20_agentflow

Multi-expert AI/ML specification review with scoring gate — model architecture, evaluation rigor, safety/alignment, and production readiness for AI systems and LLM apps

Multi-expert architecture review — boundaries, integration patterns, failure modes, evolvability

Multi-expert business analysis with advisory recommendations (no scoring gate)

Multi-expert intelligence pipeline review — discovery quality, ingestion resilience, scoring validity, platform compliance, taxonomy coherence, cost efficiency

Multi-expert mobile/native app specification review with scoring gate — Android, iOS, Swift, SwiftUI

Multi-expert personal-development review with scoring gate — learnability, adoption, human-centeredness, and capability impact for talent/learning/AI-augmentation designs

Multi-expert research and discovery panel with source evaluation, API feasibility, collection strategy, and intelligence gap analysis. Use when evaluating data sources, APIs, scraping strategies, OSINT collection plans, or any research/discovery effort that needs expert validation before implementation.

Multi-expert security review — threat modeling, secrets handling, LLM/prompt injection, data classification, incident response

Multi-expert specification review with scoring gate and quality assessment

Multi-expert test-strategy review — coverage, fixture quality, AC testability, regression discipline

4-reviewer parallel code review — disjoint lenses (correctness, concurrency, contracts-vs-callers, test-adequacy) — FIPD triage with gated auto-apply for verified + mechanical Fix-* findings

Code quality, security, and performance scan with FIPD-classified findings

Execute an implementation plan with review checkpoints and verification gates

List all 34 Shepherd commands organized by category

Break tasks into independent subagents for parallel execution when no shared state or sequential dependencies exist

Build clickable wireframe prototypes in code to validate flow and structure before hi-fi design

Evidence-based verification before claiming work is complete -- run commands and confirm output before any success claims

Create isolated git worktree workspaces for feature branches with safety verification

Complete development work -- verify tests, present merge/PR/cleanup options, handle branch finishing

Quality-gate Stage-2 code review. Reviews a diff across three lenses (correctness bugs, CLAUDE.md adherence, missed simplifications) in one pass and emits a machine-readable REVIEW-VERDICT line. Not for interactive use — invoked by quality_gate.run_stage2_review via claude -p. Does NOT invoke the /code-review plugin.

Run the Definition of Done gate check before deployment. Use when verifying work is complete, before committing, or when checking if a task meets exit criteria.

Run the Definition of Done gate check before deployment. Verifies work is complete, tests pass, and task meets all exit criteria before committing.

Start the Shepherd autonomous execution loop. Reads tasks from TODO-Today queue, executes them with quality gates, and processes until queue is empty or paused.

Requirements discovery through Socratic dialogue. Explores intent, constraints, and design before implementation.

Cancel active Ralph loop and report iteration count

Structured root-cause diagnosis before proposing fixes

Targeted documentation for components, functions, and APIs

Run the Definition of Ready gate check before implementing a task. Verifies a task is ready for implementation including spec-panel score >= 7.0 requirement.

Development effort estimation with T-shirt sizing and risk adjustment

Educational code explanations with adaptive depth and insight highlights

Save session context to HANDOVER.md for seamless continuation. Captures last task, queue position, git state, decisions, and resume checklist.

Generate compressed project map for token-efficient codebase understanding

Daily session startup scan — assess pipeline state and determine next actions.

Execute the Shepherd 14-step Scrumban inner loop for autonomous task processing. Use when starting autopilot mode, processing TODO-Today queue items, or running the autonomous dev loop.

Sync all agentflow skills to global ~/.claude/commands/ so they work in every project. Run after adding new skills.

Create an implementation plan from a spec or requirements before touching code

Start persistent Ralph loop for iterative task completion

Deep web research with parallel search, source validation, and evidence-based findings

Run a retrospective after every 10 completed User Stories. Reflects on recent work, identifies patterns, and generates actionable improvement items.

Request self-review of completed work and handle external code review feedback with technical rigor