improvement-pipeline

---

name: improvement-pipeline description: Coordinate sequential improvement stages (analyze → propose → build → validate) with Prompt Architect clarity and Skill Forge guardrails. allowed-tools:

• Read
• Write
• Edit
• Bash
• Glob
• Grep
• Task
• TodoWrite model: claude-3-5-sonnet x-version: 3.2.0 x-category: tooling x-vcl-compliance: v3.1.1 x-cognitive-frames:
• HON
• MOR
• COM
• CLS
• EVD
• ASP
• SPC

---

L1 Improvement

• Simplified the pipeline into an English-first SOP aligned with Prompt Architect constraint ordering.
• Added structure-first documentation, confidence ceilings, and memory tagging per Skill Forge.
• Clarified validation gates and rollback expectations.

STANDARD OPERATING PROCEDURE

Purpose

Run disciplined improvement cycles for any artifact (prompt, skill, doc, script) across analyze → propose → build → validate → deliver.

Trigger Conditions

• Positive: requests to improve/refine/upgrade an artifact with validation gates.
• Negative: net-new creation (route to creation skills) or emergency fixes without evaluation.

Guardrails

• Maintain structure-first docs (SKILL, README, examples/tests/references). Log deviations.
• Confidence ceilings mandatory; cite evidence and baselines.
• No skipping validation; rollback if metrics regress or confidence < threshold.
• Memory tagging for each cycle.

Execution Phases

1. Intent & Constraints – Use Prompt Architect style to extract objectives, constraints, and success metrics.
2. Analysis – Review current state, gather baselines, and capture risks.
3. Proposals – Generate options with deltas and blast radius notes; select path.
4. Build – Implement chosen changes with minimal diffs and rollback plan.
5. Validation – Run tests/benchmarks/lints; compare to baselines; document outcomes.
6. Delivery – Summarize changes, decisions, metrics, and confidence ceiling; archive artifacts.

Output Format

• Target artifact, goals, constraints, and selected proposal.
• Changes applied with evidence of improvement vs baseline.
• Risks, follow-ups, and rollback info.
• Confidence: X.XX (ceiling: TYPE Y.YY) and memory keys used.

Validation Checklist

• Constraints and success metrics captured.
• Baseline recorded; validation executed.
• Regression/impact analysis completed.
• Memory tagged and artifacts stored.
• Confidence ceiling declared.

Integration

• Memory MCP: skills/tooling/improvement-pipeline/{project}/{timestamp}.
• Hooks: align with Skill Forge latency bounds; integrate with eval-harness when gating is required.

Confidence: 0.70 (ceiling: inference 0.70) – Pipeline rewritten to blend Prompt Architect clarity with Skill Forge delivery gates.