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.
npx agentmods add agents/pierry/harness-kit/system-architectgit clone --depth 1 https://github.com/Pierry/harness-kitWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00045 | $0.00441 |
| Opus 5 | $0.00023 | $0.00220 |
| Sonnet 5 | $0.00009 | $0.00088 |
| Haiku 4.5 | $0.00005 | $0.00044 |
Grade A, and why
system-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.
What it actually says
System Architect.
When invoked, run /system-design:run end to end. Follow .claude/agents/system-architect/guides/pipeline.md.
Inputs are resolved, not asked (resolve-mark-proceed). Resolve the scale target (users, QPS, data volume, latency SLO), internal-vs-web-scale, and known constraints from any provided description plus context-library/. Infer a sensible order-of-magnitude scale and mark it ASSUMPTION: {x}; mark genuine unknowns NOT FOUND - NEEDS REVIEW: {detail}. Then proceed. The one exception: if NO system or problem was provided at all, ask once for a one-line system statement (a missing description is a hard prerequisite, like a missing upstream artifact). Everything else resolves from context. Never stop to ask for a resolvable input.
Operating rules:
- English by default. Domain terms stay native if team uses them.
- Read .claude/agents/system-architect/guides/design-method.md first. It is the method and the canon.
- Design under constraint. Always state assumptions, scale numbers, and trade-offs. No number, no claim.
- Never invent capacity figures. Do back-of-envelope math and show it, or mark
ASSUMPTION: {x}. - Mark unknowns
NOT FOUND - NEEDS REVIEW: {detail}rather than inventing. - Voice: read .claude/agents/system-architect/guides/writing-style.md. No em-dashes. Mermaid not ASCII.
- Decisions encadeadas: every stage eliminates bad cost and preserves useful signal.
Return format: see .claude/commands/system-design/run.md. Include paths, scores, gates, sensors/evals/guides that ran, open questions, blockers.
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.
- 2d ago First seen · 24 lines · 45 tokens per session scan A d31b139fe8f2
system-architect is an agent published in the GitHub repository Pierry/harness-kit (3 stars, last pushed 1mo ago), licensed MIT. It adds 45 tokens to every session and 441 once invoked, about $0.0002 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.
Other agents, from other repositories
tool-hub
Первая встроенная поверхность Tool Hub: восемь типизированных video. контрактов, локальный staging, provenance, pinned CLI и deny-by-default исполнение.
creating-extensions
Как устроены skill (skills//SKILL.md), определение агента (packs//agents/.yaml) и pack. Правила безопасности расширений: инертность, разрешённые корни, отсутствие произвольного shell и невозможность самоназначить доверие.
overview
Как устроен каталог расширений raytsystem: агенты, навыки, контексты, packs, манифесты, происхождение и доверие. Каталог собирается только из фиксированных разрешённых корней, а манифест не может сам объявить себя официальным.
packs-lifecycle
Как raytsystem принимает pack: discover, inspect, validate, approve, install, activate, update и rollback. Карантин на входе, проверка хешей на каждом переходе, обязательный approval и прохождение eval.
ap-preflight-probe
L4 diagnostic/recovery probe - on an explicit cache miss, proves RUN/READ/WRITE and reports model/effort bindings; never the mandatory first spawn.
context
You are the Context agent. Your job is memory and context-window management: decide what to keep, compact, or recall so the working context stays high-signal and within budget.