ADR-0029: Planning interaction contract and phase legibility

[4lgn]

Implements ADR-0029, a comprehensive refinement of the planning workflow based on field experience. This change introduces an interaction contract that budgets user decisions, establishes build-safety as the stopping rule (not frontier exhaustion), and adds a legibility layer presenting the thirteen internal stages as six user-facing phases.

Summary

The first full run of the Launchrail rail revealed that planning could become exhausting and procedural rather than decision-focused. This change addresses that by:

1. Splitting decision ownership: Users own product promises, priorities, and irreversible tradeoffs; the agent owns reversible implementation details (marked Provisional).
2. Labeling every uncertainty: Each question is tagged decide-now, agent-default, research, prototype, or defer — only decide-now questions become user interviews.
3. Budgeting user attention: Sessions target ~6 user decisions with rounds capped at 3 questions; checkpoints every 2 rounds let users steer the process.
4. Stopping at build-safety: Grills end when the next slice can be built safely, not when every question is answered. Foundation grills open with a "risk cut" of ~5 kill-or-reshape assumptions.
5. Legibility layer: The thirteen stages are presented as six phases (Intent, Exploration, Decisions, Blueprint, Build, Ship) with an explicit rail banner at every transition.

Key Changes

• New ADR-0029 (docs/adr/0029-planning-interaction-contract.md): Documents the interaction contract, decision labeling, session budgets, and the phase view.
• Updated workflow.md (both locations): Adds the phase view section, rail banner format, and session summary structure; clarifies that the document now carries four contracts (phase view, stage table, interaction contract, conductor rules).
• Updated launch-grill/SKILL.md: Reframes the grill's purpose from "relentless" to "budgeted"; emphasizes build-safety over exhaustion; adds the risk cut for foundation grills; introduces the triage table (decide-now / agent-default / research / prototype / defer); documents approved prototypes' authority.
• Updated launch-wayfinder/SKILL.md: Strengthens "planning only" — execution is removed from the map; tickets must name the decision they unblock; maps don't nest; inherited grills use a tight budget.
• Updated launch-spec/SKILL.md: Synthesis preserves grill labels (Locked / Provisional / Deferred).
• Updated launch-tickets/SKILL.md: Adds rail banner close-out.
• Updated launch-ralph/SKILL.md: Introduces output budget — two reports per run (pre-launch echo and close-out recap), everything else silent.
• Updated launch/SKILL.md and conductor: Reports position as phases with explicit rail banner at transitions.
• Updated launch-project-alignment/SKILL.md: Minor clarifications.
• Updated issue-tracker.md: Documents spec milestone as a rollup view.
• Version bump: .launchrail-lock.json updated to v1.12.0; all managed skill checksums updated.

Implementation Details

• The phase view is a presentation layer only — stages remain the normative contract for ownership, artifacts, and gates.
• The rail banner is a fixed fenced block rendered at every transition, never loose prose, ensuring consistent legibility.
• Session summaries contain exactly four blocks: Locked, Provisional, Deferred, Next command — no other prose.
• The risk cut for foundation grills focuses on ~5 assumptions that could kill or reshape the product; the long tail defaults or defers.
• Approved prototypes and handoff packages are treated as decision records with authority — proposed cuts require concrete safety or cost reasons.

https://claude.ai/code/session_011aDBhRGtVGr8GcbJbfRHoE

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}