doc-research-improvements

Research-driven critique and enhancement of Amelia architecture or ideas docs. Use with GPT-5.2 when given a design/architecture doc to analyze concepts, find sources, and propose improvements plus OSS candidates.

$ Installer

git clone https://github.com/existential-birds/amelia /tmp/amelia && cp -r /tmp/amelia/.claude/skills/doc-research-improvements ~/.claude/skills/amelia

// tip: Run this command in your terminal to install the skill


name: doc-research-improvements description: Research-driven critique and enhancement of Amelia architecture or ideas docs. Use with GPT-5.2 when given a design/architecture doc to analyze concepts, find sources, and propose improvements plus OSS candidates.

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

ParameterDescriptionRequired
doc_pathPath to the doc to analyze.Required (unless doc_text)
doc_textRaw doc content if not reading from file.Optional
focus_areasComma list: architecture, data-model, orchestration, security, observability, testing, UX, scaling, cost.Optional
depthquick (top 3 items) or standard (top 5-8 items).Optional
constraintsExplicit constraints to respect (ex: local-first, SQLite, no background jobs).Optional

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.