ln-010-dev-environment-setup

What this skill does

> **Paths:** File paths (`references/`, `../ln-*`) are relative to this skill directory.

# Dev Environment Setup

**Type:** L2 Domain Coordinator
**Category:** 0XX Shared

Runtime-backed coordinator for environment setup. The runtime is the execution SSOT. Worker outputs are standalone summaries, not chat prose.

## MANDATORY READ

Load these before execution:
- `references/coordinator_runtime_contract.md`
- `references/environment_setup_runtime_contract.md`
- `references/environment_worker_runtime_contract.md`
- `references/coordinator_summary_contract.md`
- `references/agent_skill_roots_contract.md`
- `references/environment_state_contract.md`
- `references/environment_state_schema.json`

MCP servers are targets of this workflow, not a prerequisite for starting it. If a server is disconnected or unavailable, continue with file and CLI inspection and report that surface as `disconnected` or `skipped` instead of failing the coordinator.

## When to Use

- First-time environment setup
- Agent/MCP drift after installs or updates
- Marketplace, plugin, MCP, or Codex execution-default drift across Claude and Codex
- Instruction file audit or repair

## Inputs

| Parameter | Required | Default | Description |
|-----------|----------|---------|-------------|
| `targets` | No | `both` | `claude`, `codex`, or `both` |
| `dry_run` | No | `false` | Plan without mutating |
| `plugins` | No | `agile-workflow` | Pass-through to ln-013. Use an explicit plugin list, or `all` to request every marketplace plugin. Optional plugins are never installed silently. |
| `auto_install_providers` | No | `false` | Pass-through to ln-013 MCP provider check. When `false`, provider checks are detection-only. |
| `apply_ide_override` | No | `false` | Pass-through to ln-012 Phase 6b. When `true`, ln-012 may write `claudeCode.initialPermissionMode` and `claudeCode.allowDangerouslySkipPermissions` to Cursor / VSCode user settings after explicit user consent. When `false` (default), Phase 6b is detection-only and reports drift without mutating IDE settings. |

## Runtime

Runtime family: `environment-setup-runtime`

Identifier:
- `targets-{normalizedTargets}`

Phases:
1. `PHASE_0_CONFIG`
2. `PHASE_1_ASSESS`
3. `PHASE_1B_PROVIDER_SELECTION`
4. `PHASE_2_DISPATCH_PLAN`
5. `PHASE_3_WORKER_EXECUTION`
6. `PHASE_4_VERIFY`
7. `PHASE_5_WRITE_ENV_STATE`
8. `PHASE_6_SELF_CHECK`

Terminal phases:
- `DONE`
- `PAUSED`

## Phase Map

### Phase 1: Assess

Collect one environment snapshot:
- agent availability and versions
- MCP registration/connection state
- hook state
- marketplace/plugin alignment state
- Codex skill-root discovery health: active roots, cache roots, duplicate skill names, `known_marketplaces.json` install-location drift, whether cache is visible under `~/.codex/skills`, whether `~/.codex/skills` is a whole-root junction to `~/.claude/plugins`, and whether `approval_policy=never` plus `sandbox_mode=danger-full-access` are already aligned
- graph provider dependency status: for each detected project language, check whether the corresponding graph provider binaries are installed (e.g. `basedpyright` for Python, `csharp-ls` for C#). Use `mcp__hex-graph__install_graph_providers` with `mode: "check"` if hex-graph MCP is connected
- graph index freshness: whether hex-graph is connected and the project has been indexed in this session (detection-only, no indexing here)
- instruction file state
- disabled flags from `.hex-skills/environment_state.json` if present
- task management provider availability detection (Linear MCP, gh CLI / Projects v2 readiness, file always available); explicit selection happens in Phase 1b
- research tool detection (Ref → Context7 → websearch)
- git worktree availability
- Claude Code IDE extension state: glob `~/.cursor/extensions/anthropic.claude-code-*` and `~/.vscode/extensions/anthropic.claude-code-*`. For each found, read `claudeCode.initialPermissionMode` and `claudeCode.allowDangerouslySkipPermissions` from the matching IDE user settings.json. Detection-only — no writes here.

Checkpoint payload:
- `assess_summary`

### Phase 1b: Tracker Provider Selection

User selects the task tracker provider (`linear`, `github`, `file`). Selection is persisted in `.hex-skills/environment_state.json` -> `task_management.provider` and reused by every task-related skill (`ln-200`, `ln-220`, `ln-300`, `ln-401`, `ln-130`, etc.).

**MANDATORY READ:** Load `references/tracker_provider_contract.md` before this phase.

Codex `exec --full-auto` cannot prompt the user mid-run. Selection MUST go through runtime `pause` + `pending_decision` + `record-decision` (NOT `AskUserQuestion`).

Decision tree:

1. Load Phase 1 availability data and existing `task_management` from env state.
2. Compute `available = filter([linear, github, file], byAvailability)`:
   - `linear` available iff Linear MCP server is connected.
   - `github` available iff `gh auth status` ok AND repository detected AND (no project_number yet, OR existing project has Status field with all 7 canonical options: Backlog, Todo, In Progress, To Review, To Rework, Done, Canceled).
   - `file` always available.
3. If `task_management.provider` is already set AND `task_management.status == "active"`:

```bash
node references/scripts/environment-setup-runtime/cli.mjs pause \
  --identifier {identifier} \
  --reason "Confirm tracker provider" \
  --payload '{"kind":"tracker_provider_confirmation","question":"Keep current tracker {provider}?","choices":["keep","switch"],"resume_to_phase":"PHASE_1B_PROVIDER_SELECTION"}'
```

Resume with `set-decision`. On `keep` -> reuse existing provider. On `switch` -> fall through to step 4.

4. If no current provider OR user chose `switch` AND `available.length > 1`:

```bash
node references/scripts/environment-setup-runtime/cli.mjs pause \
  --identifier {identifier} \
  --reason "Select tracker provider" \
  --payload '{"kind":"tracker_provider_selection","choices":<available>,"resume_to_phase":"PHASE_1B_PROVIDER_SELECTION"}'
```

5. If `available.length == 1`, do NOT pause. Auto-select with `selected_by: "single_option"` and write the reason (e.g. `"Only file mode available because: Linear MCP disconnected, gh CLI not authenticated"`).
6. Checkpoint Phase 1b:

```bash
node references/scripts/environment-setup-runtime/cli.mjs checkpoint \
  --identifier {identifier} \
  --phase PHASE_1B_PROVIDER_SELECTION \
  --payload '{"provider_selection":{"chosen":"<provider>","available":<available>,"reason":"<text>","selected_by":"user|single_option"}}'
```

Checkpoint payload:
- `provider_selection.chosen`
- `provider_selection.available`
- `provider_selection.reason`
- `provider_selection.selected_by`

### Phase 2: Dispatch Plan

Build selective dispatch plan. Only invoke workers that have work.

Dispatch precedence:

- If marketplace/plugin drift, Codex discovery violation (including whole-root junction), or Codex execution-default drift is present, `ln-013-config-syncer` becomes mandatory before Codex can be reported as healthy.
- If hex-graph MCP is registered and either (a) graph provider deps are missing for detected project languages, or (b) the project has not been indexed, `ln-012-mcp-configurator` becomes mandatory even when MCP registration is already complete. ln-012 owns both graph provider dependency installation (PHASE_3) and graph indexing (PHASE_5).

Workers:
- `ln-011-agent-installer`
- `ln-012-mcp-configurator`
- `ln-013-config-syncer`
- `ln-014-agent-instructions-manager`

Standalone skills packaged in setup-environment but not dispatched by ln-010:
- `ln-015-hex-line-uninstaller` remains direct-invocation only because it removes existing Claude-side integration surfaces.

Checkpoint payload:
- `dispatch_plan`

### Phase 3: Worker Execution

Invoke only selected workers. Do not re-probe the whole environment between worker calls.

Phase 1 assessment is the shared discovery snapshot. Materialize child manifests from that snapshot and pass them into managed worker runs.

Fo