juno-code 1.0.50 → 1.0.53

package/README.md

@@ -30,6 +30,23 @@ juno-code init --task "Your task description" --subagent claude
 juno-code init --task "Your task description" --subagent pi
 ```
 
+### Shell Completion (Tab Autocomplete)
+
+```bash
+# Install completion for your current shell
+juno-code completion install
+
+# Or explicitly target a shell
+juno-code completion install bash
+juno-code completion install zsh
+juno-code completion install fish
+
+# Check status
+juno-code completion status
+```
+
+After installation/reload, `juno-code c<TAB><TAB>` suggests available subcommands.
+
 ---
 
 ## The Ralph Method: Where It All Started
@@ -200,6 +217,22 @@ literal text with `backticks`
 EOF
 ```
 
+### Prompt-time command substitution (per iteration)
+
+`juno-code` also supports explicit prompt-time shell substitutions that run inside the working directory on **every engine iteration**:
+
+- `!'command'`
+- `!\`\`\`command\`\`\``
+
+Examples:
+
+```bash
+juno-code claude -i 3 -p "Summarize git status: !'git status --short'"
+juno-code claude -i 2 -p "Recent commits:\n!```git log -n 5 --oneline```"
+```
+
+This avoids relying on your shell’s one-time backtick expansion and keeps command output fresh across retries/iterations.
+
 ## CLI Reference
 
 ### Core Commands
@@ -222,6 +255,9 @@ juno-code codex 'your task'
 juno-code gemini 'your task'
 juno-code pi 'your task'
 
+# Pi live interactive run (auto-exits on non-aborted completion)
+juno-code pi --live -p '/skill:ralph-loop' -i 1
+
 # AI-powered test generation
 juno-code test --generate --framework vitest
 juno-code test --run
@@ -243,6 +279,7 @@ juno-code view-log .juno_task/logs/claude_shell_*.log --output json-only --limit
 | `-v, --verbose` | Human-readable verbose output |
 | `-r, --resume <id>` | Resume specific session |
 | `--continue` | Continue most recent session |
+| `--live` | Pi-only: run Pi in interactive TUI mode with auto-exit on non-aborted completion |
 | `--no-hooks` | Skip lifecycle hooks |
 | `--on-hourly-limit <action>` | Quota limit behavior: `wait` (auto-retry) or `raise` (exit) |
 | `--force-update` | Force reinstall all scripts and services |
@@ -252,12 +289,44 @@ juno-code view-log .juno_task/logs/claude_shell_*.log --output json-only --limit
 ### Session Management
 
 ```bash
-juno-code session list           # View all sessions
-juno-code session info abc123    # Session details
-juno-code --resume abc123 -p 'continue'  # Resume session
-juno-code --continue -p 'keep going'     # Continue most recent
+juno-code session list                # View all sessions
+juno-code session info abc123         # Session details
+juno-code --resume abc123 -p 'continue'   # Resume session
+juno-code --continue -p 'keep going'      # Continue most recent (backend-native)
+juno-code continue 'next prompt'          # Reuse last session id + runtime settings snapshot
 ```
 
+Each `juno-code` run also appends execution history to:
+
+- `.juno_task/session_history.json` (unlimited, newest-first)
+
+Per-run entries include: initial prompt + timestamp, subagent/model/settings, total cost, turn/message counts, session IDs, and last-message timestamp.
+
+CLI run summaries also surface these fields live in the terminal:
+- `Statistics -> Total Cost`
+- `Statistics -> Completed At`
+- `Statistics -> Average Duration` (humanized unit: ms/s/m/h)
+- `Session ID(s)` entries with per-session cost when available
+
+For `juno-code continue`, the latest session context is persisted into the project env file (`.env.juno` by default) using **shell-scoped keys** so panes/tabs do not overwrite each other:
+- `JUNO_CODE_LAST_SESSION_ID_SCOPE_<HASH>`
+- `JUNO_CODE_LAST_EXECUTION_SETTINGS_SCOPE_<HASH>` (JSON runtime-settings snapshot)
+
+Scope detection prefers terminal markers (for example `TMUX_PANE`, `WEZTERM_PANE`, `TERM_SESSION_ID`) and falls back to the parent shell PID. You can override scope resolution explicitly with `JUNO_CODE_CONTINUE_SCOPE=<name>`.
+
+Short help text for scripts that need pane-scoped continue state:
+- Session key pattern: `JUNO_CODE_LAST_SESSION_ID_SCOPE_<HASH>`
+- Settings key pattern: `JUNO_CODE_LAST_EXECUTION_SETTINGS_SCOPE_<HASH>`
+- Deterministic override: set `JUNO_CODE_CONTINUE_SCOPE=<your-pane-id>` before running `juno-code` so external scripts can target a stable scope name across runs.
+
+Script endpoint for hash/status lookups:
+```bash
+juno-code continue-scope --json          # current scope hash + status
+juno-code continue-scope A1B2C3 --json   # lookup by short hash prefix (5-6 chars)
+```
+
+`continue-scope` returns `status` as one of: `running`, `finished`, `not_found`, `error`.
+
 ### Feedback System
 
 ```bash
@@ -289,9 +358,9 @@ juno-code skills status
 
 | Agent | Directory | Skills |
 |-------|-----------|--------|
-| Claude | `.claude/skills/` | `ralph-loop`, `plan-kanban-tasks`, `understand-project` |
-| Codex | `.agents/skills/` | `ralph-loop` |
-| Pi | `.pi/skills/` | (planned) |
+| Claude | `.claude/skills/` | `kanban-workflow`, `ralph-loop`, `plan-kanban-tasks`, `understand-project` |
+| Codex | `.agents/skills/` | `kanban-workflow`, `ralph-loop`, `plan-kanban-tasks`, `understand-project` |
+| Pi | `.pi/skills/` | `kanban-workflow`, `ralph-loop`, `plan-kanban-tasks`, `understand-project` |
 
 ### Service Management
 
@@ -306,6 +375,18 @@ juno-code services status
 juno-code services install --force
 ```
 
+### Auth Management (Codex → Pi)
+
+```bash
+# Import default Codex auth into Pi auth store
+juno-code auth import-codex
+
+# Use explicit input/output paths (useful for account switching/backup files)
+juno-code auth import-codex --input ~/.codex/auth.json --output ~/.pi/agent/auth.json
+```
+
+This command translates Codex CLI credentials to Pi's `auth.json` format (`type: "oauth"`) and writes/updates the `openai-codex` provider entry.
+
 ## Backends & Services
 
 ### Supported Services
@@ -315,11 +396,31 @@ juno-code services install --force
 | claude | `claude-sonnet-4-6` | `:haiku`, `:sonnet`, `:opus` |
 | codex | `gpt-5.3-codex` | `:codex`, `:codex-mini`, `:gpt-5`, `:mini` |
 | gemini | `gemini-2.5-pro` | `:pro`, `:flash`, `:pro-3`, `:flash-3` |
-| pi | `anthropic/claude-sonnet-4-6` | `:pi`, `:sonnet`, `:opus`, `:gpt-5`, `:codex`, `:api-codex`, `:gemini-pro` |
+| pi | `anthropic/claude-sonnet-4-6` | `:pi`, `:sonnet`, `:opus`, `:gpt-5`, `:codex`, `:api-codex`, `:codex-spark`, `:api-codex-spark`, `:gemini-pro` |
 
 > **Pi** is a multi-provider coding agent that supports Anthropic, OpenAI, Google, Groq, xAI, and more.
 > It requires separate installation: `npm install -g @mariozechner/pi-coding-agent`
 
+### Pi Live Mode (`--live`)
+
+Use live mode when you want Pi's interactive TUI while keeping juno-code iteration hooks/statistics.
+
+```bash
+# Canonical live flow
+juno-code pi --live -p '/skill:ralph-loop' -i 1
+
+# If :pi default model is unavailable in your Pi provider setup, pick an explicit available model
+juno-code pi --live -m :api-codex -p '/skill:ralph-loop' -i 1
+```
+
+Notes:
+- `--live` is validated as **Pi-only** (`juno-code pi ...`).
+- `--live` requires extensions enabled (`--no-extensions` is incompatible).
+- Live auto-exit is triggered on non-aborted `agent_end` only. Pressing `Esc` to interrupt the current run keeps Pi open so you can continue interacting.
+- To manually leave Pi and return control to juno-code hooks/loop, use Pi's normal exit keys (for example `Ctrl+C` twice quickly or `Ctrl+D` on an empty editor).
+- Best experience is an interactive terminal (TTY) so Pi TUI can manage screen state cleanly.
+- Pi TUI depends on the Node runtime used to launch Pi; use a modern Node version (Node 20+) in PATH.
+
 ### Custom Backends
 
 Service scripts live in `~/.juno_code/services/`. Each is a Python script that accepts standard args (`-p/--prompt`, `-m/--model`, `-v/--verbose`) and outputs JSON events to stdout.
@@ -448,6 +549,30 @@ Orchestrate N concurrent juno-code processes with queue management, structured o
 ./.juno_task/scripts/parallel_runner.sh --stop-all
 ```
 
+#### Dedicated Example: SEO landing-page batch in tmux panes
+
+Use this pattern when you want to generate many related content tasks in parallel while keeping live visibility per worker pane:
+
+```bash
+./.juno_task/scripts/parallel_runner.sh \
+  -s pi \
+  -m zai/glm-5 \
+  --kanban-filter "--tag SEO_LANDING_PAGES --limit 200 --status backlog,in_progress,todo" \
+  --parallel 5 \
+  --tmux panes \
+  --prompt-file ./tmp_prompt/content_gen.md
+```
+
+What each flag does:
+
+- `-s pi -m zai/glm-5`: run workers with Pi on a specific model.
+- `--kanban-filter "..."`: dynamically pull task IDs from kanban (here: only `SEO_LANDING_PAGES`, up to 200, only open statuses).
+- `--parallel 5`: execute up to 5 tasks concurrently.
+- `--tmux panes`: split workers into panes for side-by-side monitoring.
+- `--prompt-file ./tmp_prompt/content_gen.md`: keep a reusable, versioned instruction template instead of long inline prompts.
+
+Tip: keep the filter string quoted so it is passed as one argument to `parallel_runner.sh` and then correctly forwarded to `kanban.sh`.
+
 #### Output & Extraction
 
 - **Per-task JSON**: `{output_dir}/{task_id}.json` with exit code, wall time, extracted response
@@ -660,6 +785,30 @@ The kanban.sh script wraps juno-kanban. Here are the actual commands:
 4. Global config files
 5. Hardcoded defaults
 
+### Per-subagent default models
+
+Set model defaults per subagent without changing your global default:
+
+```bash
+juno-code pi set-default-model :api-codex
+juno-code claude set-default-model :opus
+juno-code codex set-default-model :gpt-5
+```
+
+This writes to `.juno_task/config.json`:
+
+```json
+{
+  "defaultModels": {
+    "pi": ":api-codex",
+    "claude": ":opus",
+    "codex": ":gpt-5"
+  }
+}
+```
+
+`juno-code` resolves models in this order: CLI `--model` → configured subagent default (`defaultModels` / legacy `defaultModel`) → built-in default.
+
 ### Project Env Bootstrap (`.env.juno`)
 
 `juno-code` now bootstraps a project env file automatically: