feat(conductor): coordinate planning and building sessions

CHANGELOG.md

@@ -10,6 +10,8 @@ Tracked in [ROADMAP.md](ROADMAP.md). v2.1.x candidates carrying forward from v2.
 
 ### For LaunchPad users
 
+- **Run planning and building in separate sessions.** `/lp-conductor` sets up one session to review the work of a planning session and a building session, answer their routine questions, and bring you only the decisions that need you. It works with Claude Code and Codex sessions; see [How It Works](docs/guides/HOW_IT_WORKS.md#lp-conductor).
+
 - **Commits work in git worktrees under folders with spaces.** The bootstrapped `commit-msg` hook now quotes the message-file path. Before, a worktree under a folder such as `My Projects` failed every commit with `unrecognized arguments`. In a project bootstrapped earlier, put double quotes around `{1}` in the `restamp-history` line of `lefthook.yml`.
 
 ## [v2.2.2]

README.md

@@ -191,7 +191,7 @@ LaunchPad is MIT-licensed and open-source. Every claim above maps to a file you
 | ------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | 5-file governance kernel             | Open `docs/architecture/REPOSITORY_STRUCTURE.md`, `lefthook.yml`, `.launchpad/config.yml`, `.harness/`, `docs/architecture/`. Tests verify the file set materializes on `/lp-define`.                                       |
 | Compound learning loop               | Walk `docs/solutions/` and read any entry. 14-category taxonomy with YAML-validated frontmatter.                                                                                                                            |
-| 42 commands, 39 agents, 16 skills    | Browse `plugins/launchpad/commands/`, `plugins/launchpad/agents/`, `plugins/launchpad/skills/`.                                                                                                                             |
+| 43 commands, 39 agents, 16 skills    | Browse `plugins/launchpad/commands/`, `plugins/launchpad/agents/`, `plugins/launchpad/skills/`.                                                                                                                             |
 | Multi-agent review dispatch          | Read `.launchpad/agents.yml` and the 16 review agents in `plugins/launchpad/agents/review/`. Dispatch logic is in the `/lp-review` command file.                                                                            |
 | 0.60 confidence threshold + P1 floor | Specified in the `/lp-review` command file and enforced in code. Suppression audit trail at `.harness/observations/`.                                                                                                       |
 | Stack-aware dispatch                 | `stack_scope:` frontmatter on every agent file. Test coverage: `test_define` + `test_pipeline_matrix` (1,457 passing as of v2.1.3).                                                                                         |
@@ -230,7 +230,7 @@ LaunchPad ships as a plugin for Claude Code and Codex with:
 
 | Component       | Count     | What it covers                                                                          |
 | --------------- | --------- | --------------------------------------------------------------------------------------- |
-| Slash commands  | 42        | The brainstorm, define, plan, build, review, resolve, ship, and learn lifecycle         |
+| Slash commands  | 43        | The brainstorm, define, plan, build, review, resolve, ship, and learn lifecycle         |
 | Sub-agents      | 39        | 6 namespaces: research, review, resolve, design, skills, document-review                |
 | Skills          | 16        | Reusable instruction sets for design, planning, review, compound docs                   |
 | Runtime scripts | several   | Stack detector, polyglot adapter, Jinja2 doc generator, install scripts                 |

docs/architecture/REPOSITORY_STRUCTURE.md

@@ -429,6 +429,7 @@ Create `packages/<name>/` with `package.json` (`@repo/<name>`), `tsconfig.json`,
 
 - Agent → `plugins/launchpad/agents/<namespace>/<name>.md`
 - Command → `plugins/launchpad/commands/<name>.md`
+- Command template → `plugins/launchpad/templates/<command>/`
 - Preflight profile → `plugins/launchpad/preflight-profiles/<provider>.yaml` (provider profile YAMLs consumed by `lp_preflight.py`; BL-364, v2.1.7). Adding a new provider equals adding a YAML; the engine knows nothing provider-specific.
 - Skill → `plugins/launchpad/skills/<skill-name>/SKILL.md`
 - Skill references → `plugins/launchpad/skills/<skill-name>/references/` (one level deep only)

docs/guides/HOW_IT_WORKS.md

@@ -1,6 +1,6 @@
 # How It Works
 
-LaunchPad is an agentic coding harness with two main layers under the hood: a **governance kernel** plus a **plugin for Claude Code and Codex** that ride together inside any repository. The kernel is the persistent substrate (`REPOSITORY_STRUCTURE.md` whitelist, `lefthook.yml` pre-commit gates, `.launchpad/config.yml`, `.harness/` runtime, `docs/architecture/` core docs) that survives between sessions; the plugin is the 42 slash commands, 39 sub-agents, and 16 skills that operate against the kernel. The kernel lets each agent run inherit the previous run's findings instead of starting cold; the plugin gives Claude Code or Codex the verbs to brainstorm, define, plan, build, review, ship, and learn.
+LaunchPad is an agentic coding harness with two main layers under the hood: a **governance kernel** plus a **plugin for Claude Code and Codex** that ride together inside any repository. The kernel is the persistent substrate (`REPOSITORY_STRUCTURE.md` whitelist, `lefthook.yml` pre-commit gates, `.launchpad/config.yml`, `.harness/` runtime, `docs/architecture/` core docs) that survives between sessions; the plugin is the 43 slash commands, 39 sub-agents, and 16 skills that operate against the kernel. The kernel lets each agent run inherit the previous run's findings instead of starting cold; the plugin gives Claude Code or Codex the verbs to brainstorm, define, plan, build, review, ship, and learn.
 
 Brownfield projects pick up the kernel by adding the plugin to an existing repo and running `/lp-define`, while greenfield projects materialize the kernel from scratch through the four-command v2.0 pipeline (`/lp-brainstorm` → `/lp-pick-stack` → `/lp-scaffold-stack` → `/lp-define`). Both paths converge on the same operating model. For the framing of why this works, see [README.md](../../README.md). For the day-to-day pipeline below, read on.
 
@@ -453,6 +453,14 @@ Presents plan summary with hardening notes and design status. Four options:
 
 ## Key commands reference
 
+### `/lp-conductor`
+
+Use `/lp-conductor` for a change large enough to plan and build in separate sessions. One session becomes the Conductor, checking the work of an Author who plans and an Implementer who builds. It answers routine questions and brings you the decisions that need you. Both role sessions can use Claude Code or Codex.
+
+Run `/lp-define` first, then `/lp-conductor` and answer its questions. Paste the printed prompt into a new session and keep the Conductor open. If a plan is already approved or a build is already under way, say so when it asks where the work stands; it then prints the Implementer prompt, or a short prompt for the session already building, instead. It prints the Implementer prompt when the plan is ready. After the Conductor compacts, run `/lp-conductor resume` to restore its state and check for unanswered reports.
+
+You make the decisions the Conductor brings you, and every push, PR and merge still needs your own go-ahead.
+
 ### `/lp-review`
 
 Multi-agent code review with confidence-based false-positive suppression.