Implementation Plan: Add resolve-design-review Skill and Eliminate Terminal STOP Dead-End

[Trecek]

Summary

The research recipe's review_design step currently hard-routes verdict=STOP directly to design_rejected (pipeline halt), bypassing any analysis of whether the stop triggers are actually fixable. This causes unnecessary pipeline deaths when stop triggers are mechanical methodological flaws with concrete fixes (as shown in TalonT-Org/spectral-init#222).

This plan adds:

1. A new resolve-design-review skill that triages each stop-trigger finding as ADDRESSABLE, STRUCTURAL, or DISCUSS using parallel feasibility-validation subagents, then emits either resolution=revised (loop back for revision) or resolution=failed (genuinely terminal)
2. A new resolve_design_review recipe step in research.yaml that routes STOP → resolve_design_review instead of directly to design_rejected
3. A skill contract entry for resolve-design-review in skill_contracts.yaml
4. Updated tests: fix the existing STOP-routing assertion and add new tests for the step and skill

Architecture Impact

Process Flow Diagram

%%{init: {'flowchart': {'nodeSpacing': 40, 'rankSpacing': 50, 'curve': 'basis'}}}%%
flowchart TB
    %% CLASS DEFINITIONS %%
    classDef terminal fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff;
    classDef stateNode fill:#004d40,stroke:#4db6ac,stroke-width:2px,color:#fff;
    classDef handler fill:#e65100,stroke:#ffb74d,stroke-width:2px,color:#fff;
    classDef newComponent fill:#2e7d32,stroke:#81c784,stroke-width:2px,color:#fff;
    classDef detector fill:#b71c1c,stroke:#ef5350,stroke-width:2px,color:#fff;

    %% TERMINALS %%
    START([START])
    REJECTED([design_rejected<br/>action: stop])
    EXEC([create_worktree<br/>→ Execution Phase])

    subgraph DesignPhase ["Research Recipe — Design Phase"]
        direction TB
        scope["scope<br/>━━━━━━━━━━<br/>Scope research question"]
        plan["plan_experiment<br/>━━━━━━━━━━<br/>Plan experiment<br/>(receives revision_guidance)"]
        review["● review_design<br/>━━━━━━━━━━<br/>Validate plan<br/>retries: 2"]
        revise["revise_design<br/>━━━━━━━━━━<br/>Route → plan_experiment"]
        rdr["★ resolve_design_review<br/>━━━━━━━━━━<br/>Triage STOP findings<br/>retries: 1"]
        triage{"★ Triage<br/>━━━━━━━━━━<br/>Any ADDRESSABLE<br/>or DISCUSS?"}
    end

    %% FLOW %%
    START --> scope
    scope --> plan
    plan --> review
    review -->|"verdict=GO"| EXEC
    review -->|"verdict=REVISE"| revise
    revise --> plan
    review -->|"● verdict=STOP<br/>(was: design_rejected)"| rdr
    rdr --> triage
    triage -->|"resolution=revised<br/>any ADDRESSABLE/DISCUSS"| revise
    triage -->|"resolution=failed<br/>all STRUCTURAL"| REJECTED

    %% CLASS ASSIGNMENTS %%
    class START,REJECTED,EXEC terminal;
    class scope,plan handler;
    class review,revise stateNode;
    class rdr,triage newComponent;

Loading

Color Legend:

Color	Category	Description
Dark Blue	Terminal	START, design_rejected halt, create_worktree handoff
Orange	Handler	Existing processing steps (scope, plan_experiment)
Teal	State	Existing routing/decision nodes (review_design, revise_design)
Green	New Component	★ New resolve_design_review step + triage logic

Closes #605

Implementation Plan

Plan file: /home/talon/projects/autoskillit-runs/impl-20260404-132147-193877/.autoskillit/temp/make-plan/resolve_design_review_plan_2026-04-04_132804.md

🤖 Generated with Claude Code via AutoSkillit

Token Usage Summary

Step	input	output	cached	count	time
plan	39	23.7k	1.5M	1	8m 6s
verify	23	12.0k	937.2k	1	4m 21s
implement	56	16.1k	2.7M	1	7m 30s
fix	25	9.1k	879.4k	1	5m 58s
audit_impl	17	14.8k	356.6k	1	5m 57s
open_pr	24	12.9k	799.4k	1	4m 41s
Total	184	88.6k	7.2M		36m 35s

[Trecek]

AutoSkillit PR Review — Verdict: changes_requested

Found 9 actionable warnings. See inline comments.

[Trecek]

AutoSkillit review found 9 blocking issues. See inline comments.