test(cli): add command-registry dispatch coverage gate

[blightbow]

Summary

Adds test_every_cli_command_has_dispatch_handler, an enumeration gate that prevents the recurring bug class where commands are added to COMMAND_REGISTRY but never wired into HermesCLI.process_command().

The Bug Class

COMMAND_REGISTRY (source of truth for what commands exist) and the elif chain in process_command() (source of truth for what commands work) are maintained independently with no automated reconciliation. When they drift, the command breaks silently — appearing in /help, autocomplete, and all derived surfaces (Telegram/Slack) as functional, but returning "Unknown command" at runtime.

This has occurred at least five times:

Command	First reported	Status
/model	#3821 (~3mo)	Fixed
/resume	#2591 (~2mo)	Fixed
/sessions	#22960 (~2mo)	Fixed May 2026
/indicator	#22960 (~2mo)	Open (5 competing PRs)
/whoami	#35892 (~1mo)	Open

The documented cost of not having this gate is not only recurrences of this bug class, but ballooning PRs from users converging on the same bug from everyday use. #51009 by @GitWhoIsThis (24d ago) is a superset issue that enumerated the full set of missing handlers. It remains open.

This PR

A single test in TestDerivedDicts that uses inspect.getsource to extract every dispatched canonical name from process_command() and asserts coverage over every gateway_only=False entry in COMMAND_REGISTRY. The test currently fails honestly, and that's the point.

Currently FAILS on main: ['indicator', 'whoami']. This PR is blocked until those handlers are merged (see #41869 / #40047 / #51178 and #35996). Once they land, the gate self-resolves and any future handler-less CommandDef fails CI immediately.

Prior Art in This Codebase

The enumeration-gate pattern is well-established here:

• test_commands_dict_includes_all_cli_commands — asserts COMMAND_REGISTRY ⊆ COMMANDS
• test_commands_by_category_covers_all_categories — asserts category completeness
• tests/providers/test_provider_profiles.py — asserts provider registry invariants
• tests/hermes_cli/test_auth_commands.py — asserts auth-step registry invariants

All follow for <item> in <REGISTRY>: assert <invariant>. This PR extends that pattern to the dispatch gap.

Test Plan

$ scripts/run_tests.sh tests/hermes_cli/test_commands.py -q
# (...snip...)
E       AssertionError: commands in COMMAND_REGISTRY with no dispatch handler: ['indicator', 'whoami']. Add an `elif canonical == "<name>":` branch to HermesCLI.process_command() in cli.py, or mark the command gateway_only=True if it is not meant for CLI.
E       assert not {'indicator', 'whoami'}

tests/hermes_cli/test_commands.py:186: AssertionError
=========================== short test summary info ============================
FAILED tests/hermes_cli/test_commands.py::TestDerivedDicts::test_every_cli_command_has_dispatch_handler
1 failed, 173 passed in 3.05s

=== 1 file with test failures (1 test failed) ===
  tests/hermes_cli/test_commands.py  (1 test failed)

[alt-glitch]

This was generated by AI during triage.

Related: #51009 tracks the currently missing slash-command dispatches. This gate intentionally fails until those handlers land; maintainers should choose the landing sequence.

[blightbow]

Ignore the force push, hermes attempted to "helpfully" commit and push a local fix that was unrelated to this PR. The force push was needed to restore the branch to the clean state.

[tonydwb]

Code Review Summary

Verdict: Approved

Good defensive test that catches a known bug class (missing dispatch handler) before it reaches production. Uses inspect.getsource to verify registry-to-dispatch parity. References prior instances of the bug class in the docstring. No security concerns.

---

Reviewed by Hermes Agent

[teknium1]

Thanks for identifying a real registry/dispatch drift: current main registers /indicator and /whoami in hermes_cli/commands.py:121,163-165, while classic HermesCLI.process_command() falls through to its unknown-command path at cli.py:8958-9131.

Problems

• tests/hermes_cli/test_commands.py:178 uses inspect.getsource() and regexes process_command. AGENTS.md:1380-1384 explicitly bans source-reading tests because they assert implementation shape rather than behavior.
• The proposed assertion deliberately fails against current main until the existing dispatch gaps are resolved, so it cannot land independently as a green CI change.

Suggested changes

• Replace source inspection with an executable dispatch contract; extract a narrow resolver/dispatch seam if needed and test actual handler behavior.
• Carry the gate with, or after, the /indicator and /whoami dispatch or correct surface-classification fixes.

Automated hermes-sweeper review.

[teknium1]

inspect.getsource() makes this a source-shape test, which AGENTS.md:1380-1384 explicitly bans. Please test an executable dispatch contract instead; extract a small resolver/dispatch seam if the current method is too large to exercise directly.

[blightbow]

@teknium1 The source-shape defect has been resolved with proper exercising of the code.

Regarding the green CI merge, this is a draft PR and this can't actually merge until I advance the draft. The current plan is to keep the draft in a WiP state until the missing dispatches land. Let me know if you'd rather I handle this in a different way. I decided against xfail or whitelisting approaches because I don't want to risk agents adopting that as a pattern when interacting with this test.

On the topic of draft PRs, there may be a gap in the github-code-review skill. It states "use Comment when the PR is a draft", but isDraft is never actually fetched. The skill only references title/author/refs/state/body. That might explain some of the activity in this thread.

[blightbow]

RE: gate verification, I had an agent do a quick local fix of the missing dispatches for /indicator and /whoami. I chose not to submit PRs for these as there is a demonstrated arms race of PRs competing for these fixes. The higher value contribution was closing the testing gap. (this PR)

The text that follows is provided by Claude Code.

Verification patch (local only — not proposed for merge)

Verification was three-way:

Tree state	Result
Both handlers wired (patch below)	174 passed, 0 failed — gate green
Handlers absent (this PR / main)	1 failed — ['indicator', 'whoami']
/indicator branch commented out, its text still in the file	1 failed — ['indicator']

The first two rows are the part that matters for "does this gate actually work": it goes green purely by wiring the handlers, with no changes to the test.

The third row is why the rewrite was worth doing rather than just relocating the assertion. I commented out the elif canonical == "indicator": branch while leaving that literal text in the file as a comment, so the command is genuinely broken at runtime. The new test fails. The previous inspect.getsource version, run against that same mutated source, reports:

OLD getsource gate missing: [] -> would PASS

So the original gate would have gone green on a build where /indicator returns "Unknown command" to users — which is AGENTS.md:1379-1381's stated failure mode ("passes when the implementation is subtly broken... the regex matches a call site that exists but is wired wrong") reproduced concretely. Good catch on that review point; it wasn't just a style violation.

One implementation note on the rewrite, since it deviates from the HermesCLI.__new__ helper used in tests/cli/test_cli_prefix_matching.py: the gate binds a MagicMock(spec=HermesCLI) as self and calls process_command unbound, so alias resolution and the real dispatch chain execute while handler bodies collapse to no-ops. A real instance runs the handler bodies, which is correct for single-command behavior tests but not for a ~74-command sweep — /undo and /update block on their Choice [1/2/3]: stdin prompts and /voice reaches the network. That rationale is in the test's docstring so it doesn't get "fixed" back later.

verification patch

Caveat: this is a fixture, not a contribution. It prints via bare print() rather than the _cprint idiom used elsewhere in the mixin, and it hasn't been reviewed for the edge cases the dedicated PRs handle. It is sufficient to prove dispatch reachability and nothing more.

diff --git a/cli.py b/cli.py
--- a/cli.py
+++ b/cli.py
@@ -8895,10 +8895,14 @@ class HermesCLI(CLIAgentSetupMixin, CLICommandsMixin):
             self._handle_subgoal_command(cmd_original)
         elif canonical == "skin":
             self._handle_skin_command(cmd_original)
+        elif canonical == "indicator":
+            self._handle_indicator_command(cmd_original)
         elif canonical == "voice":
             self._handle_voice_command(cmd_original)
         elif canonical == "busy":
             self._handle_busy_command(cmd_original)
+        elif canonical == "whoami":
+            self._handle_whoami_command(cmd_original)
         else:
             # Check for user-defined quick commands (bypass agent loop, no LLM call)
             base_cmd = cmd_lower.split()[0]
diff --git a/hermes_cli/cli_commands_mixin.py b/hermes_cli/cli_commands_mixin.py
--- a/hermes_cli/cli_commands_mixin.py
+++ b/hermes_cli/cli_commands_mixin.py
@@ -2734,3 +2734,59 @@ class CLICommandsMixin:
         else:
             _cprint(f"Unknown voice subcommand: {subcommand}")
             _cprint("Usage: /voice [on|off|tts|status]")
+
+    def _handle_whoami_command(self, command: str):
+        """Handle /whoami — show the active profile and session identity."""
+
+        profile = os.environ.get("HERMES_PROFILE", "default")
+        session = getattr(self, "session_id", "no session")
+        model = getattr(self, "model", "?")
+        provider = getattr(self, "provider", "?")
+        cwd = os.getcwd()
+        toolsets = getattr(self, "enabled_toolsets", []) or []
+
+        print()
+        print(f"  Profile:    {profile}")
+        print(f"  Session:    {session}")
+        print(f"  Model:      {model} ({provider})")
+        print(f"  Directory:  {cwd}")
+        print(f"  Toolsets:   {', '.join(sorted(toolsets)) if toolsets else 'default'}")
+        print()
+
+    def _handle_indicator_command(self, command: str):
+        """Handle /indicator [kaomoji|emoji|unicode|ascii] — set the TUI busy-indicator style."""
+        from cli import save_config_value
+
+        VALID = ("kaomoji", "emoji", "unicode", "ascii")
+
+        parts = command.strip().split(maxsplit=1)
+        style = parts[1].strip().lower() if len(parts) > 1 else ""
+
+        if not style:
+            import cli as _cli
+
+            current = _cli.load_cli_config().get("display", {}).get(
+                "tui_status_indicator", "kaomoji"
+            )
+            print()
+            print(f"  Current indicator: {current}")
+            print("  Available styles:")
+            for s in VALID:
+                marker = " ●" if s == current else "  "
+                print(f"   {marker} {s}")
+            print()
+            print("  Usage: /indicator <style>")
+            print()
+            return
+
+        if style not in VALID:
+            print(f"  Unknown indicator: {style}")
+            print(f"  Pick one of: {'|'.join(VALID)}")
+            return
+
+        if save_config_value("display.tui_status_indicator", style):
+            print(f"  Indicator set to: {style} (saved)")
+        else:
+            print(f"  Indicator set to: {style}")
+        if self._apply_tui_skin_style():
+            print("  TUI indicator updated.")

[blightbow]

Just an activity ping. Still blocked on 22960 and 35892. If new breaking cases emerge in the interim, I will implement a temporary blacklist of the offenders so that we can get this shipped. (it will prove the coverage gap is leaking too regularly)