bigpowers 2.87.0 → 2.87.2

package/GSD_WORKFLOW_ANALYSIS_VS_BIGPOWERS.md

@@ -0,0 +1,1015 @@
+# GSD Workflow Analysis vs. Bigpowers: Complete Mapping & Gaps
+
+**Date:** 2026-05-18  
+**Scope:** Comprehensive analysis of GSD's 6-command core loop, each step's practices, and gaps vs. Bigpowers
+
+---
+
+## Executive Summary
+
+**GSD (Get Shit Done)** is a **meta-prompting + multi-agent orchestration framework** with:
+- 6 core commands (new-project → discuss → plan → execute → verify → ship)
+- 31+ specialized agents (researchers, planners, checkers, executors, verifiers)
+- 93+ total workflows + 60+ references
+- Context isolation (fresh 200K window per agent)
+- Wave-based parallel execution
+- Supply-chain security gates (package legitimacy)
+
+**Bigpowers** is a **skill-centric lifecycle framework** with:
+- 38 skills (verb-noun pairs)
+- Inline prompts in SKILL.md
+- BMAD phases (Discover → Elaborate → Plan → Build → Sustain)
+- Ad-hoc skill composition (no formal orchestration)
+- Manual subagent dispatch via `delegate-task` / `dispatch-agents`
+
+**Key insight:** GSD is **orchestration-first** (workflow → agents); Bigpowers is **skill-first** (agent picks skill). They solve the same problem from different angles.
+
+---
+
+## 1. The GSD Core Loop (6 Commands)
+
+### Step 1: `/gsd-new-project` — Project Initialization
+
+**Purpose:** Transform vague idea → concrete project definition (PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md)
+
+**Workflow:**
+```
+Input (idea)
+  ↓
+Questioning (domain-probes.md philosophy)
+  ↓
+4× Parallel Researchers (Stack, Features, Architecture, Pitfalls)
+  ↓
+Research Synthesizer
+  ↓
+Requirements extraction
+  ↓
+Roadmapper (creates ROADMAP.md with phases, success criteria, plans)
+  ↓
+User approval
+  ↓
+STATE.md initialized
+```
+
+**Key Practices:**
+
+1. **Questioning Philosophy:**
+   - User is the visionary; Claude is the builder
+   - Extract VISION, not implementation
+   - Avoid generic categories; generate domain-specific questions
+   - Scope is locked (no scope creep during discussion)
+
+2. **Parallel Research:**
+   - 4 independent researchers spawn simultaneously (Stack, Features, Architecture, Pitfalls)
+   - Fresh context window per agent
+   - Results feed into synthesis (no early convergence)
+
+3. **Synthesizer Role:**
+   - Unifies 4 parallel outputs into SUMMARY.md
+   - Identifies gaps, contradictions, consensus
+   - Produces requirements extraction input
+
+4. **ROADMAP Artifact:**
+   - Phases with dependencies (Phase 2 depends on Phase 1)
+   - Decimal sub-phases for urgent insertions (2.1 inserted between 2 and 3)
+   - **Success Criteria:** Observable behaviors (user-perspective) per phase
+   - Plan count estimation per phase
+   - Progress table
+   - Wave analysis for execution planning
+
+5. **Auto Mode (`--auto @file.md`):**
+   - Skip deep questioning if PRD/idea doc provided
+   - Extract context automatically
+   - Still run research and synthesis
+   - Useful for brownfield projects or time-constrained work
+
+**Artifacts Created:**
+- `PROJECT.md` — Vision, Core Value, Requirements (Validated/Active/Out of Scope), Constraints, Key Decisions
+- `REQUIREMENTS.md` — Scoped, traceable (REQ-01, REQ-02, ...)
+- `ROADMAP.md` — Phases with success criteria and plans
+- `STATE.md` — Current position, performance metrics (< 100 lines)
+- `research/` — STACK.md, FEATURES.md, ARCHITECTURE.md, PITFALLS.md, SUMMARY.md
+- `config.json` — Workflow config (mode, model profiles, parallelization)
+- `CLAUDE.md` — Project-level Claude Code integration (optional)
+
+---
+
+### Step 2: `/gsd-discuss-phase [N]` — Phase Context Gathering
+
+**Purpose:** Capture implementation decisions before planning (CONTEXT.md)
+
+**Workflow:**
+```
+Read ROADMAP.md phase description
+  ↓
+Identify gray areas (ambiguities, design choices)
+  ↓
+Present gray areas to user (or batch/auto)
+  ↓
+Deep-dive selected areas → CONTEXT.md
+  ↓
+Git commit discussion log
+```
+
+**Key Practices:**
+
+1. **Gray Area Identification:**
+   - Not generic categories (UI/UX/Behavior)
+   - Phase-specific decisions (e.g., for auth: session handling, error responses, multi-device policy, recovery flow)
+   - Driven by domain understanding (what users SEE / CALL / RUN / READ)
+
+2. **Mode Options:**
+   - `--all`: Discuss all gray areas without selection step
+   - `--auto`: Auto-select recommended defaults
+   - `--batch`: Group questions for batch intake
+   - `--analyze`: Add trade-off analysis during discussion
+   - `--power`: File-based bulk question answering
+   - `--assumptions`: Surface Claude's assumptions without interactive session
+   - Default: Interactive one-by-one
+
+3. **Scope Guardrails (CRITICAL):**
+   - Phase boundary is FIXED (from ROADMAP.md)
+   - Allowed: Clarifying ambiguity ("How should posts be displayed?")
+   - Not allowed: Scope creep ("Should we add comments?")
+   - Heuristic: Does this clarify how we implement what's scoped, or add a new capability?
+   - When creep detected: Note in "Deferred Ideas", don't act on it
+
+4. **Output Structure:**
+   - `{phase}-CONTEXT.md` — Decisions, locked choices, design contracts
+   - `{phase}-DISCUSSION-LOG.md` — Audit trail of what was discussed
+   - Questions are validated: empty responses are rejected
+
+**Key Reference:** `domain-probes.md` — Phase-specific probe questions per domain type
+
+---
+
+### Step 3: `/gsd-plan-phase [N]` — Research, Plan, Verify
+
+**Purpose:** Create executable PLAN.md files with integrated research and verification
+
+**Workflow:**
+```
+Research Gate (if needed)
+  ↓
+Phase Researcher → RESEARCH.md
+  │  ├── Package Legitimacy Audit (slopcheck)
+  │  │   ├── [SLOP] = remove entirely
+  │  │   ├── [SUS] = checkpoint before install
+  │  │   └── [OK] = approved
+  │  └── Ecosystem research, tech stack validation
+  ↓
+Planner → PLAN.md files (01-01, 01-02, ...)
+  │  ├── Every task has `verify: <runnable command>`
+  │  ├── Dependencies declared (Plan 02 depends on 01)
+  │  ├── Checkpoints for [ASSUMED]/[SUS] packages
+  │  └── STRIDE threat model for supply chain
+  ↓
+Plan Checker → Verify loop (max 3 iterations)
+  │  ├── Check plans are executable
+  │  ├── Check requirements coverage (REQ-IDs)
+  │  ├── Check decision coverage (CONTEXT.md)
+  │  └── If issues: replan until clean
+  ↓
+STATE.md → "Planned" status
+```
+
+**Key Practices:**
+
+1. **Package Legitimacy Gate (v1.42.1):**
+   - Researcher runs `slopcheck install <pkg> --json` on each package
+   - Writes `## Package Legitimacy Audit` table to RESEARCH.md
+   - Verdicts: `[SLOP]` (remove), `[SUS]` (checkpoint), `[OK]` (approved)
+   - Packages from WebSearch are tagged `[ASSUMED]` (not `[VERIFIED]`)
+   - Planner inserts `checkpoint:human-verify` before `[ASSUMED]`/`[SUS]` installs
+   - **Supply-chain attack prevention:** Blocks hallucinated package names pre-registered with malicious post-install scripts
+
+2. **Plan Task Structure:**
+   - Every task has: title, description, acceptance criteria, `verify: <runnable command>`
+   - Dependencies: "Plan 03 depends on 01" → affects wave grouping
+   - Checkpoints: human-verify gates before risky operations
+
+3. **Plan Checker Verification Loop:**
+   - Max 3 iterations (prevents infinite loops)
+   - Gates:
+     - **Quality Gate:** Plan clarity, completeness
+     - **Requirements Coverage Gate:** Every REQ-ID is addressed
+     - **Decision Coverage Gate:** Every decision in CONTEXT.md is implemented
+     - **Safety Gate:** No obvious pitfalls or security issues
+   - If any gate fails: planner revises until clean
+
+4. **TDD Mode (`workflow.tdd_mode: true`):**
+   - Planner applies `type: tdd` to eligible tasks
+   - Test-first approach integrated into planning
+   - References `tdd.md` for integration patterns
+
+5. **Adaptive Context (1M models):**
+   - When context window >= 500K (Opus 4.6, Sonnet 4.6)
+   - Planner receives prior phase CONTEXT.md + SUMMARY.md + explicit dependencies
+   - Enables cross-phase awareness within a phase
+
+6. **Flags:**
+   - `--skip-research`: Skip research if domain is familiar
+   - `--auto`: Non-interactive planning
+   - `--validate`: Validate state before planning
+   - `--bounce`: External plan bounce validation (custom script)
+   - `--ingest <ADR-file>`: Use ADR instead of discuss-phase
+   - `--reviews`: Replan with cross-AI review feedback
+
+**Artifacts Created:**
+- `{phase}-RESEARCH.md` — Ecosystem research, package legitimacy audit, technical findings
+- `{phase}-{N}-PLAN.md` — Executable plan files (01-01, 01-02, ...)
+- `{phase}-VALIDATION.md` — Verification coverage (Nyquist sampling)
+
+---
+
+### Step 4: `/gsd-execute-phase [N]` — Wave-Based Parallel Execution
+
+**Purpose:** Execute all plans in parallel waves (fresh context per executor)
+
+**Workflow:**
+```
+Parse args + initialize context
+  ↓
+Analyze dependencies → Wave grouping
+  ├── Plan 01 (no deps) ─┐
+  ├── Plan 02 (no deps) ─┤ Wave 1 (parallel)
+  ├── Plan 03 (dep: 01) ─┤ Wave 2 (waits for Wave 1)
+  └── Plan 04 (dep: 02) ─┘
+  ↓
+Per-wave: Spawn executors in parallel
+  │  ├── Fresh 200K context window per executor
+  │  ├── Project context (PROJECT.md, STATE.md)
+  │  ├── Phase context (CONTEXT.md, RESEARCH.md)
+  │  ├── Specific PLAN.md to execute
+  │  ├── Commit atomically per task (conventional commits)
+  │  └── Create SUMMARY.md per plan
+  ↓
+Post-wave: Run pre-commit hooks once (not per-agent)
+  ↓
+All waves complete → Verifier
+```
+
+**Key Practices:**
+
+1. **Wave Analysis:**
+   - Grouping independent plans for parallel execution
+   - Dependency graph: Plan B cannot start until Plan A commits
+   - If 10 plans with no deps: all run in Wave 1 simultaneously
+   - Adaptive: can execute specific wave with `--wave N`
+
+2. **Executor Agent:**
+   - Fresh 200K context (or adaptive up to 1M)
+   - Receives: PLAN.md, project context, phase context
+   - Creates code + SUMMARY.md per plan
+   - Commits atomically per task (conventional commits: feat, fix, refactor, docs)
+   - Never blocks — next wave starts immediately after prior wave completes
+
+3. **Parallel Safety Mechanisms:**
+   - **--no-verify commits:** Executors skip pre-commit hooks (prevents lock contention in Rust projects)
+   - **Orchestrator runs hooks once per wave:** After all parallel tasks complete
+   - **STATE.md locking:** Atomic creation (O_EXCL) + stale lock detection (10s timeout) + spin-wait with jitter
+   - Prevents read-modify-write race condition
+
+4. **Context Enrichment (1M Models):**
+   - Executor receives all prior wave SUMMARY.md + phase CONTEXT.md
+   - Enables cross-plan awareness (Plan 03 knows what Plan 01-02 shipped)
+   - Standard 200K windows get truncated prompts (cache-friendly ordering)
+
+5. **Checkpoint Handling:**
+   - If a task has `checkpoint:human-verify`: pause execution, ask user
+   - User approves or rejects (blocks proceeding until resolved)
+
+6. **Post-Execution Gates:**
+   - **Schema Drift Gate:** Compare `.planning/codebase/STRUCTURE.md` (last mapped commit) vs HEAD
+     - Detects new directories, barrel exports, migrations, route modules
+     - Warns or auto-remaps based on `workflow.drift_action`
+   - **Decision Coverage Gate (non-blocking):** CONTEXT.md decisions → shipped artifacts
+
+**Artifacts Created:**
+- `{phase}-{N}-SUMMARY.md` — Per-plan execution outcome, what was shipped
+- `.planning/codebase/STRUCTURE.md` updated (drift detection)
+- Git commits (one per task, conventional format)
+
+---
+
+### Step 5: `/gsd-verify-work [N]` — Post-Execution Verification
+
+**Purpose:** Walk through what was built, diagnose failures, generate fix plans
+
+**Workflow:**
+```
+User acceptance testing (manual)
+  ├── Walk through phase features
+  ├── Test acceptance criteria from ROADMAP.md success criteria
+  ├── Note issues found
+  ↓
+Verifier agent
+  ├── Reads all PLAN.md, SUMMARY.md, REQUIREMENTS.md
+  ├── Checks against phase goals
+  ├── Detects failures with dedicated debug agents
+  └── Creates VERIFICATION.md with:
+      ├── status: (passed | needs-review | needs-repair)
+      ├── Results per success criterion
+      └── Fix plans (if needed)
+  ↓
+If repairs needed:
+  └── `/gsd-execute-phase --gaps-only` → Re-execute only gap-closure plans
+```
+
+**Key Practices:**
+
+1. **UAT (User Acceptance Testing):**
+   - Manual walk-through by user
+   - Test against phase success criteria (from ROADMAP.md)
+   - Document issues as you find them
+
+2. **Verifier Agent:**
+   - Independent review of execution
+   - Cross-checks PLAN.md against SUMMARY.md and shipped code
+   - Identifies gaps or misalignments
+   - Generates fix plans if needed (not imperative — just plans)
+
+3. **Verification Status:**
+   - `passed` → Phase is complete, move to next
+   - `needs-review` → Borderline; requires second opinion
+   - `needs-repair` → Failures detected; fix plans generated
+
+4. **Gap Closure:**
+   - If `needs-repair`: Run `/gsd-plan-phase --gaps` (replanning with VERIFICATION.md as input)
+   - Then `/gsd-execute-phase --gaps-only` (execute only the closure plans)
+   - Verify again
+
+**Artifacts Created:**
+- `{phase}-VERIFICATION.md` — Status, results, fix plans (if needed)
+- `{phase}-UAT.md` — UAT session notes (optional user-written)
+
+---
+
+### Step 6: `/gsd-ship [N]` — Release & Complete Milestone
+
+**Purpose:** Create PR, merge, archive completed phase, tag release
+
+**Workflow:**
+```
+Coverage check (>= 80% overall, >= 95% business logic)
+  ↓
+Git branching strategy:
+  ├── If interactive: user chooses (merge/keep/discard)
+  ├── If auto: merge if all checks pass
+  ↓
+Create PR (gh pr create)
+  ├── Links PLAN.md, SUMMARY.md, VERIFICATION.md
+  ├── Auto-suggests reviewers (if enabled)
+  ├── Conventional title + body
+  ↓
+Merge to main
+  ↓
+Git cleanup (worktree removal if isolation used)
+  ↓
+If phase is last in milestone:
+  ├── Archive all phase artifacts
+  ├── Move to `<details>` in ROADMAP.md
+  ├── Create git tag (vX.Y.Z)
+  └── Update MILESTONES.md
+```
+
+**Key Practices:**
+
+1. **Coverage Gates:**
+   - Overall coverage must be >= 80%
+   - Business logic coverage must be >= 95%
+   - Blocks merge if gates fail
+
+2. **PR Quality:**
+   - Conventional Commits title (feat, fix, refactor, docs)
+   - Linked artifacts (test results, deployment evidence)
+   - Auto-linked to REQUIREMENTS.md (REQ-IDs mentioned)
+
+3. **Worktree Cleanup:**
+   - If `workflow.use_worktrees: true`: removes git worktree
+   - Returns to main checkout
+   - State clean for next phase
+
+4. **Milestone Completion:**
+   - Last phase in milestone triggers archival
+   - All phase artifacts collapsed in ROADMAP.md
+   - Tag created (v1.0, v1.1, etc.)
+   - MILESTONES.md updated (shipped date, summary)
+
+---
+
+## 2. GSD Agent Taxonomy (31 Agents)
+
+GSD has **31 specialized agents**, organized by spawn category:
+
+### Researchers (4 parallel agents)
+- `gsd-project-researcher` — Project-level decisions (stack, frameworks, patterns)
+- `gsd-phase-researcher` — Phase-specific ecosystem research
+- `gsd-ui-researcher` — UI/UX research for design phases
+- `gsd-advisor-researcher` — User-profiling advisor (spawned during discuss-phase if profile exists)
+
+### Synthesizers
+- `gsd-research-synthesizer` — Unifies 4 parallel research outputs
+
+### Planners
+- `gsd-planner` — Creates detailed PLAN.md files
+- `gsd-roadmapper` — Creates ROADMAP.md from requirements
+
+### Checkers (sequential verification loop, max 3 iterations)
+- `gsd-plan-checker` — Reviews plans for quality, completeness, gate coverage
+- `gsd-integration-checker` — Cross-phase integration validation
+- `gsd-ui-checker` — UI implementation quality review
+- `gsd-nyquist-auditor` — Verification coverage validation (Nyquist sampling)
+
+### Executors
+- `gsd-executor` — Executes plan tasks, commits, creates SUMMARY.md (parallel within waves)
+
+### Verifiers
+- `gsd-verifier` — Post-execution verification, identifies gaps
+
+### Mappers
+- `gsd-codebase-mapper` — Analyzes codebase, produces `.planning/codebase/*.md` (4 parallel: tech, arch, quality, concerns)
+
+### Debuggers & Auditors
+- `gsd-debugger` — Interactive debugging for live issues
+- `gsd-ui-auditor` — Visual audit and accessibility checks
+- `gsd-security-auditor` — Security-specific audit
+
+### Doc Writers
+- `gsd-doc-writer` — Documentation generation
+- `gsd-doc-verifier` — Documentation validation
+
+### Analyzers & Profilers
+- `gsd-assumptions-analyzer` — Surfaces implementation assumptions
+- `gsd-user-profiler` — User behavioral profiling
+
+### Specialized
+- `gsd-pattern-mapper` — Detects codebase patterns
+- `gsd-code-reviewer` — Code quality review
+- `gsd-code-fixer` — Fixes code issues
+- `gsd-ai-researcher` — AI-specific research
+- `gsd-domain-researcher` — Domain-specific research
+- `gsd-eval-planner` — Evaluation planning
+- `gsd-eval-auditor` — Evaluation audit
+- `gsd-framework-selector` — Framework selection
+- `gsd-debug-session-manager` — Debug session tracking
+- `gsd-intel-updater` — Intelligence updates
+
+**Agent Lifecycle:**
+1. Orchestrator reads `agents/*.md` (YAML frontmatter: name, description, tools, color)
+2. Model resolution: `gsd-sdk query resolve-model <agent-name>` → returns opus|sonnet|haiku|inherit
+3. Spawn: `Agent(subagent_type="gsd-executor", prompt=..., model=resolved_model)`
+4. Fresh context: 200K tokens (or 500K+ for 1M-class models with context_window config)
+5. Tools available: Read, Write, Edit, Bash, Grep, Glob, WebSearch, etc. (agent frontmatter declares)
+
+---
+
+## 3. GSD References (60+ Shared Knowledge Docs)
+
+GSD has segregated technical knowledge into `get-shit-done/references/*.md` (workflows and agents reference these, not inline):
+
+**Core:**
+- `checkpoints.md` — Checkpoint types (human-verify, integration, safety)
+- `gates.md` — 4 canonical gate types (Confirm, Quality, Safety, Transition)
+- `model-profiles.md` — Model tier assignments per agent
+- `verification-patterns.md` — How to verify different artifact types
+- `planning-config.md` — Full config.json schema
+- `git-integration.md` — Commit and branching patterns
+- `questioning.md` — Dream extraction philosophy (how to elicit vision)
+- `tdd.md` — TDD integration patterns
+
+**Thinking Model References:**
+- `thinking-models-planning.md` — Thinking class model usage for planning
+- `thinking-models-execution.md` — Thinking class usage for execution
+- `thinking-models-debug.md` — Thinking class usage for debugging
+- `thinking-models-verification.md` — Thinking class usage for verification
+- `thinking-models-research.md` — Thinking class usage for research
+
+**Workflow Mechanics:**
+- `agent-contracts.md` — Formal interface between orchestrators and agents
+- `context-budget.md` — Context window budget allocation
+- `revision-loop.md` — Plan revision iteration patterns
+- `universal-anti-patterns.md` — Common anti-patterns to detect
+- `gate-prompts.md` — Gate/checkpoint prompt templates
+- `domain-probes.md` — Domain-specific probing questions
+- `artifact-types.md` — Planning artifact definitions
+- `phase-argument-parsing.md` — Phase number parsing conventions
+
+**Security:**
+- `slopcheck` integration for package legitimacy (see ARCHITECTURE.md)
+- Prompt injection detection (`gsd-prompt-guard.js`)
+- Supply-chain threat modeling (STRIDE in plans)
+
+---
+
+## 4. Bigpowers Skill Inventory (38 Skills)
+
+For reference, Bigpowers skills mapped by BMAD phase:
+
+| Phase | Skill | Equivalent in GSD |
+|-------|-------|-------------------|
+| **Discover** | survey-context | (no 1:1; orchestrator reads STATE.md) |
+| | elaborate-spec | `/gsd-discuss-phase` (gate discussions) |
+| **Elaborate** | model-domain | `/gsd-new-project` → PROJECT.md |
+| | define-language | (embedded in PROJECT.md / REQUIREMENTS.md) |
+| | challenge-design | (embedded in `/gsd-discuss-phase`) |
+| | grill-with-docs | `gsd-ui-researcher` + doc grilling |
+| | deepen-architecture | `gsd-codebase-mapper` (4 parallel) |
+| **Plan** | scope-work | `/gsd-new-project` → SCOPE (via REQUIREMENTS) |
+| | slice-tasks | `/gsd-plan-phase` → PLAN.md files |
+| | define-success | ROADMAP.md success criteria |
+| | plan-work | `/gsd-plan-phase` → PLAN.md with verify: |
+| | plan-refactor | (no explicit skill; embedded in plan-phase) |
+| **Spike** | spike-prototype | (embedded in `/gsd-new-project` research phase) |
+| **Build** | kickoff-branch | `/gsd-execute-phase` → git branch |
+| | guard-git | (no 1:1; GSD uses git hooks) |
+| | hook-commits | (GSD has pre-commit integration) |
+| | develop-tdd | `workflow.tdd_mode: true` in plan-phase |
+| | delegate-task | (no 1:1; orchestrators are thin) |
+| | dispatch-agents | `/gsd-execute-phase` (wave-based parallel) |
+| | execute-plan | `/gsd-execute-phase` → `gsd-executor` agents |
+| | wire-observability | (embedded in executor + status-line hooks) |
+| **Bug** | investigate-bug | `/gsd-debug` workflow |
+| | diagnose-root | `gsd-debugger` agent |
+| | validate-fix | (embedded in verify-phase) |
+| **Review** | audit-code | (embedded in plan-checker + execution) |
+| | request-review | `/gsd-code-review` workflow |
+| | respond-review | (embedded in plan-phase --reviews) |
+| **Integrate** | commit-message | Conventional Commits (automated in executors) |
+| | release-branch | `/gsd-ship` workflow |
+| **Sustain** | inspect-quality | `/gsd-verify-work` → `gsd-verifier` |
+| | organize-workspace | `/gsd-workspace` for multi-repo isolation |
+| **Utility** | terse-mode | (no 1:1; token budgets prevent need) |
+| | craft-skill | (no 1:1; GSD doesn't support skill authoring) |
+| | edit-document | (embedded in orchestrators, not a skill) |
+| **Bootstrap** | using-bigpowers | `/gsd-new-project` (intro to GSD) |
+| | seed-conventions | (CLAUDE.md generated during new-project) |
+
+---
+
+## 5. Gap Analysis: Bigpowers vs. GSD
+
+### 5.1 GSD Has (Bigpowers Lacks)
+
+#### 1. **Formal Orchestration Layer**
+- **GSD:** Workflows are explicit state machines with gate definitions
+- **Bigpowers:** Orchestration is implicit in skill selection flow
+- **Gap:** Bigpowers skills don't declare dependencies or gates; orchestration is manual
+
+#### 2. **Parallel Agent Spawning with Isolation**
+- **GSD:** `execute-phase` spawns N executors with fresh 200K context each, parallel within waves
+- **Bigpowers:** `dispatch-agents` and `delegate-task` exist, but orchestration is left to the user
+- **Gap:** No built-in wave analysis or dependency management
+
+#### 3. **Supply-Chain Security Gates**
+- **GSD:** Package Legitimacy Audit (slopcheck) + [SLOP]/[SUS]/[OK] verdicts
+- **Bigpowers:** No security gate for package recommendations
+- **Gap:** Hallucinated package names can slip through
+
+#### 4. **Success Criteria in Roadmap**
+- **GSD:** ROADMAP.md has observable success criteria per phase
+- **Bigpowers:** RELEASE-PLAN.md has WSJF and focus area, but no observable criteria
+- **Gap:** Verification success is not pre-defined; harder to know when a phase is done
+
+#### 5. **Comprehensive Reference Library**
+- **GSD:** 60+ segregated references (gates, patterns, thinking models, domain probes)
+- **Bigpowers:** Practices embedded in CONVENTIONS.md + inline SKILL.md comments
+- **Gap:** Knowledge is distributed; harder to find and reuse
+
+#### 6. **Context Propagation Model**
+- **GSD:** Explicit document flow (PROJECT.md → ROADMAP.md → CONTEXT.md → RESEARCH.md → PLAN.md → VERIFICATION.md)
+- **Bigpowers:** Ad-hoc file creation (specs/*.md) with no declared propagation
+- **Gap:** Agents don't know which docs to read in which order
+
+#### 7. **Model Profiles & Adaptive Context**
+- **GSD:** Model assignment per agent (opus for planning, sonnet for research) + 1M context enrichment
+- **Bigpowers:** Model is inherited from session; no per-skill assignment
+- **Gap:** No cost-aware model routing; no adaptive context for large models
+
+#### 8. **Package Discovery & Legitimacy**
+- **GSD:** Slopcheck integration with verdicts
+- **Bigpowers:** Manual package selection in plan
+- **Gap:** No automated legitimacy checking
+
+#### 9. **Assumption Surfacing**
+- **GSD:** `/gsd-discuss-phase --assumptions` surfaces Claude's assumptions about the phase
+- **Bigpowers:** No equivalent; assumptions are implicit
+- **Gap:** User doesn't see what Claude thinks before planning begins
+
+#### 10. **Automated Code Analysis**
+- **GSD:** `gsd-codebase-mapper` (4 parallel agents) produce `.planning/codebase/*.md`
+- **Bigpowers:** `survey-context` is manual or light
+- **Gap:** No systematic codebase analysis before planning starts
+
+---
+
+### 5.2 Bigpowers Has (GSD Lacks)
+
+#### 1. **Verb-Noun Skill Names**
+- **Bigpowers:** `develop-tdd`, `validate-fix`, `challenge-design` (Uncle Bob naming)
+- **GSD:** Commands are `/gsd-command-name` (hyphenated, not verb-noun consistent)
+- **Win for Bigpowers:** Names are more discoverable and memorable
+
+#### 2. **BMAD Phase Clarity**
+- **Bigpowers:** Discover → Elaborate → Plan → Build → Sustain (explicit phase model)
+- **GSD:** Phases are numbered (1, 2, 3, ...) without semantic labeling
+- **Win for Bigpowers:** Phases are self-explanatory
+
+#### 3. **Explicit Verification Commands**
+- **Bigpowers:** `validate-fix` skill with runnable verify steps
+- **GSD:** Embedded in `gsd-verifier` agent
+- **Win for Bigpowers:** Verification is a first-class skill, not an embedded step
+
+#### 4. **Skill-Driven Composition**
+- **Bigpowers:** User picks skills in any order (flexible)
+- **GSD:** Workflows are prescriptive (new-project → discuss → plan → execute → verify → ship)
+- **Win for Bigpowers:** More flexible for non-standard workflows
+
+#### 5. **No Prescribed Session Structure**
+- **Bigpowers:** Skills can be invoked in any order; flexible orchestration
+- **GSD:** `/gsd-new-project` is always first (hard requirement)
+- **Win for Bigpowers:** Better for brownfield projects or partial workflows
+
+#### 6. **Explicit Consolidation (SKILL-INDEX.md)**
+- **Bigpowers:** Single canonical skill index (our analysis proposal)
+- **GSD:** Skills must be discovered from CLI listing or docs
+- **Win for Bigpowers:** Easier inventory management
+
+---
+
+### 5.3 Critical Gaps in Both
+
+#### 1. **Brownfield Project Support**
+- **GSD:** `/gsd-map-codebase` exists (4 parallel mappers)
+- **Bigpowers:** Minimal; `survey-context` is light
+- **Missing:** Systematic codebase analysis for existing codebases
+- **Recommendation:** Both need deeper codebase-first planning
+
+#### 2. **Async/Offline Handoff**
+- **GSD:** `continue-here.md` for session resumption
+- **Bigpowers:** STATE.md is static; no resumption format
+- **Missing:** Structured handoff between sessions or agents
+- **Recommendation:** Define continuation format (like GSD's)
+
+#### 3. **Cross-AI Review Loops**
+- **GSD:** `/gsd-plan-review-convergence` (replan until no HIGH concerns)
+- **Bigpowers:** `request-review` → `respond-review` (one cycle)
+- **Missing:** Automated convergence loops
+- **Recommendation:** Add convergence loop to Bigpowers
+
+#### 4. **Versioning & Branching Strategy**
+- **GSD:** `branching_strategy` in config.json (merge/keep/discard)
+- **Bigpowers:** CONVENTIONS.md mentions `gh pr` but no strategy config
+- **Missing:** Formal branching strategy management
+- **Recommendation:** Add to CLAUDE.md or config equivalent
+
+---
+
+## 6. Detailed Practice Comparison by Step
+
+### Step: Project Initialization
+
+| Aspect | GSD | Bigpowers |
+|--------|-----|-----------|
+| **Entry Point** | `/gsd-new-project` (interactive) | `using-bigpowers` → `elaborate-spec` + `model-domain` |
+| **Questioning** | Deep domain-specific probes (questioning.md philosophy) | Conversational; no prescribed probes |
+| **Research** | 4 parallel agents (Stack, Features, Architecture, Pitfalls) | No systematic research step |
+| **Requirements** | Explicit REQ-XX tracking, Validated/Active/Out of Scope | Via SCOPE.md, no explicit tracking |
+| **Roadmap** | Phases with success criteria, plan counts, wave analysis | RELEASE-PLAN.md with WSJF, no success criteria |
+| **Scope Guardrails** | HARD GATE: No scope creep, even if user suggests | No guardrail; user can add scope freely |
+| **Auto Mode** | `--auto @prd.md` skips questions, keeps research | No equivalent |
+| **Artifacts** | PROJECT.md, REQUIREMENTS.md, ROADMAP.md, research/, STATE.md, config.json | CLAUDE.md, CONVENTIONS.md, SCOPE.md (manual) |
+
+### Step: Phase Context (Discuss/Plan)
+
+| Aspect | GSD | Bigpowers |
+|--------|-----|-----------|
+| **Gray Area ID** | Phase-specific (not generic categories) | `challenge-design` is conversational |
+| **Decision Capture** | CONTEXT.md with locked choices | No explicit decision artifact |
+| **Scope Boundaries** | HARD GATE: Clarify vs. creep distinction | Ad-hoc |
+| **Documentation** | DISCUSSION-LOG.md (audit trail) | No equivalent |
+| **Mode Flexibility** | 6+ modes (--auto, --batch, --analyze, --power, --all, --assumptions) | Single conversational mode |
+| **Artifacts** | {phase}-CONTEXT.md, {phase}-DISCUSSION-LOG.md | No explicit artifact |
+
+### Step: Planning
+
+| Aspect | GSD | Bigpowers |
+|--------|-----|-----------|
+| **Research** | Mandatory (unless --skip-research) | Not explicitly in skill |
+| **Package Security** | Slopcheck + [SLOP]/[SUS]/[OK] gates | No security check |
+| **Task Format** | Every task has `verify: <command>` | Plan.md uses `→ verify:` per Karpathy |
+| **Dependency Management** | Plans declare dependencies; affects wave grouping | Not declared; affects execution order |
+| **Plan Verification** | Max 3-iteration checker loop | One-pass execution; no loop |
+| **Checkpoints** | `checkpoint:human-verify` before [ASSUMED] packages | No checkpoint mechanism |
+| **TDD Integration** | `workflow.tdd_mode: true` applies type:tdd to tasks | `develop-tdd` skill exists |
+| **Artifacts** | {phase}-RESEARCH.md, {phase}-{N}-PLAN.md, {phase}-VALIDATION.md | PLAN.md (single file) |
+
+### Step: Execution
+
+| Aspect | GSD | Bigpowers |
+|--------|-----|-----------|
+| **Wave Analysis** | Automatic grouping by dependencies | No grouping; sequential or manual parallelism |
+| **Parallel Agents** | Multiple executors per wave; fresh context each | `dispatch-agents` requires manual call |
+| **Context Isolation** | Fresh 200K per executor | Inherited from orchestrator |
+| **Adaptive Context** | 1M models receive prior wave SUMMARY.md + CONTEXT.md | No adaptive enrichment |
+| **Commit Safety** | --no-verify per executor + hooks once per wave | Ad-hoc; no formal safety mechanism |
+| **STATE.md Locking** | Atomic writes with O_EXCL + stale detection | Not formalized |
+| **Checkpoint Handling** | Pause execution, ask user, resume | No equivalent |
+| **Post-Execution Gates** | Schema Drift Gate + Decision Coverage Gate | Not formalized |
+| **Artifacts** | {phase}-{N}-SUMMARY.md, git commits | Code commits; SUMMARY.md manual |
+
+### Step: Verification
+
+| Aspect | GSD | Bigpowers |
+|--------|-----|-----------|
+| **UAT** | Manual walk-through + Verifier agent | Not formalized |
+| **Verifier Agent** | Reads PLAN.md, SUMMARY.md, REQUIREMENTS.md; identifies gaps | `validate-fix` is manual |
+| **Status Tracking** | VERIFICATION.md with passed|needs-review|needs-repair | No formal status |
+| **Gap Closure** | `/gsd-plan-phase --gaps` + `/gsd-execute-phase --gaps-only` | `validate-fix` then re-execute manually |
+| **Artifacts** | {phase}-VERIFICATION.md, (optional) {phase}-UAT.md | No formal artifact |
+
+### Step: Release
+
+| Aspect | GSD | Bigpowers |
+|--------|-----|-----------|
+| **Coverage Gates** | >= 80% overall, >= 95% business logic | Not formalized |
+| **PR Creation** | Automatic `gh pr create` with linked artifacts | Manual `release-branch` skill |
+| **Branching Strategy** | Config-driven (merge/keep/discard) | User decides |
+| **Worktree Cleanup** | Automatic if `use_worktrees: true` | Manual (via `guard-git`) |
+| **Milestone Archival** | Automatic collapse + tag creation | Manual (via `complete-milestone`) |
+| **Artifacts** | PR linked to PLAN.md, SUMMARY.md, VERIFICATION.md | Commit with message (via `commit-message`) |
+
+---
+
+## 7. Key Insights & Recommendations
+
+### 7.1 Architecture Philosophy
+
+**GSD: Orchestration-first**
+- User calls `/gsd-command` → Orchestrator loads context → Spawn agents → Collect → Update state
+- Workflows are state machines with explicit gates
+- Agents are thin specialists (each does one role)
+- Decision flow is: orchestrator controls the sequence, agents execute within that sequence
+
+**Bigpowers: Skill-first**
+- User calls skill → Skill prompt runs → Ad-hoc agent spawning if needed
+- Skills are user-facing entry points (verb-noun names)
+- Orchestration is manual (user chains skills)
+- Decision flow is: user picks the next skill, skill runs to completion
+
+**Hybrid Opportunity:** Bigpowers could keep skill-first naming but add thin orchestrators that call skills in sequence. e.g.:
+```
+/gsd-new-project → survey-context → elaborate-spec → model-domain → ... → plan-work
+```
+
+### 7.2 Context Isolation
+
+**GSD:** Fresh 200K per agent is a killer feature.
+- Prevents context rot (quality degradation as context window fills)
+- Each agent starts clean with only what it needs
+- Requires explicit context propagation (orchestrator loads + passes)
+
+**Bigpowers:** Relies on inherited session context
+- No context rot protection
+- Simpler for small projects, breaks down as complexity grows
+
+**Recommendation:** Adopt context isolation in Bigpowers via orchestrators
+- Each skill spawn gets `<files_to_read>` + `<context>` sections
+- Orchestrator manages handoff between skills
+
+### 7.3 Security Gates
+
+**GSD's supply-chain gate is unique:**
+- Blocks hallucinated package names (major AI coding risk)
+- Integrated into research → plan → execute pipeline
+- Easy to add to Bigpowers: extend `plan-work` with slopcheck call
+
+**Recommendation:** Add to Bigpowers `plan-work`:
+```markdown
+## Package Legitimacy
+Before recommending external packages:
+1. Run slopcheck on each package
+2. Tag results: [SLOP] remove, [SUS] add checkpoint, [OK] approve
+3. Add to plan: checkpoint:human-verify before [SUS] installs
+```
+
+### 7.4 Success Criteria & Verification
+
+**GSD:** ROADMAP.md success criteria are user-perspective observable behaviors.
+- Phase 1 Goal: "User can sign up with email"
+- Success Criteria: 1. "Signup form displays" 2. "Email validation works" 3. "Password reset works"
+- Verifier checks these explicitly
+
+**Bigpowers:** RELEASE-PLAN.md has "Objective" but no success criteria
+- "v1.17.0 — Guardrails" (what is success here?)
+- Harder to know when a release is done
+
+**Recommendation:** Add success criteria to RELEASE-PLAN.md per release:
+```markdown
+### v1.17.0 — Guardrails (WSJF 3.2)
+
+**Success Criteria:**
+- [ ] "Zoom-out mandate" implemented in plan-work as HARD-GATE
+- [ ] Surgical changes checklist in audit-code passes 100% tests
+- [ ] CONVENTIONS.md explicitly forbids "touch unrelated code"
+```
+
+### 7.5 Orchestration Formality
+
+**GSD:** Workflows are rigid (new-project → discuss → plan → execute → verify → ship)
+- Pro: Clear execution path, no decision paralysis
+- Con: Less flexible for non-standard scenarios
+
+**Bigpowers:** Skills are flexible (user picks next skill)
+- Pro: Adapts to any workflow
+- Con: Easy to skip steps, hard to enforce mandatory gates
+
+**Recommendation:** Bigpowers could add "orchestration modes":
+- **Standard:** Enforce discover → elaborate → plan → build → verify → release flow
+- **Brownfield:** Skip discover, start from `survey-context`
+- **Ad-hoc:** Pick skills freely (current Bigpowers behavior)
+
+### 7.6 Agent Specialization
+
+**GSD:** 31 specialized agents
+- Researchers, Planners, Checkers, Executors, Verifiers, Mappers, Debuggers, Auditors
+- Each agent has a single role
+- Model profile per agent (opus for planning, sonnet for research)
+
+**Bigpowers:** 38 skills, each invokes implicit agents
+- No predefined agent catalog
+- Skill is the entry point; agent role is implied
+
+**Recommendation:** Define Bigpowers agent taxonomy (like GSD's):
+```
+Researchers: survey-context, elaborate-spec, model-domain
+Planners: plan-work, plan-refactor
+Checkers: audit-code, request-review
+Executors: develop-tdd, execute-plan
+Verifiers: validate-fix, inspect-quality
+Mappers: deepen-architecture
+Debuggers: investigate-bug, diagnose-root
+```
+
+---
+
+## 8. Bigpowers Enhancements (Priority Order)
+
+### **Tier 1: Critical (Foundation)**
+
+1. **Add Orchestration Layer** (like GSD workflows)
+   - Define prescriptive core loop: survey → elaborate → plan → build → verify → ship
+   - Add gates: No code without plans, no plans without research, no shipping without verification
+   - Deliverable: `ORCHESTRATION.md` + thin coordinator skills
+
+2. **Add Context Isolation to Skill Spawning**
+   - Each skill spawn gets `<files_to_read>` + `<context>` sections
+   - Prevents context rot
+   - Deliverable: Update `execute-plan`, `plan-work`, `develop-tdd` with isolation
+
+3. **Add Supply-Chain Security Gate**
+   - Integrate slopcheck into `plan-work`
+   - Tag packages [SLOP]/[SUS]/[OK]
+   - Add checkpoint:human-verify before [SUS] installs
+   - Deliverable: Updated `plan-work` skill + security reference
+
+### **Tier 2: Important (UX/Clarity)**
+
+4. **Add Success Criteria to RELEASE-PLAN.md**
+   - Per-release observable behaviors
+   - Feeds into `validate-fix` verification
+   - Deliverable: Updated RELEASE-PLAN template
+
+5. **Add Reference Library (like GSD's)**
+   - Segregate knowledge from SKILL.md into `references/`
+   - Examples: gates.md, verification-patterns.md, tdd.md, thinking-models-*.md
+   - Deliverable: `references/` directory with 15-20 core docs
+
+6. **Formalize Agent Taxonomy**
+   - Map Bigpowers skills to agent roles (Researcher, Planner, Executor, Verifier, etc.)
+   - Define model profiles (opus for planning, sonnet for research)
+   - Deliverable: `AGENTS.md` documenting all agents + roles
+
+### **Tier 3: Nice-to-Have (Ergonomics)**
+
+7. **Add Automated Wave Analysis**
+   - Parse plan dependencies; group into waves for parallel execution
+   - Deliverable: Enhancement to `execute-plan`
+
+8. **Add Assumption Surfacing**
+   - Like `/gsd-discuss-phase --assumptions`
+   - Before planning, Claude surfaces what it's assuming about the phase
+   - Deliverable: Enhancement to `plan-work` or dedicated skill
+
+9. **Add Package Discovery & Legitimacy**
+   - Automated package recommendation with verification
+   - Deliverable: Enhancement to research step of `plan-work`
+
+10. **Add Continuation Format**
+    - Like GSD's `continue-here.md`
+    - Structured session resumption between context resets
+    - Deliverable: `continuation-format.md` + tool support
+
+---
+
+## 9. Conclusion
+
+**GSD is engineered for production reliability:**
+- Orchestration is explicit and auditable
+- Context isolation prevents rot
+- Security gates prevent supply-chain attacks
+- Verification is multi-step and rigorous
+
+**Bigpowers is engineered for developer ergonomics:**
+- Skill names are memorable (verb-noun)
+- BMAD phases are self-explanatory
+- Skill composition is flexible
+- Perfect for solo developers who know what they want
+
+**Hybrid Approach:** Bigpowers + GSD orchestration + GSD security gates would be ideal:
+- Keep Bigpowers' verb-noun skill names and BMAD phases
+- Add GSD's orchestration layer and context isolation
+- Add GSD's security gates and reference library
+- Result: Bigpowers' usability + GSD's production reliability
+
+---
+
+## 10. Full GSD Workflow Call Map (6 Commands)
+
+```
+START
+  ↓
+/gsd-new-project
+  ├── Questions + 4 parallel researchers
+  ├── Synthesizer
+  ├── Roadmapper
+  ├── User approval
+  └── → PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md, research/, config.json
+  ↓
+LOOP [for each phase]:
+  ├── /gsd-discuss-phase [N]
+  │   ├── Gray area identification
+  │   ├── User discussion (6+ modes: --auto, --batch, --analyze, --power, --all, --assumptions)
+  │   └── → {phase}-CONTEXT.md, {phase}-DISCUSSION-LOG.md
+  │   ↓
+  ├── /gsd-plan-phase [N]
+  │   ├── Research (if needed)
+  │   │   ├── Phase Researcher
+  │   │   ├── Package Legitimacy Audit (slopcheck)
+  │   │   └── → {phase}-RESEARCH.md with [SLOP]/[SUS]/[OK] verdicts
+  │   │   ↓
+  │   ├── Planning (max 3 iterations if issues found)
+  │   │   ├── Planner
+  │   │   │   ├── Every task: verify: <command>
+  │   │   │   ├── Dependencies declared
+  │   │   │   ├── Checkpoints for [ASSUMED]/[SUS]
+  │   │   │   └── STRIDE threat model
+  │   │   │   ↓
+  │   │   └── Plan Checker (loop until clean)
+  │   │       ├── Quality Gate
+  │   │       ├── Requirements Coverage Gate (REQ-IDs)
+  │   │       ├── Decision Coverage Gate (CONTEXT.md)
+  │   │       └── Safety Gate
+  │   │   ↓
+  │   └── → {phase}-{N}-PLAN.md files
+  │   ↓
+  ├── /gsd-execute-phase [N]
+  │   ├── Wave analysis (dependency grouping)
+  │   ├── Per-wave: spawn executors in parallel
+  │   │   ├── Fresh 200K context per executor
+  │   │   ├── Execute task
+  │   │   ├── Commit atomically (conventional)
+  │   │   └── Create SUMMARY.md
+  │   ├── Post-wave: run pre-commit hooks once
+  │   ├── Schema Drift Gate
+  │   └── → {phase}-{N}-SUMMARY.md, git commits
+  │   ↓
+  ├── /gsd-verify-work [N]
+  │   ├── User acceptance testing (manual)
+  │   ├── Verifier agent review
+  │   │   ├── Read PLAN.md, SUMMARY.md, REQUIREMENTS.md
+  │   │   ├── Identify gaps
+  │   │   ├── Generate fix plans (if needed)
+  │   │   └── Status: passed | needs-review | needs-repair
+  │   ├── If needs-repair:
+  │   │   ├── /gsd-plan-phase --gaps
+  │   │   ├── /gsd-execute-phase --gaps-only
+  │   │   └── Re-verify
+  │   └── → {phase}-VERIFICATION.md
+  │   ↓
+  ├── /gsd-ship [N]
+  │   ├── Coverage gates (>= 80%, >= 95% business logic)
+  │   ├── Create PR (gh pr create)
+  │   ├── Merge to main
+  │   ├── Worktree cleanup (if used)
+  │   ├── If last phase in milestone:
+  │   │   ├── Archive phase artifacts
+  │   │   ├── Create git tag (vX.Y.Z)
+  │   │   └── Update MILESTONES.md
+  │   └── → Release shipped
+  │   ↓
+END LOOP
+```
+
+This is the core GSD workflow. Optional workflows exist for debugging, code review, documentation, workspace management, etc.
+