SDLC Layer 0 Integration Suggestions

SDLC Layer 0 Integration Suggestions

Generated: 2026-02-23 Plan Context: SDLC Layer Separation Architecture — Layer 0 (SDLC-Agnostic) Source: Full read of all Layer 0 items from sdlc-layer-candidates-master.md

Plan Summary (Layer 0 Scope)

Layer 0 = SAM pipeline, human touchpoints, artifact conventions, RT-ICA, verification protocol, task format, subagent contract, evidence discipline.

From Plugins

1. development-harness

Path: plugins/development-harness/CLAUDE.md

Suggested integration: Add as the canonical Layer 0 hub for the SAM 7-stage pipeline. The plan should reference this plugin as the primary implementation of the SAM pipeline with ARL-derived human touchpoints and Voltron-style composition.

Deliverable: Layer 0 hub section — "The development-harness plugin is the reference implementation of the SAM 7-stage pipeline. It owns: process orchestration, ARL human touchpoint decisions, artifact management, and fallback behavior. Language plugins provide manifests; the harness resolves roles at runtime."

Extracted content (Role Resolution mermaid — include in plan):

flowchart TD
    Start([Feature Request]) --> Detect{Detect Project Language}
    Detect -->|pyproject.toml| Python[Python plugin]
    Detect -->|package.json| TypeScript[TypeScript plugin]
    Detect -->|Cargo.toml| Rust[Rust plugin]
    Detect -->|None found| Fallback[General-purpose agents]
    Python --> Manifest[Read language manifest]
    TypeScript --> Manifest
    Rust --> Manifest
    Manifest --> Resolve[Resolve roles to agents]
    Resolve --> Override{Flow override declared?}
    Override -->|Yes| Custom[Use plugin-defined flow]
    Override -->|No| Default[Use default SAM pipeline]

2. default-development-flow

Path: plugins/development-harness/skills/development-harness/references/default-development-flow.md

Suggested integration: Add to Human Touchpoints section. The default flow defines Gate 1 (S1→S2) and Gate 2 (S4→S5), plus loop limits: 3 NEEDS_WORK, 2 NOT_CERTIFIED before human escalation.

Amendment: Plan should explicitly state loop limits as Layer 0 constants:

• NEEDS_WORK loop limit: 3 iterations per task
• NOT_CERTIFIED loop limit: 2 iterations before escalation

Extracted content (Pipeline Overview mermaid):

flowchart TD
    Start([Feature Request]) --> S1[Stage 1 — Discovery]
    S1 -->|ARTIFACT:DISCOVERY| HT1{Human Touchpoint?}
    HT1 -->|ARL — unbound constraints, domain knowledge needed| Human1[Escalate to Human]
    HT1 -->|ARL — bound constraints, sufficient context| S2[Stage 2 — Planning + RT-ICA]
    Human1 --> S2
    S2 -->|ARTIFACT:PLAN| S3[Stage 3 — Context Integration]
    S3 -->|Contextualized Plan| S4[Stage 4 — Task Decomposition]
    S4 -->|ARTIFACT:TASK per task| HT2{Human Touchpoint?}
    HT2 -->|ARL — high complexity, novel architecture| Human2[Escalate to Human]
    HT2 -->|ARL — routine, patterns exist| S5[Stage 5 — Execution]
    Human2 --> S5
    S5 -->|ARTIFACT:EXECUTION| S6[Stage 6 — Forensic Review]
    S6 -->|NEEDS_WORK| S5
    S6 -->|COMPLETE| S7[Stage 7 — Final Verification]
    S7 -->|NOT_CERTIFIED| S4
    S7 -->|CERTIFIED| Done([Feature Complete])

3. human-touchpoint-model

Path: plugins/development-harness/skills/development-harness/references/human-touchpoint-model.md

Suggested integration: Add to Human Touchpoints section. The constraint classification (Bound/Unbound/Mixed) and risk assessment (reversible, local scope, existing patterns) are Layer 0 decision logic.

Amendment: Plan should include the escalation format — when escalating, present: (1) what stage, (2) what triggered it, (3) what the agent knows, (4) what the agent does not know, (5) decision options. Never present vague "please review" requests.

Nugget: Dynamic escalation points beyond pre-scheduled gates: NEEDS_WORK loop limit, NOT_CERTIFIED loop limit, agent failure, quality gate cascade failure, contradiction detected in S3.

4. artifact-conventions

Path: plugins/development-harness/skills/development-harness/references/artifact-conventions.md

Suggested integration: Add to Artifact Conventions section. Token pattern ARTIFACT:{TYPE}({SCOPE_OR_ID}), file layout .planning/harness/, required sections per artifact type.

Amendment: Plan should list the seven artifact types and their required sections (DISCOVERY, PLAN, CONTEXT, TASK, EXECUTION, REVIEW, VERIFICATION). Coexistence with .planning/gsd/ and .planning/backlog/ is explicit.

Nugget: Each artifact includes predecessor/successor references. Traceability chain: S7 builds matrix linking requirements (S1) through plan (S2), tasks (S4), execution (S5), review (S6).

5. role-resolution-protocol

Path: plugins/development-harness/skills/development-harness/references/role-resolution-protocol.md

Suggested integration: Add to Layer 0 process — role resolution happens at pipeline start before S1. Abstract roles: architect, test-designer, code-reviewer, design-spec, linting.

Extracted content (Resolution Process mermaid):

flowchart TD
    Start([Pipeline Start]) --> ScanRoot[Scan project root for language markers]
    ScanRoot --> Markers{Which markers found?}
    Markers -->|pyproject.toml, setup.py, setup.cfg| Python[Language = Python]
    ...
    FindManifest --> ManifestExists{Manifest found?}
    ManifestExists -->|Yes| ParseManifest[Parse Role Fulfillment section]
    ManifestExists -->|No| Fallback[Use general-purpose for all roles]

6. language-manifest-schema

Path: plugins/development-harness/skills/development-harness/references/language-manifest-schema.md

Suggested integration: Add to Layer 0 schema — language manifests have four required sections (Role Fulfillment, Quality Gates, Project Detection) and one optional (Process Flow Override). Flow override must include at least one human touchpoint gate and end with CERTIFIED/NOT_CERTIFIED.

Nugget: Validation failures produce warnings but do not block the pipeline. Harness falls back to general-purpose for any section that fails validation.

7. Workflow stages (7)

Paths: plugins/development-harness/skills/workflows/{discovery,planning,context-integration,task-decomposition,execution,forensic-review,final-verification}/SKILL.md

Suggested integration: Add to SAM Pipeline section — each stage has a dedicated skill with input/output contracts. Key cross-stage rules:

• S2 Planning: RT-ICA pre-pass, BLOCK on MISSING critical items
• S4 Task Decomposition: CLEAR ordering, CoVe only when accuracy-risk medium/high
• S5 Execution: Task file IS the complete prompt; fresh session per task
• S6 Forensic Review: Producer and reviewer must be different agents
• S7 Final Verification: Goal-backward verification, not implementation-forward

Nugget: S6 principle — "AI cannot reliably self-evaluate." Forensic review uses a separate agent with fresh context.

8. planner-rt-ica

Path: plugins/development-harness/skills/planner-rt-ica/SKILL.md

Suggested integration: Add to RT-ICA section. Planner RT-ICA is distinct from execution RT-ICA:

• planner-rt-ica: Enables safe planning under uncertainty; localizes gaps; produces unblock paths; outputs APPROVED-FOR-PLANNING / APPROVED-WITH-GAPS / BLOCKED-FOR-PLANNING
• rt-ica: Blocks execution if required inputs remain missing

Amendment: Any task produced under APPROVED-WITH-GAPS MUST still pass rt-ica before execution by a specialist agent.

9. validation-protocol

Path: plugins/development-harness/skills/validation-protocol/SKILL.md

Suggested integration: Add to Verification Protocol section. Four-step protocol: (1) Reproduce failing state, (2) Define success criteria, (3) Apply fix and observe, (4) Verify result. Success = observing intended behavior, not absence of errors.

Nugget: Anti-patterns to avoid: claiming success without reproducing failure; confusing "no errors" with success; skipping verification; partial verification.

10. clear-cove-task-design, generate-task

Paths: plugins/development-harness/skills/clear-cove-task-design/SKILL.md, plugins/development-harness/skills/generate-task/SKILL.md

Suggested integration: Add to Task Format section. CLEAR ordering: Context, Objective, Inputs, Requirements, Constraints, Outputs, Verification, Handoff. CoVe checks only when accuracy-risk is medium or high.

Amendment: Task prompt template must include: Required Inputs (with assumptions and how to confirm), Verification Steps, CoVe Checks (conditional), Handoff.

11. verification-gate

Path: plugins/verification-gate/skills/verification-gate/SKILL.md

Suggested integration: Add to Verification Protocol section. Four mandatory pre-action checkpoints: (1) Hypothesis stated, (2) Hypothesis verified, (3) Hypothesis-action alignment, (4) Pattern-matching detection.

Nugget: Verification-gate's 4 checkpoints should run before S5 Execution self-verification. They prevent pattern-matching from overriding explicit reasoning. Activate before any Bash/Write/Edit/NotebookEdit.

Amendment: Remove or rephrase "5% cost for 95% reliability improvement" — grooming report 2026-02-23 marked this as REFUTED (unsubstantiated). Use qualitative phrasing instead.

12. orchestrator-discipline

Path: plugins/orchestrator-discipline/skills/orchestrator-discipline/SKILL.md, references/investigation-escalation.md

Suggested integration: Add to Layer 0 process — orchestrator context window discipline. Two anti-patterns: (1) Investigation Escalation (3+ Read/Grep/Bash on source files without Edit/Write/Task), (2) Agent Output Polling (TaskOutput block=false on running agent).

Amendment: Plan should state: "Orchestrators delegate; agents implement. No exemption categories for 'config changes', 'small edits', or 'just TOML/YAML'."

Extracted content (Correct workflow):

flowchart TD
    Start([Task arrives]) --> Q1{Does orchestrator need current codebase state?}
    Q1 -->|Yes| Explore["Delegate to Explore agent with diagnostic command"]
    Q1 -->|No| Direct[Delegate implementation to specialist agent]
    Explore --> Summary[Receive summary from agent]
    Summary --> Q2{Scope changed?}
    Q2 -->|Yes| User[Present to user for routing decision]

13. agent-orchestration, how-to-delegate

Paths: plugins/agent-orchestration/skills/agent-orchestration/SKILL.md, plugins/agent-orchestration/skills/how-to-delegate/SKILL.md

Suggested integration: Add to Delegation section. WHERE-WHAT-WHY framework. Delegation template: OBSERVATIONS, DEFINITION OF SUCCESS, CONTEXT, YOUR TASK. Pre-gathering anti-pattern: never run commands to collect data for agents.

Amendment: Plan should require: "Start every Task prompt with Your ROLE_TYPE is sub-agent." Include the 10-step delegation worksheet from how-to-delegate for orchestrators.

Nugget: Scientific delegation = Observations + Success Criteria + Available Resources - Assumptions - Prescriptions.

14. hallucination-detector

Path: plugins/hallucination-detector/commands/hallucination-audit.md

Suggested integration: Add to Evidence Discipline section. Five triggers: (1) Speculation language, (2) Causality without evidence, (3) Pseudo-quantification, (4) Completeness claims, (5) Delegation micromanagement.

Nugget: Run hallucination audit after producing answers or reviewing sub-agent output. CLAUDE.md already references /hallucination-detector:hallucination-audit when reviewing agent output.

15. the-rewrite-room/output-contracts, fidelity-rules

Paths: plugins/the-rewrite-room/skills/the-rewrite-room/references/output-contracts.md, fidelity-rules.md

Suggested integration: Add to Artifact Conventions section. STATUS block: DONE, BLOCKED, FAILED. BLOCKED = workflow did not complete; FAILED = output exists but failed validation.

Amendment: Fidelity rules apply to summarization: Read before summarizing; Extract before abstracting; Preserve counts; Distinguish absence from nonexistence; No lossy re-summarization; State confidence explicitly.

Nugget: Rule 4 — "Not mentioned in this document" vs "Doesn't exist" — precise language when information is not found.

16. plugin-creator/arl

Path: plugins/plugin-creator/skills/arl/SKILL.md

Suggested integration: Add to ARL Meta-Layer (cross-reference). ARL theory informs Layer 0: HOOTL concept, 10 gates (R1-R10), scope-feasibility matrix, decision tree for gate replacement.

Nugget: ARL is not a process to run — it is research that informs how to design processes. Relationship triangle: ARL produces theory, SAM applies it, agentskill-kaizen validates.

Amendment: Plan should note that the 4 build-from-scratch requirements (R6, R7, R8, R10) emerge from iterative refinement — invisible to single-pass pipelines.

From Skills

17. work-backlog-item

Path: .claude/skills/work-backlog-item/SKILL.md

Suggested integration: Add to Layer 0 process — bridge from BACKLOG.md to SAM planning. Arguments: #N, --auto, close, resolve, setup-github. Stops if item has existing Plan or RT-ICA returns BLOCKED.

NOTE (2026-02-27): BACKLOG.md was removed. Backlog items now live in .claude/backlog/ per-item files; GitHub Issues are the source of truth.

Amendment: Plan should document the workflow: find item → fact-check → RT-ICA → groom → SAM planning → BACKLOG.md update. GitHub Issue is canonical status when **Issue**: #N exists.

18. sam-definition

Path: .claude/skills/work-backlog-item/references/sam-definition.md

Suggested integration: Add as canonical SAM definition for Layer 0. Core principles: stateless agents, externalized memory, single responsibility, message passing, verification at boundaries, deterministic backpressure, RT-ICA gate, semantic artifact tokens, structure over instruction, AI cannot self-evaluate.

Amendment: Plan should cite the 7-stage pipeline table and artifact flow from this file.

19. github-integration, validation-plan

Paths: .claude/skills/work-backlog-item/references/github-integration.md, validation-plan.md

Suggested integration: Add to Layer 0 process — GitHub integration is optional but when used: labels (priority:, type:, status:*), milestones, setup-github command. Validation plan V1–V6 provides verification commands.

Nugget: BACKLOG.md ↔ Issue mapping. When **Issue**: #N exists, GitHub status is canonical.

20. groom-backlog-item

Path: .claude/skills/groom-backlog-item/SKILL.md

Suggested integration: Add to Layer 0 process — grooming flow: fact-check → RT-ICA → spawn groomer agents (max 5 concurrent). REFUTED claims become MISSING conditions in RT-ICA.

Amendment: Plan should state: "Fact-check runs before RT-ICA. Training data recall is NOT evidence."

21. rt-ica

Path: .claude/skills/rt-ica/SKILL.md

Suggested integration: Add to RT-ICA section. Mandatory pre-planning checkpoint. BLOCK if any condition is MISSING. Integration points: before top-level plan, before delegating to agents, before finalizing acceptance criteria.

Amendment: Plan should include the RT-ICA output format: Goal, Success Output, Conditions (reverse prerequisites), Verification (AVAILABLE/DERIVABLE/MISSING), Decision (APPROVED/BLOCKED).

22. subagent-contract

Path: .claude/skills/subagent-contract/SKILL.md

Suggested integration: Add to Subagent Contract section. DONE/BLOCKED signaling. Scope discipline: no invention, no assumption, no scope creep. Quality checklist before DONE; blocked checklist before BLOCKED.

Amendment: Plan should require all specialist agents to load this contract. BLOCKED is preferred over guessing.

23. fact-check

Path: .claude/skills/fact-check/SKILL.md

Suggested integration: Add to Evidence Discipline section. Primary-source verification only. VERIFIED/REFUTED/INCONCLUSIVE verdicts. Wave spawning (max 5 concurrent). Valid evidence: WebFetch, WebSearch, CLI output, repo source, MCP tools.

Nugget: REFUTED claims from fact-check must become MISSING conditions in RT-ICA.

24. find-cause

Path: .claude/skills/find-cause/SKILL.md

Suggested integration: Add to Evidence Discipline section. Evidence-chain protocol: every claim must have command output, file content, or direct observation. NOT evidence: docs intent, training recall, inference from absence.

Nugget: Use for investigation requests; transforms vague requests into reproducible-proof investigations.

25. verify

Path: .claude/skills/verify/SKILL.md

Suggested integration: Add to Verification Protocol section. Completion checklist: Task type, WORKS check (executable vs static), FIXED check (for bug fixes), Quality gates, Honesty check.

Nugget: "If you cannot demonstrate it working in practice with evidence, it is NOT done." CLAUDE.md requires /verify before claiming task completion.

26. delegate, cove-prompt-design, scientific-thinking, commit-staged

Paths: .claude/skills/delegate/SKILL.md, cove-prompt-design/SKILL.md, scientific-thinking/SKILL.md, commit-staged/SKILL.md

Suggested integration:

• delegate: Quick WHERE-WHAT-WHY template; for full delegation use agent-orchestration how-to-delegate
• cove-prompt-design: When to use CoVe (factual accuracy, multi-fact reasoning); three phases: generate → verify → revise
• scientific-thinking: Hypothesis → Prediction → Experiment → Conclusion; for debugging, architecture, complex refactoring
• commit-staged: Conventional commits (feat/fix/docs/scope); Layer 0 artifact for commit phase

From Agents

27. backlog-item-groomer

Path: .claude/agents/backlog-item-groomer.md

Suggested integration: Add to Layer 0 process — produces context manifest: RT-ICA Assessment, supporting skills, related agents, prior work, dependencies, blockers. Uses Glob, Grep, Read.

Nugget: If orchestrator provides pre-computed RT-ICA, groomer uses it and focuses discovery on filling MISSING conditions.

28. fact-checker

Path: .claude/agents/fact-checker.md

Suggested integration: Add to Evidence Discipline section. Verifies single factual claim. MUST use WebFetch/WebSearch/gh — training recall rejected. Returns structured VERIFIED/REFUTED/INCONCLUSIVE with citations.

Nugget: Applies CoVe before finalizing verdict — generate falsification questions, check against second source.

29. topic-specialist

Path: .claude/agents/topic-specialist.md

Suggested integration: Add to Evidence Discipline section. Domain specialist; researches primary sources only; can update or create skills with verified findings. Loads fact-check, find-cause, research-curator.

Nugget: Source priority: GitHub source code > README > issues > official docs > WebSearch.

30. context-gathering, context-refinement

Paths: .claude/agents/context-gathering.md, context-refinement.md

Suggested integration: Add to Layer 0 process — context-gathering produces "How It Currently Works" narrative; context-refinement adds "Discovered During Implementation" when drift detected.

Nugget: Context-gathering writes ONLY to the task file; forbidden from editing other files. Context-refinement only updates if significant discoveries or drift found.

31. logging

Path: .claude/agents/logging.md

Suggested integration: Add to Layer 0 process — consolidates work logs into task's Work Log section. Chronological order. Use during context compaction or task completion.

Nugget: Assessment phase before changes: read entire task file, read transcript, plan REMOVE/UPDATE/ADD.

32. code-review

Path: .claude/agents/code-review.md

Suggested integration: Add to Verification Protocol section. LLM-slop detection: reimplementing existing scaffolding, failing to follow norms, junk patterns, placeholders, hallucinated defaults. Critical/Warning/Suggestion categories.

Nugget: Use ONLY when explicitly requested or invoked by protocol. Do not use proactively.

Cross-Cutting Integration Summary

Recommended Plan Amendments

1. Explicit loop limits: 3 NEEDS_WORK, 2 NOT_CERTIFIED before human escalation
2. Verification-gate order: 4 checkpoints run before S5 Execution self-verification
3. Remove unsubstantiated claim: verification-gate "5% cost for 95% reliability" — REFUTED
4. Orchestrator discipline: No exemption categories; 3+ source reads without Edit/Write/Task = trigger
5. Fact-check before RT-ICA: REFUTED claims become MISSING conditions in RT-ICA