kon-tmux-test

name	kon-tmux-test
description	E2E testing of kon using tmux sessions; IMPORTANT: only trigger this skill when user asks for e2e testing of kon

Kon Tmux E2E Testing

End-to-end testing of kon using tmux sessions to programmatically control the TUI application.

Why Tmux?

Kon is a TUI (Textual-based) app. Running tests programmatically is hard. Tmux provides:

• tmux new-session - isolate test environment
• tmux send-keys - send keyboard input
• tmux capture-pane - capture output
• tmux has-session - check if kon is running

Test Design Philosophy

• Deterministic: Shell scripts create reproducible test environments
• Isolated config: Tests run with HOME=/tmp/kon-e2e-home so runtime settings do not mutate the real user config; auth JSON files are copied into the temp HOME when present so provider startup still works
• Separation of concerns: Shell script runs steps and captures output; kon/the reviewer evaluates results
• Output-based evaluation: Test success/failure determined by reading output files, not shell script heuristics
• UI-focused: Test triggers (@, /, runtime pickers, keybindings) by checking UI elements appear
• Filesystem verification: Tool execution is verified through files under /tmp/kon-test-project

Quick Start

# Run all e2e tests from the repo root
bash .agents/skills/kon-tmux-test/run-e2e-tests.sh

# Optional: keep the temporary HOME for debugging
KEEP_E2E_HOME=1 bash .agents/skills/kon-tmux-test/run-e2e-tests.sh

# Optional: override launch command/provider/model
KON_CMD='uv run kon --model gpt-5.5' \
  bash .agents/skills/kon-tmux-test/run-e2e-tests.sh

After running, read /tmp/kon-test-*.txt and evaluate the captured pane/config/filesystem outputs.

Test Scripts

Setup Script: setup-test-project.sh

Creates a deterministic test project structure at /tmp/kon-test-project/.

bash .agents/skills/kon-tmux-test/setup-test-project.sh

Main Test Script: run-e2e-tests.sh

Runs comprehensive e2e tests including UI triggers, runtime controls, tab completion, and tool execution.

bash .agents/skills/kon-tmux-test/run-e2e-tests.sh

Test Categories

UI Trigger Tests (LLM-independent)

• / commands: Type /, verify slash command list appears with core and newer commands
• @ file search: Type @pyproject, verify file picker appears with pyproject.toml
• /model command: Type /model, verify model selector appears, then dismiss
• /new command: Type /new, verify new conversation is started
• /resume command: Type /resume, verify session list appears, then dismiss
• /session command: Type /session, verify session info/statistics displayed

Runtime Mode Tests (LLM-independent)

• /permissions picker: Shows prompt and auto, with current mode checked
• /permissions auto/prompt: Info bar updates (✓✓ auto / ⏸ prompt) and temp config persists permissions.mode
• Shift+Tab: Cycles permission mode and persists it in temp config
• /thinking picker: Shows none, minimal, low, medium, high, xhigh, with current level checked
• /thinking minimal: Info bar model/thinking area updates to minimal
• Ctrl+Shift+T: Cycles thinking level in the info bar
• /notifications picker: Shows on and off, with current mode checked
• /notifications on/off: Status says saved and temp config persists notifications.enabled
• Info bar row2 regression: Permission mode remains row2-left while model/provider/thinking remains row2-right after runtime changes

Tab Path Completion Tests (LLM-independent)

• Unique match: Type pypr + Tab, verify completes to pyproject.toml
• Multiple alternatives: Type src/kon/ui/s + Tab, verify floating list shows selection_mode.py, session_ui.py, styles.py
• Nested unique file: Type src/kon/ui/widg + Tab, verify completes to src/kon/ui/widgets.py
• Select from list: Type src/kon/ui/s + Tab + Enter, verify first completion is applied to input

Tool Execution Tests (Filesystem verification)

• Write tool: Creates /tmp/kon-test-project/test1.txt, verified by file existence
• Edit tool: Modifies test1.txt, verified by content changing from hello to world
• List files: Shows directory contents in captured pane
• Calculation: Computes 3+3, verified in LLM output where practical

Configuration

Edit or override environment variables for run-e2e-tests.sh:

WAIT_TIME=30                    # Time for LLM to complete all tool tasks
COMMAND_WAIT_TIME=3             # Time for UI commands to settle
SESSION_NAME="kon-test"         # Tmux session name
TEST_DIR="/tmp/kon-test-project" # Test project directory for tool execution
TEST_HOME="/tmp/kon-e2e-home"    # Isolated HOME/config/session directory
KON_DIR="$PWD"                  # Kon repo directory for tab completion tests
KON_CMD="uv run kon --model gpt-5.5"
KEEP_E2E_HOME=0                 # Set to 1 to preserve temp HOME after run

Output Files

The main script writes captured outputs to /tmp/kon-test-*.txt:

• /tmp/kon-test-1-commands.txt - / slash command list
• /tmp/kon-test-2-at-trigger.txt - @pyproject file picker
• /tmp/kon-test-3-model.txt - /model selector
• /tmp/kon-test-4-new.txt - /new result
• /tmp/kon-test-5-permissions-picker.txt - /permissions picker
• /tmp/kon-test-6-permissions-auto.txt and ...-config.txt - /permissions auto
• /tmp/kon-test-7-permissions-prompt.txt and ...-config.txt - /permissions prompt
• /tmp/kon-test-8-permissions-shift-tab.txt and ...-config.txt - Shift+Tab mode cycling
• /tmp/kon-test-9-thinking-picker.txt - /thinking picker
• /tmp/kon-test-10-thinking-minimal.txt - /thinking minimal
• /tmp/kon-test-11-thinking-cycle.txt - Ctrl+Shift+T thinking cycle
• /tmp/kon-test-12-notifications-picker.txt - /notifications picker
• /tmp/kon-test-13-notifications-on.txt and ...-config.txt - /notifications on
• /tmp/kon-test-14-notifications-off.txt and ...-config.txt - /notifications off
• /tmp/kon-test-15-tab-unique.txt - Tab completion unique match
• /tmp/kon-test-16-tab-multiple.txt - Tab completion alternatives
• /tmp/kon-test-17-tab-nested-unique.txt - Nested unique file completion
• /tmp/kon-test-18-tab-select.txt - Tab completion selection
• /tmp/kon-test-19-tools.txt - Tool execution turn
• /tmp/kon-test-20-session.txt - /session stats
• /tmp/kon-test-21-resume.txt - /resume session list
• /tmp/kon-test-files.txt - Test project file listing
• /tmp/kon-test-test1-content.txt - Final test1.txt content or FILE_NOT_FOUND
• /tmp/kon-test-session-files.txt - Session JSONL paths under temp HOME
• /tmp/kon-test-final-config.txt - Final temp config

Key Tmux Gotchas

• Use Escape not Esc: tmux recognizes Escape. Esc sends literal characters.
• Always clear input between tests: Use Escape to dismiss completions, then C-u to clear text.
• Completion selectors block input: Selectors intercept Enter/Escape; dismiss them before the next test.
• Shift+Tab: The script sends CSI Z via Escape '[' 'Z' rather than relying on a tmux key name.
• Ctrl+Shift+T: The script sends CSI-u Escape '[84;6u' because C-S-t often collapses to Ctrl+T.

Test Evaluation (by Kon/reviewer)

After running the test script, evaluate results by reading the output files.

What to Check

UI Trigger Tests:

• / test: Slash command list includes themes, permissions, thinking, notifications, init, compact, handoff, export, copy, login, logout
• @ test: File picker appears and shows pyproject.toml
• /model test: Model selector appears with model list/current markers
• /new test: Started new conversation appears
• /resume test: Session list appears with prior sessions
• /session test: Session info/statistics displayed

Runtime Mode Tests:

• /permissions picker shows prompt and auto, current item checked
• /permissions auto shows ✓✓ auto, saved status, and config has mode = "auto"
• /permissions prompt shows ⏸ prompt, saved status, and config has mode = "prompt"
• Shift+Tab toggles back to auto and config has mode = "auto"
• /thinking picker shows none, minimal, low, medium, high, xhigh
• /thinking minimal shows Thinking level changed to minimal and info bar row2-right includes minimal
• Ctrl+Shift+T changes info bar thinking level from minimal to the next level
• /notifications picker shows on and off, current item checked
• /notifications on/off status says saved and config flips enabled = true/false
• Permission mode remains in row2-left and model/provider/thinking remains row2-right

Tab Path Completion Tests:

• pypr + Tab shows pyproject.toml
• src/kon/ui/s + Tab shows selection_mode.py, session_ui.py, styles.py
• src/kon/ui/widg + Tab shows src/kon/ui/widgets.py
• src/kon/ui/s + Tab + Enter applies a selected completion

Tool Execution Tests:

• /tmp/kon-test-project/test1.txt exists
• /tmp/kon-test-test1-content.txt contains world
• /tmp/kon-test-files.txt lists test1.txt
• /tmp/kon-test-19-tools.txt shows relevant tool blocks/results

Tabular Report

Provide a summary showing:

• Test name
• Status (PASS/FAIL)
• Description/failure reason
• Overall success rate

IMPORTANT: Always offer the view command

After presenting the report, ALWAYS give the user this shell command so they can inspect raw captured outputs:

for f in /tmp/kon-test-*.txt; do printf "\n\033[1;36m▶▶▶ %s\033[0m\n" "$f"; awk 'NF{found=1} found{lines[++n]=$0} END{while(n>0 && lines[n]=="") n--; for(i=1;i<=n;i++) print lines[i]}' "$f"; done

Cleanup

# Test script auto-cleans tmux session and temp HOME unless KEEP_E2E_HOME=1.
# Output files remain for evaluation (/tmp/kon-test-*.txt).
# Manual cleanup if needed:
tmux kill-session -t kon-test 2>/dev/null
rm -rf /tmp/kon-test-project /tmp/kon-e2e-home
rm -f /tmp/kon-test-*.txt

Tmux Commands Reference

# Session management
tmux new-session -d -s <name> -c <dir> '<command>'
tmux kill-session -t <name>
tmux has-session -t <name>

# Input — IMPORTANT: use full key names (Escape, Enter, not Esc)
tmux send-keys -t <name> "text"
tmux send-keys -t <name> Enter
tmux send-keys -t <name> Escape
tmux send-keys -t <name> Tab
tmux send-keys -t <name> C-c
tmux send-keys -t <name> C-u

# Output
tmux capture-pane -t <name> -p
tmux capture-pane -t <name> -p > file.txt

Tips

• Tests are deterministic: project/config structure is recreated each run.
• Runtime mode tests are LLM-independent and should be checked first.
• Tab completion tests run from the kon repo to use known paths.
• Tool tests verify filesystem state; avoid relying solely on LLM prose.
• Use KEEP_E2E_HOME=1 to inspect temp config/session files after failures.
• Run tool execution before /resume so there is a session with messages in the list.