[codex] Remove app orchestration surface

.claude/skills/tycho/SKILL.md

@@ -1,21 +1,15 @@
 ---
 name: tycho
-description: Manages Tycho-monitored projects and managed agents. Use when the user asks to deploy, check status, manage maintenance mode, create/list/run/stop/send/archive/clone agents, or control schedules for any project.
+description: Manages Tycho-monitored projects and managed agents. Use when the user asks to check status, create/list/run/stop/send/archive/clone agents, or control schedules for any project.
 ---
 
 # Tycho CLI Skill
 
-Manage Kamal-deployed projects, managed agents, and scheduled runs via the `tycho` CLI.
 
 ## Quick Reference
 
 | Group | Command | Description |
 |-------|---------|-------------|
-| **app** | `app list` | List projects with Kamal deployment |
-| | `app status <project-key>` | Check a project's health and last action |
-| | `app deploy <project-key>` | Start a deploy |
-| | `app maintenance <project-key>` | Enable maintenance mode |
-| | `app live <project-key>` | Resume live traffic |
 | **project** | `project update <project-key> --pr-url <url>` | Set / clear open PR URL |
 | **agent** | `agent create <project-key> <prompt>` | Create (and optionally run) a managed agent |
 | | `agent list [<project-key>]` | List agents, optionally filtered by project |
@@ -161,20 +155,6 @@ tycho agent clone my-project-agent-3 --run    # clone and start immediately
 
 ---
 
-## `tycho app` — Deployment Commands
-
-```bash
-tycho app list
-tycho app status my-project
-tycho app deploy my-project
-tycho app maintenance my-project
-tycho app live my-project
-```
-
-Actions run as detached background processes. Check progress via `app status` or the TUI.
-
----
-
 ## `tycho schedule` — Schedule Management
 
 ```bash
@@ -200,7 +180,6 @@ tycho project update my-project --pr-url ""   # clear
 ## Project Keys
 
 ```bash
-tycho app list                              # Kamal-enabled projects
 grep "^- key:" ~/.tycho/config/hq.yml      # all projects (including agent-only)
 ```
 

.github/ISSUE_TEMPLATE/feature_request.md

@@ -14,5 +14,4 @@ assignees: ""
 
 ## Notes
 
-Mention affected areas such as TUI, Remote UI, Kamal actions, managed agents,
 hooks, config, or docs.

AGENTS.md

@@ -4,11 +4,10 @@
 
 `bin/tycho` is the executable and boots through `HQ::CLI` in `lib/hq/cli.rb`. Keep the main Bubbletea model and screen-level update flow in `lib/hq/app.rb`, config loading in `lib/hq/registry.rb`, terminal input shims in `lib/hq/bubbletea_input.rb`, domain and process-management logic under `lib/hq/domain/`, form/composer components under `lib/hq/ui/components/`, and rendering split between the aggregator in `lib/hq/ui/rendering.rb` and focused modules under `lib/hq/ui/rendering/`.
 
-Domain files are intentionally small: `constants.rb` owns log/schema paths, `version_lookup.rb` handles RubyGems lookups, `kamal_action.rb` manages detached Kamal commands, `app_project.rb` handles project metadata and health checks, `managed_agent.rb` manages Codex/Claude-compatible execution and structured results, and `agent_store.rb` persists managed agents.
 
 Project definitions live in `~/.tycho/config/hq.yml`, system prompt templates live in `~/.tycho/config/system_prompts.yml`, and structured managed-agent output is described by `~/.tycho/config/schemas/agent_result.json`. Project status, key decisions, and roadmap live in `docs/PROJECT_STATUS.md`; update it when durable priorities, milestones, or architectural decisions change. Research and workflow notes live under `docs/`, including `docs/GOTCHAS.md`, `docs/REMOTE_SERVER.md`, `docs/research/charm-ruby.md`, `docs/research/codex-json-schema-research.md`, and `docs/research/claude-json-schema-research.md`.
 
-Runtime artifacts are written to `~/.tycho/logs/`, including app state files such as `actions.json` and `managed_agents.json`, application logs in `hq.log`, project logs under `~/.tycho/logs/projects/{project}/`, archived project logs under `~/.tycho/logs/projects/archived/`, and per-agent logs/status/result files under `~/.tycho/logs/agents/`. Keep automated checks under `test/`; the existing rendering regression coverage lives in `test/rendering_test.rb`.
+Runtime artifacts are written to `~/.tycho/logs/`, including app state files such as `managed_agents.json`, application logs in `hq.log`, project logs under `~/.tycho/logs/projects/{project}/`, archived project logs under `~/.tycho/logs/projects/archived/`, and per-agent logs/status/result files under `~/.tycho/logs/agents/`. Keep automated checks under `test/`; the existing rendering regression coverage lives in `test/rendering_test.rb`.
 
 ## Build, Test, and Development Commands
 
@@ -37,19 +36,16 @@ If you introduce new tooling, document the command here and keep it runnable fro
 
 Follow the existing Ruby style across `bin/tycho` and `lib/hq/**/*.rb`: two-space indentation, snake_case for methods and variables, SCREAMING_SNAKE_CASE for constants, and short guard clauses where they simplify flow. Keep classes and modules focused, and prefer small helper methods over deeply nested conditionals.
 
-Preserve the current separation of concerns: registry/config parsing stays out of the TUI layer, managed-agent and Kamal behavior belongs in domain objects, and screen/layout code should stay in the UI modules.
 
 Preserve the current file-level conventions: `# frozen_string_literal: true`, double-quoted strings, and concise comments only where the code is not obvious.
 
 The repo ships a `.rubocop.yml` that pins double-quoted string style and Ruby 3.4. `bundle exec rubocop` is not part of the bundle; if you run RuboCop locally, use the available executable and keep changes scoped to the requested files.
 
 ## Runtime Behavior & External Integrations
 
-Health checks run concurrently via Ruby threads. `AppProject#check_health!` uses HEAD requests against the configured healthcheck path and root URL; keep HEAD semantics because kamal-proxy maintenance detection depends on them. A root URL 503 means the app is in maintenance mode even if the health endpoint looks healthy.
 
-Kamal actions run as detached background processes through `mise exec`, using `TYCHO_MISE_BIN` first, then common local install paths, then `mise` on `PATH`. Prefer a project's `bin/kamal` binstub and fall back to `bundle exec kamal`. Logs go to `~/.tycho/logs/projects/{project}/action.log`, action state is persisted to `~/.tycho/logs/actions.json`, and actions are restored and checked for liveness on startup. Archiving a project moves its config entry from `~/.tycho/config/hq.yml` to `~/.tycho/config/hq.archived.yml`, moves `~/.tycho/logs/projects/{project}/` to `~/.tycho/logs/projects/archived/YYYY-MM-DD_project-name/`, and archives related managed-agent logs.
 
-The app auto-refreshes every 30 seconds. Action and agent status is polled every 10 seconds. Multiple projects can have concurrent actions running, and completed actions trigger a health refresh.
+The app auto-refreshes every 30 seconds. Agent status is polled every 10 seconds.
 
 Managed agents are configured from project settings in `~/.tycho/config/hq.yml`, with prompt templates loaded from `~/.tycho/config/system_prompts.yml`. Codex agents use JSON output and `~/.tycho/config/schemas/agent_result.json`; Claude and custom Claude-compatible harnesses use `--output-format stream-json` so logs stream incrementally. Use `TYCHO_CODEX_BIN` and `TYCHO_CLAUDE_BIN` to override built-in agent executables. Custom Claude-compatible wrappers belong in `custom_harnesses` with `adapter: claude` and an `execution_command`; provider-specific details should live in that wrapper or command configuration, not in HQ. Native Claude/Codex `session_id` values are persisted on managed agents and reused with `--resume` after the first run; HQ still treats `memory.jsonl` as the canonical transcript and only replays the bounded memory prompt when no native session is known. Preserve structured inquiry submission and focus-aware chat behavior when changing agent flows.
 
@@ -71,7 +67,7 @@ For input components, keep paste handling compatible with Bubbles `TextInput` an
 
 This repository now has lightweight automated coverage under `test/`. Every change should at minimum pass `bin/test` and a manual run of `bin/tycho` when TUI behavior is affected.
 
-Validate the affected key paths in the UI, especially grouped project rows, table alignment, detail views, sidebar log inspection, agent create/edit flows, agent chat and structured inquiry submission, refresh, deploy, maintenance toggle, and the `g` shortcut that opens the selected project in a terminal.
+Validate the affected key paths in the UI, especially grouped project rows, table alignment, detail views, sidebar log inspection, agent create/edit flows, agent chat and structured inquiry submission, refresh, and the `g` shortcut that opens the selected project in a terminal.
 
 For Remote UI `/ui` changes, run `bundle exec ruby test/remote_server_test.rb` and do browser verification for user-visible behavior. A safe fallback pattern is to start `bin/tycho serve` on a spare localhost port with temp env vars (`TYCHO_CONFIG_PATH`, `TYCHO_SYSTEM_PROMPTS_PATH`, `TYCHO_LOGS_ROOT`), create fixture data through the JSON API, then drive `http://127.0.0.1:{port}/ui` with Playwright + local Google Chrome. Check concrete browser facts such as focused form values surviving `refresh({ force: true })`, details toggles preserving state across polling, mutually exclusive panels closing as expected, and sticky/fixed docks staying pinned inside the viewport.
 
@@ -83,6 +79,6 @@ When adding tests, keep using simple Ruby test files under `test/` with `*_test.
 
 Recent commits use short, imperative subjects such as `Fix table column alignment across screens` and `Add custom Claude harness support`. Keep commit messages in that style and scope each commit to one logical change.
 
-Pull requests should include a brief summary, manual verification steps, and screenshots or terminal captures for visible TUI changes. Link related issues when applicable and note any changes to `~/.tycho/config/hq.yml`, `~/.tycho/config/system_prompts.yml`, structured agent schemas, hardcoded project paths, logs, or external dependencies such as `mise`, RubyGems API access, Codex, Claude, or Bedrock-backed Claude execution.
+Pull requests should include a brief summary, manual verification steps, and screenshots or terminal captures for visible TUI changes. Link related issues when applicable and note any changes to `~/.tycho/config/hq.yml`, `~/.tycho/config/system_prompts.yml`, structured agent schemas, hardcoded project paths, logs, or external dependencies such as Codex, Claude, or Bedrock-backed Claude execution.
 
 Also call out changes to Bubbletea input handling, bracketed paste behavior, or Bubbles text components because they affect chat, inquiry, and agent editor typing flows.

CHANGELOG.md

@@ -19,8 +19,7 @@ All notable changes to Tycho will be documented in this file.
 ### Notable Releases
 
 - Add persisted Remote UI server switching for configured `remote_servers`,
-  including backend proxy routes, selected-server persistence, and peer health
-  display. (#39)
+  including backend proxy routes and selected-server persistence. (#39)
 - Add Remote UI skill autocomplete and a quick agent switcher for faster chat
   composition and agent navigation. (#38)
 - Show run summaries in Remote UI conversations and allow any Remote UI
@@ -87,8 +86,6 @@ All notable changes to Tycho will be documented in this file.
 - Improve the Remote UI Summary surface and keep run summary blocks out of the
   main conversation stream. (#11)
 
-- Fix Remote readiness checks by sharing executable resolution for `mise`,
-  Kamal, Codex, Claude, and compatible harnesses. (#12)
 - Fix Claude structured output schema handling so Claude-compatible runs can
   produce and parse structured results reliably. (#13)
 - Resume stopped schedules safely after interactive scheduled agents are