Run E2E / Coin Tester on CI

When: for changes that impact the e2e apps (desktop/mobile/coin logic), once the PR is getting ready for review. Desktop & Mobile E2E only run on a schedule + manual dispatch, so trigger them deliberately. Coin Tester already auto-runs on PRs for affected coins — only dispatch it to force a chain or re-run.

Scope it down first — filtered is the default, full suite is the exception

Always run the smallest subset that covers the change. The full suite is slow, expensive, and competes for shared on-chain accounts, so run it only for genuinely cross-cutting changes — and when you do, say why in the PR comment. Derive the scope from the diff:

Filter tags: @bitcoin, @family-evm, @solana, @generic-coin-framework, @smoke, … See the test-filter guide for the grammar and the mobile (whole spec files by path/@-tag) vs desktop (test titles) matching difference.

team vs test_filter. test_filter is an OR over its patterns; the team dropdown is an AND that narrows the run to one team's specs. Combine them: team=bst + test_filter=@bitcoin runs only the bitcoin specs BST owns. Slugs: bst, buy-and-sell, coin-integration, earn, engagement, swap, wallet-xp (all is the default and changes nothing). List them with their spec counts:

node e2e/tooling/filter/resolve.mjs --list-teams --check-dir e2e/desktop/tests --runner playwright
node e2e/tooling/filter/resolve.mjs --list-teams --check-dir e2e/mobile/specs  --runner detox

Selection is per spec file, so a spec two teams share runs for both. bst and coin-integration each own about half the suite — pair those with a coin tag or you have not narrowed much. Do not put @team-<slug> in test_filter: it is rejected, with a message pointing at the dropdown.

Dispatch

These are workflow_dispatch workflows — no slash-command triggers them. Dispatch with gh, which must run outside the sandbox (dangerouslyDisableSandbox: true).

gh workflow run --ref <branch> is all you need: the workflows checkout inputs.ref || github.sha, so the optional "Specify branch" (ref) field can stay blank — it falls back to the commit --ref points to. Run-names echo the branch, so use gh run list to confirm.

BR=<branch>            # PR head branch
TAGS=<derived-filter>  # e.g. "@solana" — see the table above. Drop -f test_filter ONLY for a justified full suite.
CHAIN=<derived-chain>  # e.g. "solana" or "evm,solana" — derive from the diff; omit -f chain to auto-detect.

# Filtered runs (preferred). Narrow tests_type when the change is platform-specific.
gh workflow run test-ui-e2e-only-desktop.yml --ref "$BR" -f test_filter="$TAGS"
gh workflow run test-mobile-e2e-reusable.yml --ref "$BR" -f test_filter="$TAGS" -f tests_type="iOS & Android" -f speculos_device=nanoX

# Scope to one team (optional, ANDs with the filter above). Omit -f team for the default "all".
gh workflow run test-ui-e2e-only-desktop.yml --ref "$BR" -f team=swap -f test_filter="$TAGS"
gh workflow run test-mobile-e2e-reusable.yml --ref "$BR" -f team=swap -f test_filter="$TAGS" -f tests_type="iOS & Android" -f speculos_device=nanoX

gh workflow run test-coin-tester.yml --ref "$BR" -f chain="$CHAIN"   # or omit -f chain to auto-detect affected coins

gh run list --branch "$BR" --limit 6   # grab the run IDs

Mobile requires tests_type (Android Only|iOS Only|iOS & Android) and speculos_device (nanoS|nanoSP|nanoX|stax|flex|nanoGen5). Desktop defaults to Speculos nanoSP. Coin Tester auto-detects affected coins (or -f chain="evm,solana").

team is optional on both and defaults to all. Some combinations are refused outright, failing the run in the first minute rather than running the wrong thing:

• invert_filter with team, or with smoke_tests. --grep-invert is applied to the whole pattern, so anything the resolver folds into it is inverted too: with a team you get every other team's specs, and with Smoke you get the whole suite minus the smoke tests. Invert only ever applies to test_filter, so use it alone.
• @team-<slug> typed into test_filter. Use the dropdown; the error says so.

-f team= only exists on branches that already carry the input — a branch cut before it landed will reject the flag.

Verify the filter took effect: open the run's Summary → Workflow Context. Desktop prints Filtered pattern; Mobile prints both Filtered pattern (your raw input) and Resolved filtered pattern (what the runner selects with). Both print a Team line.

A filter that matches 0 specs is wasted, and the failure mode depends on whether a team is set:

• No team — resolve.mjs warns "filter has no matches"; on Mobile the test jobs are then skipped and the run still reports success, on Desktop it fails ("No tests executed").
• With a team — unknown teams and teams that own nothing fail in the first minute (look for Unknown E2E team / E2E selection is empty). On Mobile, an empty team ∩ filter is also a hard error (E2E selection is empty). On Desktop, the resolver can only warn when the team ∩ filter looks empty; the run may still proceed and fail later with "No tests executed".

Fix the filter and re-dispatch.

Post the run on the PR

Fill the Scope column with the actual filter used (never a hardcoded "full suite"), and add a rationale line stating what you scoped to and why the other surfaces were skipped. Post with gh pr comment <pr> --body ... (--edit-last to update):

## 🧪 Triggered test workflows

Manually dispatched on `<branch>` (commit <short-sha>, `git rev-parse --short HEAD`), scoped to <scope>:

| Workflow | Scope | Run |
|---|---|---|
| [Desktop] E2E Only | `<@tag>` filter + `team=<slug>`, Speculos nanoSP | [run <id>](https://github.com/LedgerHQ/ledger-live/actions/runs/<id>) |
| [Mobile] E2E Only | `<@tag>` filter + `team=<slug>`, <tests_type>, nanoX | [run <id>](https://github.com/LedgerHQ/ledger-live/actions/runs/<id>) |
| [Coin] Test Coin modules | `chain=<...>` (or affected coins) | [run <id>](https://github.com/LedgerHQ/ledger-live/actions/runs/<id>) |

Scope rationale: <which surfaces changed, what you filtered to, and why the rest were skipped>.

Only write "full suite" when you deliberately ran everything, and keep the rationale line explaining why the change is cross-cutting enough to need it.