doc-research-improvements

Doc Research and Improvement (GPT-5.2)

Use this skill to analyze a single doc (architecture or idea), research its referenced concepts, and propose improvements with external sources and implementation candidates.

Input Parameters

Repository Context (Align To)

These are core assumptions from docs/site/architecture/* and docs/site/ideas/*:

• Amelia is a local-first, agentic coding orchestrator with LangGraph, FastAPI, and SQLite.
• Primary workflow: Architect -> Developer -> Reviewer with human approval gates.
• Design principles: structured handoffs, verify before done, environment as truth.
• Drivers and trackers are pluggable abstractions.
• Ideas are exploratory; suggest improvements without assuming commitment.

Use these as guardrails when proposing changes.

Required Workflow

1. Ingest and summarize the doc

   • Identify goals, scope, non-goals, and assumptions.
   • Note dependencies on current architecture or roadmap phases.

2. Extract concepts, frameworks, and techniques

   • List all explicit or implied frameworks (ex: LangGraph, RAG, SOX/ICFR, BDD).
   • Classify each as architecture, process, data, evaluation, or UI.

3. Research and validate

   • Use authoritative sources: specs, official docs, standards, reputable research.
   • Capture at least 1 source per key concept. If none, label as hypothesis.
   • Summarize evidence in 1-2 lines with citations.

4. Propose improvements

   • Prioritize items by impact and feasibility.
   • Provide rationale tied to evidence and Amelia constraints.
   • Include tradeoffs and risks.
   • Call out where changes touch existing modules (ex: amelia/core/, amelia/server/).

5. Recommend OSS candidates

   • Suggest open-source libraries or frameworks needed to implement changes.
   • Include license, maturity, and integration notes.
   • Prefer Python 3.12, FastAPI, React, SQLite compatible stacks.

Output Format

Doc Snapshot

• Goal, scope, non-goals, assumptions

Extracted Concepts | Concept | Category | Why it matters | Sources |

Evidence and Research Notes

• Bullet list of sources with 1-line relevance

Improvements (Ranked)

1. Title
   • Rationale + evidence (cite)
   • Tradeoffs/risks
   • Implementation notes (files/modules)

OSS and Framework Candidates | Need | Options | License | Why it fits |

Open Questions

• Any missing inputs or decisions required

Guardrails

• Do not contradict architecture principles unless the doc explicitly proposes it.
• If the doc is exploratory, frame improvements as experiments or phased probes.
• Avoid proprietary SaaS recommendations unless the doc already assumes them.
• No hallucinated citations; if unsure, say so and suggest verification.