ai-architect

An advisory designer for artificial-intelligence systems and software agents. It plans how agents, prompts, information retrieval, evaluations, orchestration, models, memory, and safety controls should fit together, but does not write production code.

In plain words
What is it for?
Use it to design agent teams, prompt structures, retrieval-augmented generation (systems that search information before answering), evaluation checks, model tiers, memory systems, knowledge graphs, and autonomy safeguards.
Why use it?
It helps expose design problems before implementation, such as oversized prompts, poor search results, unnecessary sequential work, or unsuitable model choices.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/lucassantana-dev/sharekit/ai-architect
Clone the repo
git clone --depth 1 https://github.com/LucasSantana-Dev/sharekit

Made for: Claude Code.

Per session 88 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,975 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00088 $0.01975
Opus 5 $0.00044 $0.00988
Sonnet 5 $0.00018 $0.00395
Haiku 4.5 $0.00009 $0.00198

Measured 2d ago against content hash c81f02c15f6a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

ai-architect scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

sharekit-profile/.claude/agents/ai-architect.md · 99 lines

How it starts

The opening of the file, as written. The whole thing — 99 lines — stays where its author put it; the contents beside it link to each section on GitHub.

<Agent_Prompt> You are AI Architect — a systems designer for AI and agent infrastructure (distinct from codebase architecture). You are responsible for: agent topology, prompt & context architecture, RAG/retrieval design, eval-gate strategies, orchestration patterns (parallel, pipeline, adversarial), model-tier optimization, memory/knowledge-graph design, and autonomy tier design (T0-T3 guardrails). You are NOT responsible for: implementing agents/code (code-architect/builder), testing AI systems (test-engineer), securing agent outputs (security-reviewer), or shipping the system (deployment-automation). Role: ADVISORY — you design and recommend. Implementation and verification are the orchestrator's.

<Why_This_Matters> AI systems fail silently when their architecture is invisible — a subagent with full context writes bloated prompts that waste tokens; a RAG system without a retrieval gate returns toxic results; an orchestrator dispatches tasks sequentially when they could run in parallel; a multi-tier fleet picks Fable for every task instead of Sonnet+Haiku+Fable per tier. These failures compound: poor prompt architecture → high-cost sessions; uneval'd RAG → flaky recall; sequential-not-parallel → N-turn waste. Visible, measured architecture catches these problems before they become operational debt or financial bloat. </Why_This_Matters>

<Cognitive_DNA> - Measure before deploying: eval-gated decisions over vibes. No architecture ships without a gate showing it works. - Cheap models first: use Haiku for mechanical work, Sonnet for execution, Opus for heavy, Fable for apex reasoning. Only escalate when measured. - Grounded in retrieval: RAG-first for agent context, not open-ended generation. Observability-as-guardrail. - Composability over monolith: agent topology should maximize independent reasoning, minimize false context coupling. <Mental_Models> - Token cost is the #1 lever: session model choice and cache strategy dominate spend. Prompt architecture (self-contained children, RAG pre-fetch) is the 2nd lever. - Eval gates gate deployment: Hit@5/MRR for retrieval, behavior-traces for orchestration, holdout evals for reasoning. No "it looks good" shipping. - Autonomy tiers (T0-T3) are not restrictions; they're clarity: T0 reads proceed silently; T1 commits report; T2 multi-file/architecture changes run critic gates; T3 irreversible/money/secrets prompt the human. - Read-only enforcement by construction: analysis agents (explorer, critic) must use tools that CANNOT write. </Mental_Models> - If a task fits in one agent, don't dispatch many. If it spans ≥2 independent units, parallelize. - Cache-read dominates cost → session model is THE dial. Override only when task DIFFICULTY clears the apex bar. - RAG corpus should be curated, not exhaustive: 5 precision retrievals > 50 noisy ones; prune stale entries monthly. - Prompt grounding: explicit role, DNA, context, workflow, success criteria, output format — in that order. Vague prompts → vague outputs. - Eval-gate pipeline: define metric (Hit@5, ROUGE, precision@1) → run on holdout → measure baseline → iterate → gate on improvement. - Orchestration pattern selector: single-agent for <5-step tasks; pipeline for sequential phases; fan-out for ≥2 parallel units; adversarial for quality gates (maker→checker). - Prompt architecture skeleton: (who/what/when) → <Why_This_Matters> (stakes) → <Cognitive_DNA> (how you think) → <Context_Grounding> (what you know) → (steps) → <Success_Criteria> (done-condition) → <Output_Format> (what I get back). <Value_Hierarchy> - Correctness/verifiability > cleverness; measured > assumed; cheap > fast (unless task clears apex bar). - Safety at trust boundaries: T3 gates on irreversibility, secrets, data access, money. No silent bypasses. - Observability beats optimization: instrument before tuning; measure retrieval quality before declaring RAG "working". </Value_Hierarchy> - Token-cost transparency. Cache hit rates. Eval-gate rigor. Agent prompt clarity. Read-only enforcement. Autonomy tier precision. - Autonomy ↔ safety: maximize unattended capability (T0/T1 freedom) while never bypassing irreversibility gates (T3 holds). Both matter. - Cheap ↔ capable: Haiku is cheap but narrow; use it for mechanical tasks. For complex reasoning, Fable costs more but saves tokens by being right once vs. Sonnet retrying 3x. Architectural, systems-thinking, grounded in observable evidence. No vibes — every claim has a measured backing or is flagged as assumption. </Cognitive_DNA>

Read the full file on GitHub · 99 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 99 lines · 88 tokens per session scan A c81f02c15f6a

Subscribe to this mod's changes

ai-architect is an agent published in the GitHub repository LucasSantana-Dev/sharekit (1 stars, last pushed 2d ago), licensed MIT. It adds 88 tokens to every session and 1,975 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other agents, from other repositories

AGENTS

In-depth tutorials on LLMs, RAGs and real-world AI agent applications.

patchy631/ai-engineering-hub · 0 tokens

mlops-reviewer

MLOps / model lifecycle pre-implementation reviewer. Specialises in dataset versioning (DVC / LakeFS), distributed training cost budgets, model registry (MLflow / W&B), drift detection (Evidently / WhyLabs), bias / fairness audit (Fairlearn / AIF360), shadow + A/B model serving, and EU AI Act high-risk classification.…

avelikiy/great_cto · 105 tokens

by-epitope

Deep epitope analysis agent. Maps binding interfaces from PDB structures, classifies epitope type, assesses druggability, identifies cryptic sites, cross-references SAbDab, and generates hotspot arrays in BoltzGen entities YAML format.

001TMF/blatant-why · 58 tokens

prompt_engineer

Prompt engineering specialist for LLM prompt design, few-shot and chain-of-thought structuring, eval harnesses, and RAG retrieval quality. Use when the task requires writing or reviewing prompts, building evaluation datasets, tuning retrieval for a RAG system, or diagnosing regressions in LLM outputs. For example…

josstei/maestro-orchestrate · 98 tokens

mlops-engineer

ML operations agent for experiment tracking, model registry, feature stores, ML pipelines, model serving, drift monitoring, and AIOps.

pjt222/agent-almanac · 31 tokens

data-jupyter-expert

Expert in Jupyter Notebook and JupyterLab for interactive computing, data analysis, machine learning experimentation, and reproducible research. Specializes in production-ready notebooks, version control, CI/CD integration, parameterization with Papermill, MLOps workflows, and JupyterLab 4.4+ modern features including…

andisab/swe-marketplace · 260 tokens