Evaluate Sdlc Layers

Evaluate SDLC Layers

Systematically evaluate the SDLC Layer Separation Architecture implementation and support iterative improvement. Treats the implementation as first-pass until validated.

Arguments

• --dry-run — Run all checks, produce report only. Do not apply fixes.
• --fix — After evaluation, apply safe fixes for broken references, missing metadata, or obvious gaps. Report what was changed.
• (no args) — Evaluate and produce report; offer to fix or delegate fixes.

Evaluation Checklist

Run each check and record PASS / FAIL / SKIP with evidence.

1. Cross-Reference Validation

For each linked path in .claude/docs/sdlc-layers/ and related docs:

• sam-definition.md — exists at .claude/skills/work-backlog-item/references/sam-definition.md
• plugins/development-harness/CLAUDE.md — exists
• stateless-agent-methodology/research/arl/PROVENANCE.md — exists (sibling repo or configured path)
• Layer 0 docs → TASK_FILE_FORMAT.md — exists at .claude/docs/TASK_FILE_FORMAT.md
• Layer 1 → language-manifest-schema.md, role-resolution-protocol.md — exist in development-harness
• Layer 2 → plugins/development-harness/docs/layer-2/ — exists with README, schema, pilot profiles

Evidence: List each path checked and result (exists / 404 / wrong content).

2. Doc Completeness

• Layer 0: All 9 docs present (README, sam-pipeline, arl-touchpoints, artifact-conventions, rt-ica-gate, verification-protocol, task-file-format, evidence-discipline, orchestrator-discipline)
• Layer 1: All 6 docs present (README, layer-1-overview, language-manifest-template, linting-discovery-protocol, workflow-pattern-taxonomy, harness-role-mapping)
• Layer 2: README, layer-2-overview, stack-profile-schema, stack-profile-template; pilot profiles python-fastapi, python-cli
• ARL: arl-meta-layer.md, arl-human-probing-design.md

Evidence: Glob or Read results for each expected file.

3. Knowledge-Explorer Layer Filter

• uv run research/knowledge-explorer.py list --layer 0 — returns entries with layer: "0"
• uv run research/knowledge-explorer.py list --layer 1 — returns entries with layer: "1"
• uv run research/knowledge-explorer.py list --layer 2 — returns entries with layer: "2"
• Entries without layer metadata are excluded when --layer is used (expected)

Evidence: Paste command output for each.

4. Research Entry Layer Metadata

• evaluation-testing/harness-engineering-openai.md — has layer: "0"
• api-frameworks/fastapi.md, api-frameworks/tornado.md — have layer: "2", language, stack
• developer-tools/copier-astral.md — has layer: "1" (or 2 if stack-scaffold)
• research/README.md — has "Layer Mapping" section

Evidence: Grep for layer: in frontmatter of each.

5. Integration Points

• work-backlog-item SKILL — documents --language, --stack; references layer docs
• groom-backlog-item SKILL — documents ARL human-probing integration; references arl-human-probing-design
• language-manifest-schema.md — has "Inherits from Layer 0"; typecheck: (none); Conventions schema
• role-resolution-protocol.md — has "Layer 0 gates apply before role resolution"
• plugins/development-harness/CLAUDE.md — references layer model

Evidence: Grep or Read for key phrases.

6. Consistency with Plan

• Plan deliverables (from attached plan) — compare File and Directory Changes table to actual files
• Dependency order — Layer 0 → Layer 1 → Layer 2 → Research → SAM/ARL → ARL probing → work-backlog-item

Evidence: List any plan items not yet implemented or diverged.

Output Format

Produce a structured report:

## SDLC Layer Evaluation Report
Date: {YYYY-MM-DD}

### Summary
- Cross-Reference: {PASS|FAIL|PARTIAL} — {brief}
- Doc Completeness: {PASS|FAIL|PARTIAL}
- Knowledge-Explorer: {PASS|FAIL|PARTIAL}
- Research Metadata: {PASS|FAIL|PARTIAL}
- Integration Points: {PASS|FAIL|PARTIAL}
- Plan Consistency: {PASS|FAIL|PARTIAL}

### Findings
1. [Category] {finding} — {suggested fix}
2. ...

### Recommended Actions
- [ ] {action 1}
- [ ] {action 2}

Iteration

After evaluation:

1. If --fix: Apply safe fixes (broken paths, missing frontmatter fields, obvious typos). Report each change.
2. If no --fix: Present findings; offer to create backlog items or apply fixes.
3. Re-run: After fixes, re-run evaluation to confirm improvements.

Experiments

Flow experiments and learnings live in sam-flow-experiments. Clone via SSH: git clone git@github.com:Jamie-BitFlight/sam-flow-experiments.git. When iterating, consider running experiments against concept fixtures to validate changes.

References

• .claude/docs/sdlc-layers/
• verify skill — evidence discipline
• groom-backlog-item — orchestration pattern