whole connections can fail there.
102
+ - Android WebViews, experimental.
87
103
 
88
- Coding agents can run saved scenarios, read the JSON evidence, propose a change and rerun the same experiment. Recovery Probe supplies repeatable experiments and evidence; it is not a general bug detector or a production self-healing system.
104
+ It doesn't simulate a whole device going offline, WebSocket drops, token expiry, sleep or slow networks.
89
105
 
90
- Reports stay local, and Recovery Probe sends no telemetry. Scenarios, endpoints, channels, selected text, screenshots, login state and command logs may be private, so do not publish them blindly. Request and response bodies and headers are not collected, and observation never records read contents.
106
+ ## Privacy
91
107
 
92
- Tests and discovery do not call an AI service. Only the opt-in repair workflow, with `--allow-source-upload`, sends the selected source files and bounded observations to your configured OpenAI model; login state and command logs are not part of that upload. It sets `store: false`, which is not a promise of zero provider retention, so review your employer's policy and the provider's data terms first. A repair is a reviewable candidate, not a guaranteed fix.
108
+ Reports stay on your machine and there's no telemetry. Request and response bodies aren't recorded, and discovery and observation keep screen text only as hashes. Reports can still contain endpoint paths, selectors, text you picked as a marker and, with `--screenshots`, what was on screen, so check them before you share them. Nothing calls an AI service except the opt-in [repair](docs/repair.md) workflow.
93
109
 
94
- ## Scope
110
+ Only test apps and accounts you're allowed to, and keep debugging ports on localhost.
95
111
 
96
- Recovery Probe checks request-level recovery deterministically and observes whether the screen reflects the reads it received. It does not simulate whole-device offline states, WebSocket interruptions, token expiry, browser sleep, slow networks or duplicate submissions.
112
+ ## Upgrading from 0.5.0
97
113
 
98
- Structural discovery compares healthy and faulted screens without a hand-written content selector, but it needs a selected read and a stable healthy baseline. No workflow infers your product's intended recovery policy.
114
+ Restart the development app after upgrading so its IPC hook matches the CLI. Public package imports and the hook's path are unchanged. The old command spellings still work until 1.0, with a migration notice; new commands distinguish usage errors (exit 3) and setup failures (exit 4) from inconclusive results (exit 2). Update CI scripts that interpret exit codes. With the new `demo` and JSON-runner `check` commands, `--out` names a new report folder; their old spellings still write one file. See the [changelog](CHANGELOG.md) for the full migration notes.
99
115
 
100
- The recovery, discovery and refresh workflows reload the renderer to establish controls. That clears in-memory caches, so those modes cannot prove account isolation. Observation keeps that state, but it does not infer identities or data ownership.
116
+ ## Docs
101
117
 
102
- Attach only to authorized test sessions, keep debugging endpoints on loopback, and keep the target window visible and uncovered.
118
+ [Guided start](docs/start.md) · [Desktop and browsers](docs/desktop.md) · [IPC](docs/ipc.md) · [Discovery](docs/discovery.md) · [Observe](docs/observe.md) · [Refresh](docs/refresh.md) · [Network faults](docs/net.md) · [Android](docs/android.md) · [Suspects](docs/suspects.md) · [Compare](docs/compare.md) · [Watch](docs/watch.md) · [Verify](docs/verify.md) · [MCP](docs/mcp.md) · [API](docs/api.md) · [JSON runner](docs/runner.md) · [CI](docs/ci.md) · [Repair](docs/repair.md) · [Changelog](CHANGELOG.md)
103
119
 
104
- ## Development
120
+ ## Contributing
105
121
 
106
- See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, checks and test suites, and [SECURITY.md](SECURITY.md) for reporting security issues. CI runs package checks on Node.js 22 and 24, Chromium integration, and real Electron checks.
122
+ See [CONTRIBUTING.md](CONTRIBUTING.md). Report security issues as described in [SECURITY.md](SECURITY.md).
107
123
 
108
124
  ## License
109
125