Manual first-run walkthrough
Issue #25, verification run on 2026-10-12 against loop/issue-25 at 2393e75. Testing-plan row: T11.
The real TUI was driven through tmux (send-keys / capture-pane). Below, the throwaway HOME is shown as ~.
1. Build and install
text
$ uv build --out-dir ~/.cache/i25/dist
Successfully built ~/.cache/i25/dist/tiny_harness-0.5.0-py3-none-any.whl
$ uv venv ~/.cache/i25/w && uv pip install -p ~/.cache/i25/w ~/.cache/i25/dist/*.whl2. Bare tiny-harness with an empty HOME
text
$ HOME=$(mktemp -d) tiny-harness
created default configuration ~/.tiny-harness/config.toml
configuration error: required secret is not set (OPENAI_API_KEY)
$ echo $?
2
$ stat -c '%a %n' ~/.tiny-harness ~/.tiny-harness/config.toml
700 ~/.tiny-harness
600 ~/.tiny-harness/config.toml- First run: the default file is created, then the run stops with exit 2. The generated
[openai]names a hosted model, and its secret comes only from the environment. - Pointing at Ollama: following the plan,
[openai]was then edited to the loopback Ollama (base_url = "http://127.0.0.1:11434/v1",api = "chat_completions",model = "qwen3:1.7b",models = ["qwen3:1.7b", "qwen3:8b"]).
3. The TUI: one conversation over several messages
First message. It parks in INPUT_REQUIRED · awaiting-next-message. The Task tab shows Settings and Sub-tasks.
text
tiny-harness · support-agent v0.1.0 · A2A 1.0 · http://127.… task d846…0207 · INPUT_REQUIRED · awaiting-next-message
╭──────────────────────────────────────────────────────────────────────────────╮ Task Plan Trace
│ you › Reply with one short sentence: what is 2+2? │╸━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
│ │ Goal
│ ── task → SUBMITTED │ Reply with one short sentence: what
│ ── task → WORKING │
│ agent › The task is to reply with "2+2 equals 4" as a short sentence. No │ the-looper reporter
│ tools are required for this simple calculation. │ tiny-harness assignee
│ complete_step │ Settings
│ step: answer_task │ model configured default
│ output: "2+2 equals 4" │ effort provider default
│ │ Sub-tasks
│ ── task → INPUT_REQUIRED · awaiting-next-message · your turn │ none
╰──────────────────────────────────────────────────────────────────────────────╯
▊ Enter to send as the-looper · / for commands ▎
^d complete ^c cancel f2 task/plan/trace f4 parent ▏^p paletteSecond message, /status and /effort low. The second message is on the same task. /status and /effort low reply with ⌘ lines, and the server log's model-call lines were the same before and after them (2 → 2), so neither made a model call.
text
tiny-harness · support-agent v0.1.0 · A2A 1.0 · http://127.… task d846…0207 · INPUT_REQUIRED · awaiting-next-message
╭──────────────────────────────────────────────────────────────────────────────╮ Task Plan Trace
│ output: "2+2 equals 4" │╸━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
│ │ Goal
│ ── task → INPUT_REQUIRED · awaiting-next-message · your turn │ Reply with one short sentence: what
│ you › And 3+3? One sentence. │
│ ── task → WORKING │ the-looper reporter
│ ▁▁ │ tiny-harness assignee
│ agent › complete_step │
│ step: answer_task │ Settings
│ output: "3+3 equals 6" │ model configured default
│ ── task → INPUT_REQUIRED · awaiting-next-message · your turn │ Sub-tasks
│ you › /status │
│ ▌ ⌘ Task d8464e7a-8cd3-4eea-9f52-37f4c6a80207 · Reply with one short │
│ ▌ sentence: what is 2+2? │
│ ▌ State: INPUT_REQUIRED · awaiting-next-message │
│ ▌ Model: qwen3:1.7b (default) │
│ ▌ Effort: provider default │
│ ▌ Plan: none │
│ ▌ Sub-tasks: none │
│ you › /effort low │
│ ▌ ⌘ Effort set to low. │
╰──────────────────────────────────────────────────────────────────────────────╯
▊ Enter to send as the-looper · / for commands ▎
^d complete ^c cancel f2 task/plan/trace f4 parent ▏^p palette/sub-task. The reply names the new id, and the sub-task row appears.
text
tiny-harness · support-agent v0.1.0 · A2A 1.0 · http://127.… task d846…0207 · INPUT_REQUIRED · awaiting-next-message
╭──────────────────────────────────────────────────────────────────────────────╮ Task Plan Trace
│ ── task → WORKING │ Goal
│ agent › complete_step │ is 2+2?
│ step: answer_task │
│ output: "3+3 equals 6" │ Participants
│ ── task → INPUT_REQUIRED · awaiting-next-message · your turn │ tiny-harness assignee
│ you › /status │ Settings
│ ▇▇ │ model configured default
│ ▌ ⌘ Task d8464e7a-8cd3-4eea-9f52-37f4c6a80207 · Reply with one short │ effort low
│ ▌ sentence: what is 2+2? │ Sub-tasks · Enter opens
│ ▌ State: INPUT_REQUIRED · awaiting-next-message │ d846…07.1 · INPUT_REQUIRED · Draft a
│ ▌ Model: qwen3:1.7b (default) │ one-line summary for the customer
│ ▌ Effort: provider default │
│ ▌ Plan: none │
│ ▌ Sub-tasks: none │
│ you › /effort low │
│ ▌ ⌘ Effort set to low. │
│ you › /sub-task Draft a one-line summary for the customer │
│ ▌ ⌘ Starting a sub-task. │
│ ▌ Created sub-task d8464e7a-8cd3-4eea-9f52-37f4c6a80207.1. │
╰──────────────────────────────────────────────────────────────────────────────╯
▊ Enter to send as the-looper · / for commands ▎
^d complete ^c cancel f2 task/plan/trace f4 parent ▏^p paletteOpening the sub-task. Tab moves focus to the sub-task list, and Enter opens it. The breadcrumb shows parent › task …07.1. The sub-task waits for its first message and has inherited effort low.
text
tiny-harness · support-agent v0.1.0 · A2A 1.0 ·… d846…0207 › task d846…07.1 · INPUT_REQUIRED · awaiting-next-message
╭──────────────────────────────────────────────────────────────────────────────╮ Task Plan Trace
│ ── task → SUBMITTED │╸━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
│ │ Goal
│ ── task → INPUT_REQUIRED · awaiting-next-message · your turn │ Draft a one-line summary for the
│ │ Sub-tasks
╰──────────────────────────────────────────────────────────────────────────────╯
▊ Enter to send as the-looper · / for commands ▎
^c cancel f2 task/plan/trace ^d complete f4 parent q quit ▏^p paletteCompleting both tasks:
ctrl+don the sub-task:completed by the-looper,COMPLETED.f4back to the parent. The parent wasWORKING, because the sub-task's completion notes the parent and starts its turn, as the agent's own sub-tasks do.ctrl+don the parent: the turn is interrupted, and the task endsCOMPLETED.ctrl+dagain:── already COMPLETED, so the key is inert.
text
tiny-harness · support-agent v0.1.0 · A2A 1.0 · http://127.0.0.1:8080/ task d846…0207 · COMPLETED
╭──────────────────────────────────────────────────────────────────────────────╮ Task Plan Trace
│ ── task → INPUT_REQUIRED · awaiting-next-message · your turn │ Goal
│ you › /status │ is 2+2?
│ ▌ ⌘ Task d8464e7a-8cd3-4eea-9f52-37f4c6a80207 · Reply with one short │ Participants
│ ▌ sentence: what is 2+2? │ the-looper reporter
│ ▌ State: INPUT_REQUIRED · awaiting-next-message │ tiny-harness assignee
│ ▌ Model: qwen3:1.7b (default) │
│ ▌ Effort: provider default │ Settings
│ ▌ Plan: none │ model configured default
│ ▌ Sub-tasks: none │ effort low
│ ▆▆ │ Sub-tasks · Enter opens
│ you › /effort low │ d846…07.1 · COMPLETED · Draft a
│ ▌ ⌘ Effort set to low. │
│ you › /sub-task Draft a one-line summary for the customer │
│ ▌ ⌘ Starting a sub-task. │
│ ▌ Created sub-task d8464e7a-8cd3-4eea-9f52-37f4c6a80207.1. │
│ ── task → WORKING │
│ agent › completed by the-looper │
│ ── task → COMPLETED │
╰──────────────────────────────────────────────────────────────────────────────╯
▊ Enter to send as the-looper · / for commands ▎
^c cancel f2 task/plan/trace ^d complete f4 parent q quit ▏^p palettetext
tiny-harness · support-agent v0.1.0 · A2A 1.0 · http://127.0.0.1:8080/ task d846…0207 · COMPLETED
╭──────────────────────────────────────────────────────────────────────────────╮ Task Plan Trace
│ you › /status │ Goal
│ ▌ ⌘ Task d8464e7a-8cd3-4eea-9f52-37f4c6a80207 · Reply with one short │ is 2+2?
│ ▌ sentence: what is 2+2? │
│ ▌ State: INPUT_REQUIRED · awaiting-next-message │ Participants
│ ▌ Model: qwen3:1.7b (default) │ the-looper reporter
│ ▌ Effort: provider default │ tiny-harness assignee
│ ▌ Plan: none │
│ ▌ Sub-tasks: none │ Settings
│ you › /effort low │ effort low
│ ▁▁ │ Sub-tasks · Enter opens
│ ▌ ⌘ Effort set to low. │ d846…07.1 · COMPLETED · Draft a
│ you › /sub-task Draft a one-line summary for the customer │
│ ▌ ⌘ Starting a sub-task. │
│ ▌ Created sub-task d8464e7a-8cd3-4eea-9f52-37f4c6a80207.1. │
│ ── task → WORKING │
│ agent › completed by the-looper │
│ ── task → COMPLETED │
│ ── already COMPLETED │
╰──────────────────────────────────────────────────────────────────────────────╯
▊ Enter to send as the-looper · / for commands ▎
^c cancel f2 task/plan/trace ^d complete f4 parent q quit ▏^p palette4 and 5. --config wins; a symlinked home is refused
text
$ tiny-harness --config examples/demo/config.ollama.toml --help | tail -4
--version show program's version number and exit
--config CONFIG TOML configuration file (secrets from env)
configuration: examples/demo/config.ollama.toml (--config)
rc=0
$ ls -A $HOME
(empty)
$ ln -s elsewhere ~/.tiny-harness; tiny-harness
configuration error: refusing the default configuration: the directory is a symlink (~/.tiny-harness)
rc=2Findings
- Every T11 step passed.
- Keyboard reach: Tab cycles four stops, and the sub-task list is one of them. A keyboard-only user reaches it, but there is no direct key for it. That is enough for R9.4; a dedicated key could be a follow-up.
- First run: the first run ends with exit 2 until a model is set up (an
OPENAI_API_KEY, orbase_url). The message names the missing variable. This matches R2 and the fail-closed rule: there is no silent fallback.