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/lazymac2x/agent-wiki/_agent_specgit clone --depth 1 https://github.com/lazymac2x/agent-wikiWhat 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 | $0.00044 | $0.01926 |
| Opus 5 | $0.00022 | $0.00963 |
| Sonnet 5 | $0.00009 | $0.00385 |
| Haiku 4.5 | $0.00004 | $0.00193 |
Grade A, and why
_AGENT_SPEC 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.
How it starts
The opening of the file, as written. The whole thing — 93 lines — stays where its author put it; the contents beside it link to each section on GitHub.
_AGENT_SPEC — 코어 에이전트 카드 계약
agents/*.agent.md한 장 = 한 코어 에이전트. 상단·프로젝트단위·비즈로직 보유. wiki는 JIT로 끌어다 쓴다. 이 파일은 모든 카드가 따르는 메타 계약(reference). "왜 이런 구조인가"의 서사는 [[core-agent]] 패턴이 갖는다(L4 분리). 카드는 이 계약을 복붙하지 말고[[_AGENT_SPEC]]로 참조한다(L1 단일 SOT).
1. 카드란 무엇인가
- 코어 에이전트 = 사람이 말 거는 진입점. 도메인 비즈로직(불변식·게이트·소유 데이터흐름)을 자기 카드에 갖고, 일반 지식은 wiki에서 JIT.
- dual-home 금지: 외부코드 프로젝트 =
agents/<name>.agent.md(브레인+포인터) +wiki/domain(공유지식). from-scratch =projects/<name>/. 카드는 항상 한 곳. - 카드는 얇게 — 디테일은 자식노트로 쪼개고
[[link]]. 임계(200줄/25KB) 전에 compaction-on-write. → [[token-budget]]
2. 카드 필수 섹션 (이 순서)
- 역할 & 소유 비즈로직 — 이 에이전트만 아는 불변식/게이트(예: chokepoint, 서킷브레이커).
- 카드 메타(선언) — 아래 yaml 블록.
- wiki_access — 접근 허용 노트를
[[link]]prose로 (allowlist = 항해 경계). - self-heal / HITL 훅 — 런타임 이상수복·비가역 게이트.
- skill 브리지 — 재구현 금지, 당신 환경의 기존 스킬/루틴 호출(에이전트 메모리 경로).
# 카드 메타 선언 (오케스트레이터가 파싱) — 값은 카드마다 다름
model: opus # 티어: opus=오케스트레이터 · haiku=워커/1차스크리닝 · 타패밀리=judge → [[model-tiering]]
preload: [self_card, project_llms] # 사전적재는 이 둘만. 나머지 전부 JIT
loop: pdca | ooda # 빌드/기능=pdca · 런타임이상수복=ooda
gate: rubric7 ≥90, P0/P1=0 # 승격 게이트 → [[rubric]] [[evaluator-optimizer]]
hitl: [prod_write, irreversible, conf<0.85] # AskUserQuestion 의무
3. 사전적재 vs JIT (지연로딩 — 핵심)
- 사전적재(preload) = 딱 둘: ① 자기 카드 1장 ②
projects/<name>/llms.txt(있으면). 끝. - 그 외 모든 wiki 본문은 JIT.
@import/전체 wiki Read = 런치 때 전량 펼쳐짐 = 토큰절약 아님(L7). 가벼운 식별자(slug/경로/쿼리)만 컨텍스트에 둔다. - 검색 디테일은 sub-agent에 격리하고 1~2k 토큰 요약만 회수(메인 컨텍스트 오염 방지).
4. JIT 검색계약 — search → get → reflect (depth≤3)
frontmatter가 신호다. 본문 읽기 전에 frontmatter만 필터(progressive disclosure).
# (search) frontmatter 라인만 ripgrep — 후보 slug 수집. 본문 안 읽음
rg -l -e '^description:.*<키워드>' -e '^type: <원하는타입>' wiki/ harness/ projects/
# (get) 후보의 frontmatter 헤더만 미리보기 → 관련성 판정
rg -m1 -A10 '^---$' <후보.md> # 또는 Read(limit=12)
- search — frontmatter(name/description/type/track) 키워드 매치로 후보 slug 목록만.
- get — 후보의 frontmatter만 먼저 본다. 통과한 1~2개만 본문 Read.
- reflect — 답에 부족하면 그 노트의
[[link]]1홉 따라간다. 최대 3홉(depth≤3). 그 안에 못 찾으면 "모른다"로 정직 종료(가짜 답 금지).
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 · 93 lines · 0 tokens per session scan A efa408b4f569
_AGENT_SPEC is an agent published in the GitHub repository lazymac2x/agent-wiki (43 stars, last pushed 2mo ago), licensed MIT. It adds 44 tokens to every session and 1,926 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 2026-08-30.
Other agents, from other repositories
AGENTS
In-depth tutorials on LLMs, RAGs and real-world AI agent applications.
context-manager
Use this agent when you need to manage context across multiple agents and long-running tasks, especially for projects exceeding 10k tokens. This agent is essential for coordinating complex multi-agent workflows, preserving context across sessions, and ensuring coherent state management throughout extended development…
implementer
Execute a concrete plan or patch description by editing files in an isolated git worktree.
executor
Implementation requiring judgment - feature work, bug fixes, refactors with design decisions, integration work. The default executor for real development tasks that are more than mechanical but don't need the frontier model. Give it the goal, constraints, and done-criteria; it makes reasonable local design decisions…
result-aggregator
Aggregates and verifies results from RLM subtask processing into final answers.
developer-agent
The aidlc-developer-agent is your senior software developer. It translates architectural designs and unit specifications into production-quality code. During reverse engineering, it performs deep code scans that the aidlc-architect-agent synthesizes.