fix(platforms): clear home channel when setup prompt left blank (#12423)

[DavidMetcalfe]

Problem

Interactive platform setup wizards (Feishu, Discord, Slack, Mattermost, WhatsApp) could not clear an existing *_HOME_CHANNEL env var. If a user re-ran setup and left the home-channel prompt blank, the old value was silently preserved. Users had to manually edit ~/.hermes/.env to clear it.

Reported in #12423 for Feishu; the same pattern existed in 4 other adapters.

Changes

• All 5 adapters: Added remove_env_value to the lazy-import block and an else branch that calls it when the prompt is left blank, clearing the persisted env var.
• Whitespace handling: All adapters now .strip() the prompt result before the emptiness check, so whitespace-only input clears instead of being saved.
• Feishu UX hint: Added a print_info line before the prompt explaining that leaving it blank clears a previously saved home channel.
• Code smell fix: Used else: if remove_env_value(...) instead of elif remove_env_value(...) to avoid side effects in a conditional expression (caught by review).

Tests

• Updated _run_setup_feishu helper to mock remove_env_value and track removed keys.
• Added TestSetupFeishuHomeChannel with 4 tests:
  • Blank input removes existing home channel
  • Blank input without prior key still invokes remove
  • Non-empty input saves and does not remove
  • Whitespace-only input clears (not saves)
• All 18 tests pass.

Verification

python -m pytest tests/gateway/test_setup_feishu.py -q

Notes

• Supersedes stale PR fix(cli): clear Feishu home channel when setup prompt left blank (#12423) #12451 (targeted pre-refactor file path hermes_cli/gateway.py; function moved to plugins/platforms/feishu/adapter.py).
• The other 4 adapters' prompts already say "leave empty to set later" — the else branch now makes that accurate (empty = clear). Prompt wording could be updated in a follow-up for full consistency.

[teknium1]

Thanks for tracing the Feishu report through the plugin migration and applying the existing remove_env_value() helper. The five changed paths still have the reported truthy-only persistence behavior on current main (for example, plugins/platforms/feishu/adapter.py:5616-5619), and the proposed removal mechanism is appropriate.

Problems

• The same bug class remains in the adjacent plugin setup flows: Matrix only saves MATRIX_HOME_ROOM when home_room is truthy (plugins/platforms/matrix/adapter.py:4532-4534), and WeCom only saves WECOM_HOME_CHANNEL when home is truthy (plugins/platforms/wecom/adapter.py:1819-1821). Both are active cron/home delivery targets (cron/scheduler.py:215,225).
• The new regression coverage is Feishu-only. Existing Slack coverage confirms blank input is not saved (tests/gateway/test_slack_plugin_setup.py:45-57), but does not exercise removal of an existing target.

Suggested changes

• Apply the same clear-on-blank behavior to Matrix and WeCom.
• Add seeded-existing-value, blank, whitespace, and non-empty tests for each affected setup wizard.

Automated hermes-sweeper review.

[DavidMetcalfe]

Thanks for the trace and the salvageability tag — both findings addressed in two follow-up commits on top of 394f6ca07:

bf5115ff8 — fix(platforms): clear home channel on blank prompt for matrix and wecom

Matrix (plugins/platforms/matrix/adapter.py) and WeCom (plugins/platforms/wecom/adapter.py) now mirror the same remove_env_value() + .strip() + else-clear pattern as the other 5 adapters. Both are in cron/scheduler.py::_HOME_TARGET_ENV_VARS (line 254 and 264), so the cron-silent-preserve failure mode is closed for every active cron home delivery target.

201a39e78 — test(gateway): add home-channel clear-on-blank coverage for 6 adapters

5 new test files (Discord/Mattermost/WhatsApp/Matrix/WeCom) + a TestSlackHomeChannelClear class augmenting the existing Slack test. Each adapter gets 4 scenarios:

• test_blank_removes_existing_home_channel — pre-seeded env, blank prompt → remove_env_value called AND key removed from saved env.
• test_blank_without_prior_home_still_attempts_remove — no pre-seeded env, blank prompt → remove_env_value still called once (confirms the else-branch isn't dead code).
• test_nonempty_saves_home_channel — fresh env, real value → save_env_value called AND remove_env_value NOT called (no accidental clear on save).
• test_whitespace_only_clears_home_channel — pre-seeded env, " " input → cleared (regression guard on .strip()).

Local verification:

$ pytest tests/gateway/test_discord_plugin_setup.py \
        tests/gateway/test_mattermost_plugin_setup.py \
        tests/gateway/test_whatsapp_plugin_setup.py \
        tests/gateway/test_matrix_plugin_setup.py \
        tests/gateway/test_wecom_plugin_setup.py \
        tests/gateway/test_slack_plugin_setup.py \
        tests/gateway/test_setup_feishu.py
============================== 44 passed in 0.56s ==============================

44 tests across 7 adapters, all green. The Matrix test additionally patches tools.lazy_deps.feature_missing/ensure so the wizard's mautrix auto-install path is a no-op; the WeCom test stubs qr_scan_for_bot_info and prompt_choice so the QR path is skipped. CI should be green on the next push.

Cross-vendor dual review (Flash + GPT-OSS) on the rework passed with no BLOCKERs; both reviewers independently accepted the adapter patches, mock semantics, prompt-choice sequence, scope, and commit boundary. Two SHOULD-FIX findings (None-safety on prompt() and Unicode whitespace) were verified against hermes_cli/cli_output.py:43-48 (prompt() returns str, never None) and str.strip() semantics (handles all Unicode whitespace per the Unicode database) — both false positives.

Happy to split the test commit per-adapter or follow up on the prompt-text-consistency NIT separately if you'd prefer.

[teknium1]

Merged via #70196 — your commit was cherry-picked/reapplied onto current main with your authorship preserved in git history: your blank-home-channel fix was cherry-picked, fixes #12423.

Thanks for the contribution!