agentleFS
advenimusadvenimus/conduit-desktop/.claude/skills/verify/SKILL.md

verify

Skill19 starsChanged 49 days agoApache-2.0View on GitHub

Run every live end-to-end suite (backup, copies, lifecycle, mcp, ownership, password, resilience, smoke, sync; 50 scenarios) against the real Conduit app and a local Appwrite, then summarize pass or fail. Use only when the user explicitly asks for every live suite (for example "run all the live tests" or "run the full verify") or for a release candidate. For a change in one area, use the targeted skill instead: verify-sync, verify-mcp, verify-ownership or verify-data. Fully isolated from the user's dev app and data; takes about 12 minutes and opens many app windows.

  • Reads credentials

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use genie-hacks, workfly and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • automagik-devSkill

    genie-hacks

    Browse, search, and contribute community hacks — real-world patterns for provider switching, teams, skills, hooks, cost optimization, and more.

    Also in 3 projects that use this

  • automagik-devSkill

    workfly

    Discover a procedure and build its saved workflow — dynamic discovery, drafted script, adversarial verification, landed in the catalog.

    Also in 3 projects that use this

  • TheKREATORSSkill

    instinct-export

    Export instincts from project/global scope to a file

    Also in 3 projects that use this

  • TheKREATORSSkill

    instinct-import

    Import instincts from file or URL into project/global scope

    Also in 3 projects that use this

  • AviadPSkill

    risk-assess

    Assess the risk and blast radius of a proposed code change before implementing it. Evaluates CI impact, framework-wide vs localized scope, rollback path, and identifies blindspots. Use when user says "what's the risk", "is this safe", "assess the impact", "risk assess", "blast radius", or before applying framework-wide or CI-affecting changes. Do NOT use for code quality review (use code-quality) or plan validation (use design-validator).

    Also in 3 projects that use this

  • diego-tobalinaSkill

    build-fix

    Fix build and compilation errors systematically. Run when build fails.

    Also in 6 projects that use this

  • +2 more from automagik-dev
  • +1 more from TheKREATORS

The file

We split this skill into the rules it gives an agent

10 rules. 0 are common, 4 are distinctive, and 6 are very specific to this project.

  • 0

    Common

    Most of the 168 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run every suite only when the user asks for it or for a release candidate.

    line 13
  2. Very specific:

    Every scenario checks the shared file: two key epochs with the old salt and verifier redacted, a WAL header, the iOS 1.0.5 open, and no …

    line 58
  3. Very specific:

    Never make a run pass by weakening an assertion, deleting a check, skipping a scenario or raising a timeout without a measured reason.

    line 167
  4. Distinctive:

    If a scenario passes only sometimes, treat it as a real race and report it.

    line 170
  5. Very specific:

    Only the local stack on localhost.

    line 214
  6. Very specific:

    Never touch the user's dev profile (~/Library/Application Support/conduit/conduit-dev, the "Conduit Dev" app), their running dev app or its …

    line 216
  7. Distinctive:

    Never answer a macOS keychain prompt.

    line 218
  8. Very specific:

    Only kill processes that belong to the run (their command line mentions /tmp/cv-<id> or .verify/<runId>).

    line 219
  9. Very specific:

    Never run docker compose down -v on the local stack or delete its env file unless the user asks: that wipes the stack.

    line 221
  10. Distinctive:

    Never print the API key, the service secret or session tokens.

    line 223

“Common” is measured: kept by at least 6 in 10 of the 168 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/verify: all live suites

Runs npm run verify. The harness in scripts/verify/ builds the app, launches real Electron devices side by side, drives them like a user (and over MCP), checks the server state in the local Appwrite, and cleans up after itself. You can run it start to finish without asking the user for anything. The harness reference is scripts/verify/README.md.

Run every suite only when the user asks for it or for a release candidate. For any other change, run only the suite that covers it, with its skill below; they add per-scenario detail and suite-specific troubleshooting:

SkillSuitesCommand
verify-sync (.claude/skills/verify-sync/SKILL.md)syncnpm run verify:sync
verify-mcp (.claude/skills/verify-mcp/SKILL.md)mcpnpm run verify:mcp
verify-ownership (.claude/skills/verify-ownership/SKILL.md)ownershipnpm run verify:ownership
verify-data (.claude/skills/verify-data/SKILL.md)backup, copies, lifecycle, password, resiliencenpm run verify:data, or node scripts/verify/run.mjs <suite>

Smoke has no skill of its own (node scripts/verify/run.mjs smoke).

What it proves

smoke (harness health)

  • two-devices: a Free user creates a vault in the shared cloud folder; a second device in local mode shows the vault hub.
  • ipc-and-mcp: in local mode an entry is created, renamed and deleted, and an MCP client on the device sees each change.

sync (personal vault sync and device limits, docs/MULTI_DEVICE_SYNC.md)

  • free-takeover-server: on Free, device B opens a vault A has open, is offered a take-over, takes it and gets A's data; A is locked out and offered "Use here instead".
  • take-back: A presses "Use here instead" and gets the vault back; B is locked out.
  • pro-merge-and-conflict: on Pro both devices stay open; edits to different fields merge, and a same-field edit and an edit-versus-delete become conflicts that are resolved in the review panel, ending the same on both devices.
  • downgrade-and-upgrade: Pro to Free with two devices open locks out exactly one, with the plan-limit message; back to Pro it reopens with no take-over and the other device stays open.
  • signout-releases-lease: signing out frees the device's server lease within seconds.
  • signed-out-claims: with nobody signed in, the owner claim inside the vault file still drives the take-over and locks A out.
  • legacy-writer: an in-place edit by an old desktop build (0.17) is absorbed with no conflict.
  • file-safety: no -wal, -shm or -journal file ever appears in the cloud folder during the suite, and every vault file has a WAL header and opens the way iOS 1.0.5 opens it.

mcp (MCP tools on real devices)

  • tools-list: tools/list has the vault tools and entry_list works on an unlocked vault.
  • no-daily-quota: a Free user makes 60 tool calls (the old cap was 50); none is refused and no quota file is written.
  • writes-sync: on Pro, MCP note and document writes show on the device and reach a second device.
  • has-conflict: a same-field conflict makes entry_info report has_conflict: true until it is resolved in the review panel.
  • locked-and-elsewhere: a locked vault returns code: VAULT_LOCKED; after a Free take-over from another device it returns reason: open_elsewhere and the device shows the locked-out dialog.
  • audit: every MCP call of the run is in the MCP audit log, in order, with the right outcome; secrets such as api_key show as [REDACTED].

ownership (vault ownership and plan limits, docs/PLAN_ENFORCEMENT.md; npm run verify:ownership, about 3 minutes)

  • Owner and grace banners, grace ending with [Make my own copy], release and hand-over, release surviving a background re-acquire, the owner tag signed out and offline, the account device cap, the minimum app version, and the cloud backup plan gate. Details and wording: .claude/skills/verify-ownership/SKILL.md.

password (master-password changes on a synced vault, spec 4.7 and 4.8; node scripts/verify/run.mjs password, about 1 minute)

  • change-while-both-open: on Pro, A changes the password (Vault > Change Password...); B shows "Syncing paused" naming the device, stops publishing, takes [Later] and keeps editing (a host and a secret); the old password is refused there; with the new one B's edits reach A, both show the same entries and every secret decrypts; after a lock the old password gets "Invalid master password" on both and the new one unlocks.
  • change-while-other-closed: B stops with an unpublished edit; A changes the password; B's old password opens only B's own copy and pauses syncing (spec 4.8 unlock policy); the new password merges B's edit and A's later edit on both devices.
  • erase-recently-deleted: the change with "Also permanently delete items in Recently deleted" redacts the grave in the shared file and leaves nothing to restore on either device.
  • older-file-old-password: when the cloud drive brings back the file from before the change, the old password gets "Master password changed" naming the device; the new password there unlocks and republishes under the new key.
  • Every scenario checks the shared file: two key epochs with the old salt and verifier redacted, a WAL header, the iOS 1.0.5 open, and no -wal, -shm or -journal file.

lifecycle (vault lifecycle, spec 4.7, 5.9 and Settings > Sync; node scripts/verify/run.mjs lifecycle, about 3 minutes)

  • recently-deleted: on Pro, an entry with a password (and a password history row) deleted on A shows in Recently deleted on A and B; [Restore] on B brings it back on both with its password; deleted again and erased with "Delete permanently" on A, it leaves both lists (an "Erased permanently" record stays), the shared file's grave has row_json NULL and redacted = 1, every register is redacted, and no name, host, username, notes or password cipher (current or old) is left in the file's bytes.
  • rename-and-rebind: A renames the vault with File > Rename Vault...; A keeps working at the new path; about 33 s later B shows "Found the vault file under its new name." with [Undo], follows the new path (sync state, vault_get_path, recent vaults), both keep syncing, nothing recreates the old file, and after a relaunch B's hub offers only the renamed file and opens it.
  • make-separate-vault: opening a copy of a synced vault on the same device offers "'Vault.conduit' is a copy of ..."; [Use as a separate vault] writes a new file with a new lineage, genesis and vault id; [Open it] opens it with the same password; its edits never reach the original, which stays byte-for-byte the same.
  • damaged-working-copy: A's working copy is corrupted while A is closed and B publishes a new entry; at unlock A shows "This computer's copy is damaged"; [Rebuild from shared file] opens the vault with the shared file's data (B's entry, A's password), parks the damaged copy byte for byte, and A syncs again.
  • idle-auto-lock: "Lock the vault when idle" saved in Settings > Security persists in the dialog, settings.json and across a relaunch; with the system idle time stubbed in the main process, 299 s idle keeps the vault open, 300 s locks it at the next 30 s check and releases the lease; the OS screen lock locks at once; with the setting Off the screen lock does nothing.
  • settings-sync-tab: Settings > Sync shows "one device at a time" and limit 1 for Free, "any number of devices at once" and limit -1 for Pro, and the devices list ("(this device)", "Open now"; the other device turns "Closed" after it quits).

copies (copies of a vault, spec 5.5, 5.8 and 5.10; node scripts/verify/run.mjs copies, about 2 minutes)

  • provider-conflict-copy: on Pro, a third device's published edit lands next to the shared file as "Vault (conflicted copy ...).conduit"; A merges it within a second with the toast "Merged changes from a copy Dropbox made." and publishes it; the copy stays in place byte for byte, and B, opening later, has the edit and lists the copy as "Nothing new" with [Move to Trash].
  • user-copy-review: "Vault 2.conduit" edited by desktop 0.17 (3 entries deleted, 1 host changed) shows the notice "... 4 changes that aren't in your vault, including 3 deletions." and is never merged on its own; [Ignore this copy] and [Don't merge] after the preview ("Different values (1)", "Deleted in this copy (3)") change nothing on A, B or the shared file; [Merge] applies the edit and the deletions, except an item B changed while the preview was open, which becomes one "Deleted and edited" review item ([Keep item] keeps it); the deleted items show in Recently deleted.
  • side-files-pause: -wal/-shm next to the shared file show the side-files banner on both devices; A's new entry and edit stay on A and are not published (checked for 10 s, several times the measured normal publish time) while the status is "Paused (older Conduit open)"; [Conduit is closed on my other computers] moves both files into A's sidefiles-<ts>/ folder (not deleted), publishing resumes within 5 s (measured 0.3 s), B gets the held edits and both banners go.
  • mass-delete-undo: on Pro, A deletes a folder with 12 entries; B applies it, takes one pre-merge snapshot (on disk) and shows "A sync from another device deleted or changed 13 items." with [Review]; a single delete before takes no snapshot; after an unrelated later edit from A, Undo lists exactly the folder and its 12 entries, "Undid 13 changes." restores them with their values and folder on both devices, and keeps the later edit and the earlier single delete.
  • different-copies: on Pro, A syncs .../Dropbox/Vault.conduit and B a copy at .../OneDrive/Vault.conduit (same lineage); B shows "Two copies of this vault" at unlock (from A's presence in the copy) and A about 15 s later (from B's session row, since A's file never held B's presence); [Merge them...] with the other file on each device previews its one change, merges it and publishes, so both copies hold both edits.

backup (local and cloud backups, spec 5.10, 8.1 and 8.2; node scripts/verify/run.mjs backup, about 2 minutes)

  • local-backup-rollback: on Free, Settings > Backup turns local backup on into a folder under the run dir; two new entries make a backup file on their own about 5 s later; decrypted, it is a WAL-header SQLite file with no free pages and this vault's lineage, with no -wal, -shm, -journal or .tmp next to it. After a password change and a delete, "Restore from backup" (the app's own preview dialog, fed by local_backup_restore, since the app has no local-restore screen) lists the deleted entry and "1 newer value will be replaced"; [Roll this vault back] brings the entry and the old password back, keeps the replaced password in password history with changed_by "rollback" (and the earlier manual change), matches the backup exactly and is published. With the shared folder read-only, a backup still holds an entry that only the working copy has, so backups read the working copy.
  • rollback-syncs: the same rollback on Pro device A reaches device B (entry, password and the "rollback" history row) about 2.5 s later.
  • restore-as-new-vault: [Restore as a new vault...] writes a file with the backup's entries, a new lineage and a WAL header, without writing the original; [Open it] opens it with the same password; the original keeps its lineage, vault id, entries and history.
  • cloud-backup-plan-gate: on Free, the Cloud Backup toggle is disabled with a "Pro and Team" badge and "Upgrade to Pro or Team to back up your vault to the cloud.", cloud_sync_enable and cloud_backup_restore are refused with the plan messages, and the vaults bucket stays empty. On Pro, the toggle turns backup off and on again; the upload lands under <user id>/<vault id>/ (storage.objects), cloud_backup_list shows it, and a Backup Manager restore of it goes through the preview and brings a deleted entry back. Downgraded to Free while open, the still-shown toggle is refused with the toast "Could not change cloud backup: Cloud backup needs the Pro or Team plan." and nothing is uploaded.
  • downgraded-backup-stops: a Pro vault with cloud backup on is locked, the plan is set to Free, and after the next unlock an edit uploads nothing for 15 s (an edit on Pro reached the bucket in about 5.2 s) and cloud backup shows off.

resilience (offline use and reconnect, cached tier, team vaults, export and import; spec 4.2, 6.5-6.8; node scripts/verify/run.mjs resilience, about 2.5 minutes)

  • offline-open: on Free, with the device's backend link cut after sign-in, a vault is created, locked and opened again; it shows "Offline: device check paused" in Settings > Sync, keeps limit 1, and its edits reach the shared file while the server has no lease row. When the link returns, an active lease row for that vault appears with no user action (about 60 s, the offline retry) and the badge clears.
  • offline-takeover-reconnect: on Free, A's link is cut after its offline edits publish; B (its own drive folder, fed A's file) is offered a take-over because the server still sees A, takes it and has A's offline edits. A keeps working, unaware, until its link returns; its next heartbeat (about 30 s) shows "This vault is now open on ..." with "Your changes from this device were saved.", A soft-locks, reports its final save (row still displaced by B's take-over), and its shared file holds every edit, including one made after the take-over.
  • cached-pro-offline: a Pro user signs in once online on two devices, which then relaunch with the backend cut (cached auth mode). While a signed-out device holds the owner claim, the device with an 8-day-old cached tier and the one with a cache dated a day ahead get the one-device take-over dialog (limit 1, source default); the device with the normal cache opens with no prompt (limit -1, source tier-cache) and its edit reaches the online device, the only one with a server lease.
  • team-vault-regression: a Team user creates a team vault on A (identity key, team_vault_create), adds entries that reach vault_entries, and B (identity key recovered with the passphrase) opens it and gets them by team sync. No personal-sync lineage folder, sync lineage in the team vault file, running personal sync or personal_vault_sessions row appears for it; after "Lock Current Vault" both devices reopen the personal vault and its edits sync both ways.
  • import-export: on Pro, export_execute writes a folder, a credential, an SSH entry using it and a web entry; the preview reads it and a wrong passphrase cannot. Imported into another vault that has an open same-field conflict on both devices, the entries arrive with new ids, the folder, the remapped credential and their secrets, sync to the other device, and the conflict stays open with both values.

Prerequisites (the harness checks or fixes these itself)

NeedWhat the harness does
macOS or Linux, Node 20+Preflight fails fast with a clear message otherwise.
Docker runningPreflight runs docker info; if it fails, start Docker Desktop and rerun.
Local Appwrite at localhost:8085 (project conduit-dev)Made once with npm run appwrite:setup (docs/LOCAL_APPWRITE.md); the harness reads conduit-dev.env in the stack folder and refuses any endpoint that is not localhost. If the stack is down it runs docker compose up -d in the stack folder and waits up to 3 minutes.
Current server codeWhen anything under appwrite/ changed since the last deploy (stamp file .deployed-conduit-dev in the stack folder), it redeploys the schema, the seed rows (tiers, app_config) and the api function. There is no SQL, no dump and no schema check.
Electron, playwright-core, MCP SDKPreflight finds them or says what to install.
Offline scenarios (resilience)A per-device TCP proxy to localhost:8085 (lib/net-proxy.mjs, ctx.backendProxy()) passed to that device as CONDUIT_DEV_APPWRITE_ENDPOINT, which only unpackaged preview builds honor, and only for a loopback URL; cutting it takes the backend away from that one device while every other device and run keeps it.
PortsIts own Vite on a free port, never 1420, so it runs beside the user's dev app.
IsolationEach test device gets /tmp/cv-<id>/<name>/ with its own HOME, app data, Chromium profile, single-instance lock and MCP socket, runs with --use-mock-keychain, and cannot claim conduit:// links or the global shortcut. Test users are verify-<runId>-<n>@conduit.local.

Run it

From the repo root:

npm run verify                                   # all suites
node scripts/verify/run.mjs sync mcp             # several suites
node scripts/verify/run.mjs sync --only take-back          # one scenario (repeat --only for more)
node scripts/verify/run.mjs mcp --only audit --keep        # keep /tmp/cv-<id> device roots to inspect
node scripts/verify/run.mjs --help

Exit code 0 means every phase, scenario and cleanup step passed; 1 means something failed; 2 means bad arguments. Suites run in file-name order: backup, copies, lifecycle, mcp, password, resilience, smoke, sync (sync last, so its file-safety watch also sees what the other suites left in the cloud folder). Test windows appear on screen: leave them alone.

The run id looks like 20260926-163755-1596d7; the first output line prints it. Device roots live in /tmp/cv-<id>, where <id> is the last six characters of the run id (/tmp/cv-1596d7), and results.json cleanup.keptRoots holds the exact path after --keep.

How long

About 12 minutes for all 41 scenarios (two consecutive full runs: 735.5 s and 738.9 s). Setup is about 9 s; scenario time per suite in a full run: backup 120 s, copies 105 s, lifecycle 163 s, mcp 48 s, password 52 s, resilience 150 s, smoke 4 s, sync 62 s. Most of it is the app's own waits: the 30 s missing-file debounce, the idle lock's 30 s check, the 30 s heartbeat and 60 s offline retry, the 5 s backup debounce and fixed windows that prove nothing is published or uploaded.

Give the command a 20 minute timeout, and run it in the background if your tool caps a command at 10 minutes. A cold start of the Appwrite stack, or a redeploy after appwrite/ changed, can add a few minutes.

Read the results

The last lines print a table and RESULT: PASS|FAIL in <time> (.verify/<runId>/results.json). Everything for the run is in .verify/<runId>/ (gitignored):

  • results.json: status, phases[] (preflight, appwrite, build, vite), scenarios[] with status, durationMs, timestamped steps, error (stack) and failure[] (per device screenshot path and main and renderer log tails), and cleanup (failures, leftover processes).
  • shots/: numbered PNGs; <n>-<device>-FAIL-<suite>-<scenario>.png for every open device on a failure.
  • logs/: run.log, <device>.main.log, <device>.renderer.log, <device>.mcp.log, build.log, vite.log, cleanup.log. Keys and tokens are redacted.

List the failures of the newest run:

node -e "const d=require('fs').readdirSync('.verify').filter(n=>/^\d{8}-/.test(n)).sort().pop();const r=require('./.verify/'+d+'/results.json');console.log(d,r.status);for(const s of [...r.phases,...r.scenarios].filter(s=>s.status!=='pass'))console.log(s.name??s.suite+'/'+s.scenario,'|',String(s.error).split('\n')[0])"

The table has no per-suite totals. Print passed / total and time per suite of the newest run (time is the sum of scenario times, without device start-up and shut-down between scenarios):

node -e "const d=require('fs').readdirSync('.verify').filter(n=>/^\d{8}-/.test(n)).sort().pop();const r=require('./.verify/'+d+'/results.json');const t={};for(const s of r.scenarios){const x=t[s.suite]??={p:0,n:0,ms:0};x.n++;x.ms+=s.durationMs;if(s.status==='pass')x.p++}for(const [k,x] of Object.entries(t))console.log(k,x.p+'/'+x.n,(x.ms/1000).toFixed(1)+'s')"

When something fails

  1. Find the first failure in results.json and read its steps, error, screenshot and the device logs. Decide whether it is an app bug, a harness bug or the environment (see Troubleshooting).
  2. Re-run just that scenario: node scripts/verify/run.mjs <suite> --only <scenario> (add --keep to look inside the device roots afterwards).
  3. Fix the root cause, then add or update a unit test for it where the code allows.
  4. Never make a run pass by weakening an assertion, deleting a check, skipping a scenario or raising a timeout without a measured reason. If a check disagrees with the spec (docs/MULTI_DEVICE_SYNC.md), say so and cite the section instead of quietly changing it.
  5. If a scenario passes only sometimes, treat it as a real race and report it.
  6. Finish by re-running the fixed scenario with --only, then the one suite it belongs to. Do not re-run every suite unless the user asks for it.

Cleanup guarantees

On pass, fail or Ctrl+C (press once; a second press aborts cleanup) the run quits every test device, stops its Vite, deletes its test users (their leases go with them) and their cloud backups in the vaults bucket, removes /tmp/cv-<id>/ and .verify/<runId>/dist-electron/, then kills and reports any process that still mentions the run. A leftover process or failed cleanup step makes the run FAIL. Each run also deletes verify-* users older than 6 hours and, before it starts, prunes .verify/ to the 20 newest run folders (so 21 exist after a run). The local Appwrite stack is left running. The build phase also rewrites the gitignored mcp/dist/ in the repo; that is expected and not undone.

--keep keeps only the device roots and .verify/<runId>/dist-electron; test users and their lease rows are still deleted. After --keep, delete /tmp/cv-<id> and .verify/<runId>/dist-electron yourself when done.

Quick check afterwards (works in zsh and bash; the trailing slash matters because /tmp is a symlink on macOS): find /tmp/ -maxdepth 1 -name 'cv-*' prints nothing and ps ax | grep -F /tmp/cv- | grep -v grep finds nothing (unless another run is going). Do not use ls -d /tmp/cv-*: zsh prints no matches found when the folder is clean.

Troubleshooting

SymptomCauseFix
Docker is not running at preflightDocker Desktop is offStart Docker Desktop, wait until docker info answers, rerun.
appwrite phase fails, or the API does not answerStack never set up, half-started, or short of memoryIf the message names a missing conduit-dev.env, run npm run appwrite:setup. Check curl -s http://localhost:8085/v1/health/version and docker ps --filter name=appwrite. Run npm run appwrite:stop and rerun; the harness starts it again. Give Docker Desktop at least 4 GB of memory.
appwrite deploy fails in the appwrite phaseBuild or schema error in appwrite/Run npm run test:server, then npm run appwrite:deploy to see the error; fix the code, not the check. The function's deployment and execution logs are in the console (http://localhost:8085).
No user_profiles row ... is the api function deployed with its users.*.create event?The function is not deployed or failednpm run appwrite:deploy, then check the function's executions in the console.
Port 8085 busyAnother Appwrite stack is runninglsof -nP -iTCP:8085 -sTCP:LISTEN. Ask the user before stopping another stack. Vite ports are picked automatically.
resilience/import-export fails: imported entries not visiblePre-existing, not from the Appwrite move: it fails the same way on main with Supabase (the entries land in a collapsed folder)Report it as pre-existing.
A device times out on a page call, or a macOS keychain prompt is on screenSome code path reached the real keychainNever answer the prompt. Let the run fail (cleanup kills the device) or SIGKILL only that device's pid. Find the call in <device>.main.log and fix the isolation (scripts/verify/lib/launcher.mjs).
EINVAL on the MCP socket, or too long for the MCP socket pathUnix socket path over 104 bytesKeep device names 1 to 12 characters and device roots at /tmp/cv-<id>/<name>; never move them under $TMPDIR or the repo.
Device name "<name>" was already used by <suite>/<scenario>Two scenarios, often in different suites, launch the same name; a device root lasts the whole run, so the second would inherit the first one's sign-in and device idRename the newer scenario's devices with its suite's prefix (backup b, copies c, lifecycle l, mcp m, password p, resilience r, sync s).
createVault: <path> already existsTwo scenarios use the same <cloud>/<folder>/ (the hub would open the old file and wait for "Set a master password" in vain)Name the folder with the suite's prefix, like its devices.
playwright-core is not installed, MCP SDK not resolvable, Electron binary missingDependencies not installednpm install and (cd mcp && npm install).
build phase failsTypeScript errorRead logs/build.log; run npx tsc --noEmit -p electron/tsconfig.json.
[cleanup] FAIL or leftover process killedA device or client did not shut downRead logs/cleanup.log and fix the quit path; the run counts as failed.

Safety rules

  • Only the local stack on localhost. Never point the harness or any env var at production Appwrite (appwrite.cloudwise.ca) or another cloud project, and never set CONDUIT_ENV=production.
  • Never touch the user's dev profile (~/Library/Application Support/conduit/conduit-dev, the "Conduit Dev" app), their running dev app or its Vite on port 1420, or ~/.claude.json.
  • Never answer a macOS keychain prompt.
  • Only kill processes that belong to the run (their command line mentions /tmp/cv-<id> or .verify/<runId>).
  • Never run docker compose down -v on the local stack or delete its env file unless the user asks: that wipes the stack.
  • Never print the API key, the service secret or session tokens.

Report back

One line with the overall result and time, then one line per suite (passed / total and time, from the per-suite command in "Read the results"), then each failure with its root cause, the fix or the open question, and the run folder (.verify/<runId>). Quote the RESULT: line as printed. Never say a run passed unless it ran and printed RESULT: PASS.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.