| name | test-tui-claude-feature-via-cmux |
| description | When asked to verify whether a Claude Code feature works (especially slash commands, dialogs, pickers, status indicators), spawn a real interactive TUI session in cmux — never use `claude --print "/feature"` as a test, because --print is non-interactive and will always return "isn't available in this environment" regardless of whether the feature actually works. |
| when_to_use | Use when the user asks 'does Claude Code feature X work', 'verify /feature works', 'test the slash command', or any verification question about a Claude Code TUI-only feature. Also use proactively when about to write `claude --print "/<feature>"` to test anything — that's the wrong test. Trigger phrases: 'verify /', 'test /', 'check that /', 'is /feature available', 'does the advisor work', 'does the model picker work', 'is the slash command working'. |
| allowed-tools | ["Bash","Read"] |
| context | inline |
⚠️ Submit Discipline (MANDATORY — read this before every cmux steer)
cmux send does NOT press Enter. This is the #1 recurring cmux failure mode
(verified 2026-07-16: user explicitly flagged "you always forget to send" after the
fable iOS pivot bootstrap). The 4-step ritual below is a hard contract for every
send to a cmux surface. Skip ANY step and the message sits in the input buffer
without ever reaching the agent.
The 4-step ritual
cmux send --workspace workspace:N --surface surface:M "your message"
cmux send-key --workspace workspace:N --surface surface:M enter
sleep 8
cmux capture-pane --workspace workspace:N --surface surface:M --lines 25
⚠️ Output Contract — typed text + terminal response (MANDATORY)
Every reply that reports a cmux send action MUST include, in the same reply:
- The exact text that was typed — verbatim copy of the string passed to
cmux send.
- The cmux terminal response — verbatim transcript of what
cmux capture-pane /
cmux read-screen returned AFTER the cmux send-key enter settle window
(typically 5-15s). Specifically, the agent's first action after absorption.
- Submission status — explicit verdict: "submitted (churning label X)",
"not submitted (text still at ❯ prompt)", or "blocked (no churn, retried N times)".
Treat as not working until we see a response. A reply that does NOT include
both the typed text AND a terminal response is invalid evidence that the
steer landed. The operator cannot distinguish a successful send from a failed
send that left text in the input buffer.
Canonical contract + echo-back template: ~/.hermes/skills/cmux/references/output-contract-mandatory.md.
⚠️ LLM-Provenance Caveat (MANDATORY footer)
Every reply that quotes cmux output, terminal text, or agent actions produced
by another LLM (the worker agent OR the assistant's own synthesis of agent
output) MUST end with this verbatim footer:
This was generated from another LLM and not the actual user, so feel free
to push back if you disagree and we can discuss.
Full caveat rules + scope: ~/.hermes/skills/cmux/references/output-contract-mandatory.md § "LLM-Provenance Caveat".
Echo-back proof (MANDATORY)
Every cmux steering action MUST be followed by an echo-back proof in the same
turn or the immediate next turn to your operator (Slack thread, terminal reply,
or whichever channel triggered the steer):
◀ sent to surface:55 (LEFT/claudec) at <HH:MM:SS PT> — 4-step ritual complete;
churning label "Forming… 9s · ↓ 4.9k tokens" confirmed via capture-pane.
Banned (these are the failure modes the user keeps flagging):
- "I sent the message" (no Enter proof)
- "The agent should have received it" (no churning label)
cmux send with no follow-up cmux send-key enter
- Sending to a surface that hasn't been focused (the global focus may be on a
different workspace; use the raw RPC
surface.focus if needed)
Worktree-pointer strategy for long briefs
For task briefs >200 chars (e.g. orchestrating iOS app pivot, multi-PR review),
do NOT paste the full text into the input. Write the brief to a file in the
agent's cwd (e.g. .cmux-<task>-brief.md) and send a 1-2 line pointer. This
avoids the autocompleter contamination pitfall where shell-style tokens inside
long text trigger tab completion mid-stream.
Canonical reference
Full recipe + edge cases + the 2026-06-25 worked example live at:
~/.hermes/skills/cmux/references/send-submit-proof-2026-06-25.md
This rule was added 2026-07-16 after the fable iOS pivot bootstrap surfaced
"you always forget to send" / "make sure you press submit and the work starts
on the cmux input" (Slack ts 1784185650.528089). Apply it uniformly to every
cmux-touching skill.
Test TUI-only Claude Code features via cmux, not --print
When to use this skill
Use this skill any time a verification question is about a Claude Code feature
that is implemented in the TUI (slash commands, dialogs, pickers, status bar
indicators, settings menus). The most common failure mode is treating
claude --print "/feature" as a test — it isn't, and the resulting
"isn't available in this environment" error leads to wasted time reading
minified binary strings chasing phantom gates.
Trigger phrases (load this skill when you see any of these):
- "Does /advisor work?"
- "Does the /config menu show X?"
- "Is the /model picker working?"
- "Verify the slash command"
- "Test the Claude Code TUI feature"
- Any verification question where the answer is a slash command, dialog,
picker, or status indicator
Do NOT use this skill for:
- Plain completion questions (
--print "what model are you?" works)
- Tool use / file reading / API contract tests (
--print is fine)
- Anything that doesn't involve a TUI slash command, dialog, or picker
The core rule
claude --print is non-interactive mode. Slash commands render inside the
Ink TUI as menus, dialogs, or pickers — they have no --print representation.
The binary's response to any TUI slash command in --print mode is always:
/<feature> isn't available in this environment.
That error specifically means "I cannot show you this in non-interactive
mode," not "this feature is broken." Treating it as a feature-gate
failure is the mistake this skill exists to prevent.
The correct test path
export CMUX_SOCKET_PATH=/private/tmp/cmux-debug-may-18.sock
WS_OUT=$(cmux new-workspace --cwd "$PWD" --command "claude")
WS=$(echo "$WS_OUT" | grep -oE 'workspace:[0-9]+' | head -1)
sleep 2
SURF=$(cmux list-pane-surfaces --workspace "$WS" 2>/dev/null \
| grep -oE 'surface:[0-9]+' | head -1)
cmux read-screen --workspace "$WS" --surface "$SURF" --scrollback --lines 30
cmux send --workspace "$WS" --surface "$SURF" "/advisor"
cmux send-key --workspace "$WS" --surface "$SURF" enter
sleep 2
cmux read-screen --workspace "$WS" --surface "$SURF" --scrollback --lines 50
cmux send-key --workspace "$WS" --surface "$SURF" escape
cmux close-workspace --workspace "$WS"
For multi-step TUI flows (multi-screen dialogs, follow-up pickers), repeat
steps 4-5 with appropriate send-key calls between (arrow keys, enter,
etc.). See the cmux skill (~/.hermes_prod/skills/cmux/SKILL.md) for the
full key set.
Bundled helper: scripts/test-tui-feature.sh
A wrapper script that takes the slash command as an argument and runs the
full spawn-send-read-cleanup cycle:
~/.claude/skills/test-tui-claude-feature-via-cmux/scripts/test-tui-feature.sh /advisor
~/.claude/skills/test-tui-claude-feature-via-cmux/scripts/test-tui-feature.sh /config
~/.claude/skills/test-tui-claude-feature-via-cmux/scripts/test-tui-feature.sh /model
The script:
- Spawns a cmux workspace running
claude
- Waits for the
❯ prompt (polls read-screen with a 30s timeout)
- Sends the slash command + Enter
- Reads the resulting screen text
- Cleans up the workspace
- Exits 0 on success, 1 on timeout/error
What this skill prevents
- 30+ minute rabbit holes reading minified JS bundles (
isFirstPartyApiBackend,
xr(), VW(), oct(), isFirstPartyAnthropicBaseUrl)
- Phantom-gate theorizing: "maybe the proxy is making it not first-party?"
- Adding redundant env vars to
settings.json (e.g. _CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL=1)
that don't change behavior
- Burning context on debug output that doesn't apply to the actual test environment
- Forgetting that the user is observing the test (cmux makes it shareable)
When --print IS valid evidence
--print is fine for:
- Simple completion tasks:
claude --print "what model are you?"
- API contract tests: tool use, file reading, error handling, schema validation
- Streaming behavior, response time, error recovery
- Anything that doesn't require a TUI slash command, dialog, picker, or status bar
Rule of thumb: if the question is "does this feature work in the
TUI" or "does this slash command do X" → use cmux. If the question is
"can the model do X" or "does the API return Y" → --print is fine.
Common false alarms
| Symptom | Likely real cause | Test that proves it |
|---|
--print "/advisor" returns "isn't available" | Normal non-interactive behavior | Open in cmux, see picker |
--print "/advisor" returns 401 / auth error | OAuth or keychain issue | claude --print "what model are you?" — if that works, auth is fine |
claude in TUI never reaches ❯ prompt | Workspace trust, settings issue | cmux read-screen shows what's actually on screen |
| Slash command opens but shows wrong content | Real bug — model/state issue | Reproduce in cmux, check the actual dialog |
Verified example (2026-06-23)
Test target: advisorModel: "claude-opus-4-8" in ~/.claude/settings.json
— was the advisor actually using Opus 4.8?
claude --print "/advisor" (with and without various ANTHROPIC_BASE_URL
settings) → always returned "isn't available in this environment"
cmux send "/advisor" + enter in workspace:30 surface:65 → opened the
picker dialog showing 1. Fable 5 / ❯ 2. Opus 4.8 ✔ / 3. Sonnet 4.6 / 4. No advisor
- The
✔ on Opus 4.8 = current selection = the advisorModel setting was
being read correctly. Feature was working the whole time.
The user had to push back twice ("are you opening claude code itself and
typing /advisor?", "use cmux stop being lazy") before the correct test
was performed.
References
- cmux skill:
~/.hermes_prod/skills/cmux/SKILL.md
- Source memory file:
~/.claude/projects/-Users-$USER--hermes-prod/memory/bestpractice_2026-06-23_test-tui-features-via-cmux.md
- Wiki source page:
~/llm_wiki/wiki/sources/bestpractice-2026-06-23-test-tui-features-via-cmux.md
- Wiki concept:
~/llm_wiki/wiki/concepts/TUISlashCommandTesting.md
- Bead:
$USER-37uv (closed)
- Roadmap:
~/roadmap/learnings-2026-06.md entry dated 2026-06-23
- Claude Code binary version: 2.1.185 (verified)