Ralplan

Ported from oh-my-codex ralplan. OMX runtime conventions ($macro invocation, omx CLI, .omx/ state directory) are replaced with WorkBuddy idioms (Skill tool, Agent tool, task list, .workbuddy/memory).

Ralplan (Consensus Planning Alias)

Ralplan is a shorthand alias for plan --consensus. It triggers iterative planning with Planner, Architect, and Critic roles until consensus is reached, with RALPLAN-DR structured deliberation (short mode by default, deliberate mode for high-risk work). An advisory ontology reviewer (Scholastic-style) may inform the plan for ontology-heavy evidence but is not part of the durable consensus gate.

Usage

ralplan "task description"
ralplan --interactive "task description"
ralplan --deliberate "task description"

Flags

• --interactive: Enables user prompts at key decision points (draft review and final approval). Without it the workflow runs fully automated and outputs the final plan.
• --deliberate: Forces deliberate mode for high-risk work (adds pre-mortem + expanded test planning). Can also auto-enable when the request signals high risk (auth/security, migrations, destructive changes, production incidents, compliance/PII, public API breakage).

Behavior

This skill simply invokes the plan skill in consensus mode:

skill: plan   (with consensus; add --interactive / --deliberate as requested)

The consensus workflow (full detail in the plan skill):

1. Planner creates an adaptive plan (right-sized, not exactly five steps) and a compact RALPLAN-DR summary (Principles 3-5, Decision Drivers top 3, Viable Options ≥2 with bounded pros/cons; invalidation rationale if only one remains; deliberate mode adds pre-mortem + expanded test plan).
2. User feedback (--interactive only): present the draft + Principles/Drivers/Options via AskUserQuestion (Proceed to review / Request changes / Skip review). Otherwise auto-proceed.
3. Architect reviews for soundness via a separate Agent call (strongest steelman antithesis, a real tradeoff tension, synthesis); await completion before step 4.
4. Critic evaluates via a separate Agent call, only after step 3 (principle-option consistency, fair alternatives, risk clarity, testable criteria, verification steps).
5. Re-review loop (max 5): any non-APPROVE verdict re-runs Planner→Architect→Critic until APPROVE or 5 iterations; then present the best version.
6. On Critic approval (--interactive only): present approval options via AskUserQuestion (ultragoal / team / explicit ralph fallback / specialized goal-mode follow-up / Request changes / Reject). Final plan includes ADR, available-agent-types roster, staffing guidance, team launch hints, team verification path, Goal-Mode Follow-up Suggestions. Otherwise output the final plan and stop.
7. (--interactive only) On approval: invoke ultragoal (default), team, the selected specialized goal-mode follow-up (autoresearch-goal / performance-goal), or ralph only when explicitly selected — never implement directly.

Steps 3 and 4 MUST run sequentially as role-specific Agent calls. Always await the Architect result before the Critic call.

Planning / Execution Boundary

Ralplan is a planning mode. While active with no explicit execution handoff, implementation write tools are out of scope. It may inspect the repo and write only planning artifacts (.omw/context/, .omw/plan/, .omw/specs/, task-list/state records).

Canonical flow: ralplan -> durable consensus artifact -> explicit execution lane -> ultragoal | team | ralph.

Durable Consensus Handoff Contract

Ralplan is not complete merely because .omw/plan/prd-*.md and .omw/plan/test-spec-*.md exist — those are planning artifacts, not consensus evidence. Before any execution handoff, persist a durable handoff record distinguishing:

• planning_artifacts: PRD/test-spec paths.
• ralplan_architect_review: completed Architect review with approving verdict.
• ralplan_critic_review: completed Critic review with approving verdict, recorded only after Architect.
• ralplan_consensus_gate.complete: true only when both reviews are present, approving, and in Architect→Critic order.

If Architect is missing/blocked, stay in Architect review or report the blocker. If Critic is missing/blocked/non-approving, stay in Critic/re-review or report the max-iteration outcome.

Goal-Mode Follow-up Suggestions

When ralplan outputs a handoff, include:

• ultragoal — default goal-mode follow-up for durable sequential completion tracking.
• autoresearch-goal — research projects with a research deliverable/evaluator.
• performance-goal — optimization/performance work with measurable criteria.
• team — first-class parallel execution option.
• ralph — explicit fallback only when persistent single-owner verification is intentionally selected.

For parallelizable durable-goal delivery, recommend ultragoal + team together.

Pre-context Intake

Before consensus planning or execution handoff, ensure a grounded context snapshot exists at .omw/ralph/context-{slug}-{timestamp}.md (task statement, outcome, known facts, constraints, unknowns, touchpoints). Reuse the latest relevant snapshot when available. If ambiguity is high, gather brownfield facts via repository inspection, then run deep-interview --quick <task>. For external/official docs, invoke best-practice-research before finalizing the handoff. Do not hand off to execution until intake is complete; if urgency forces progress, document risk tradeoffs explicitly.

Pre-Execution Gate

Execution modes spin up heavy multi-agent orchestration and waste cycles on vague requests. Ralplan-first intercepts underspecified execution requests (e.g. "ralph add authentication") and redirects them through consensus planning. The gate auto-passes when any concrete signal is present (file path, issue/PR number, camelCase/snake_case symbol, test runner, numbered steps, acceptance criteria, error reference, code block, or an explicit force:/! override). One signal is enough.