Project-AMBER: Skill for Claude Code

.github/skills/sage/SKILL.md

sage is a skill for Claude Code, Codex from JJHbrams/Project-AMBER. It costs 120 tokens per session (1,494 once invoked), scanned A, original, MIT.

A skill for sending one unusually difficult question to a top-level provider model and returning its advice to the main agent. It is intended for hard design choices, persistent bugs, cross-system failures, or computer-control tasks, not routine work.

In plain words
What is it for?
Use it to prepare a focused brief and request a one-time read-only judgment, plan, or root-cause analysis.
Why use it?
It gives the main agent an outside judgment when ordinary debugging or self-review has not resolved an important problem.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions Claude Code; runs codex exec; mentions Codex.

This is JJHbrams/Project-AMBER's own configuration. It tells Claude Code and Codex how to work on Project-AMBER itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything Project-AMBER configures →

Reuse

Borrowing it

Nothing to install: this file belongs to JJHbrams/Project-AMBER. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/JJHbrams/Project-AMBER/master/.github/skills/sage/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/JJHbrams/Project-AMBER

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for sage

README.md
[![agentmods](https://agentmods.dev/badge/skills/jjhbrams/project-amber/sage/github.svg)](https://agentmods.dev/skills/jjhbrams/project-amber/sage)
Your own site
<a href="https://agentmods.dev/skills/jjhbrams/project-amber/sage"><img src="https://agentmods.dev/badge/skills/jjhbrams/project-amber/sage/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for sage

Your own site · 80×15
<a href="https://agentmods.dev/skills/jjhbrams/project-amber/sage"><img src="https://agentmods.dev/badge/skills/jjhbrams/project-amber/sage.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 120 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,494 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.00120 $0.01494
Opus 5.5 $0.00048 $0.00598
Sonnet 5.5 $0.00024 $0.00299
Haiku 4.5 $0.00012 $0.00149

Measured 1mo ago against content hash 56910eeff44d, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

sage 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 1mo 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.

.github/skills/sage/SKILL.md · 112 lines

How it starts

The opening of the file, as written. The whole thing — 112 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Sage (현자)

Orchestrator 는 계속 중간 티어(Opus / Sol)로 돈다. 현자는 상시 티어가 아니라 예외 호출이다. 한 번의 호출로 끝낼 질문 하나를 들려 보내고, 답을 받아 orchestrator 가 실행한다.

Phase 0 — 승급 판정

다음 중 하나라도 해당하면 승급한다.

승급 사유 예
설계 갈림길 되돌리기 비싼 구조 결정, 상호 배타적 아키텍처 후보
난치 원인 규명 통상 디버깅·로그·테스트로 2회 이상 못 잡은 버그, 재현이 불안정한 결함
교차 시스템 추론 여러 런타임·프로세스·경계를 동시에 물고 있는 실패
판단 검증 orchestrator 자신의 계획을 스스로 검증할 수 없을 때
PC 제어 화면·GUI·데스크톱 조작이 필요한 작업 (Phase 3)

다음은 승급하지 않는다. 그냥 평소대로 처리한다.

  • 파일 몇 개 수정, 리팩토링, 테스트 추가
  • 코드 위치 찾기·읽기·설명
  • 이미 원인을 아는 버그의 수정
  • "어렵다"가 아니라 "길다"인 작업 — 그건 분할 대상이지 승급 대상이 아니다

승급 판정이 애매하면 승급하지 않는다. 현자는 비싸다.

Phase 1 — 브리프 작성

현자는 이 세션의 맥락을 모른다. 호출 하나로 닫히는 자립 브리프를 만든다.

[SAGE_BRIEF]
question: <현자가 답해야 할 단 하나의 질문>
why_escalated: <위 표의 승급 사유 + 이미 시도해서 실패한 것>
context: <필요한 배경 사실만. 추측과 사실을 구분해 적는다>
artifacts: <읽어야 할 파일 경로, 로그 발췌, 실패한 명령과 그 출력>
constraints: <바꾸면 안 되는 것, 이미 확정된 결정, 런타임·플랫폼 제약>
deliverable: <판단만 | 판단+계획 | 원인규명>

question 이 둘 이상이면 브리프를 쪼개지 말고 가장 비싼 질문 하나만 남긴다. 나머지는 그 답이 나온 뒤 orchestrator 가 직접 처리한다.

Phase 2 — 자문 호출 (읽기 전용, 기본 모드)

세션이 돌고 있는 provider 의 최상위 모델로 대칭 승급한다.

세션 호출 방법 모델
Claude Code sage 서브에이전트 Fable
Codex sage 서브에이전트 GPT-6-Astra (model_reasoning_effort = high)
Copilot CLI sage 서브에이전트 해당 CLI 최상위

현재 세션에 sage 서브에이전트가 없으면 Codex 로 직접 넘긴다.

codex exec -m gpt-6-astra -c model_reasoning_effort=high -s read-only "<[SAGE_BRIEF] 전문>"

현자는 읽기 전용이다. 파일을 고치거나 상태를 바꾸는 명령을 실행하지 않는다. 반드시 [SAGE] 형식(verdict / reasoning / evidence / risks / plan / confidence / unknowns)으로 답을 받는다.

Phase 3 — 제어 모드 (PC 제어가 필요할 때만)

화면·GUI·데스크톱 조작은 Codex 의 computer_use 로만 수행한다. 이 경로는 샌드박스 밖에서 실제 기계를 건드리므로 사용자의 명시적 승인 없이는 실행하지 않는다.

승인 요청 시 다음을 먼저 보여준다.

[SAGE_ACT]
goal: <조작으로 달성할 것>
target: <대상 앱·창·경로>
actions: <수행할 조작의 순서>
blast_radius: <잘못되면 무엇이 바뀌는가 / 되돌릴 수 있는가>

사용자가 승인한 뒤에만 실행한다.

codex exec -m gpt-6-astra -c model_reasoning_effort=high \
  --enable computer_use -s workspace-write "<[SAGE_ACT] + 조작 지시>"

Read the full file on GitHub · 112 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. 1mo ago First seen · 112 lines · 120 tokens per session scan A 56910eeff44d

Subscribe to this mod's changes

sage is a skill published in the GitHub repository JJHbrams/Project-AMBER (2 stars, last pushed 14d ago), licensed MIT. It adds 120 tokens to every session and 1,494 once invoked, about $0.0005 per session on Opus 5.5. 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 2026-09-13.

Related

Other skills, from other repositories

penfield

Persistent memory for AI agents. Store decisions, preferences, and context that survive across sessions. Build knowledge graphs that compound over time. Hybrid search (BM25 + vector + graph) recalls what matters when you need it.

penfieldlabs/penfield-mcp · 47 tokens

memory-audit-discoverability

A review guide for checking whether stored memories can be found at the right time. It focuses on where memories are attached, when they are triggered, whether aliases are missing, and whether a parent has too many children.

Dataojitori/nocturne_memory · 35 tokens

memory-audit-pattern-extraction

A method for investigating repeated mistakes by comparing related memories and checking whether an earlier reminder failed. It looks at where the reminder was stored, when it was created, and whether it was strong enough to prevent the mistake.

Dataojitori/nocturne_memory · 48 tokens

memory-audit

An entry point for reviewing and maintaining an AI agent's stored memories. It describes how to remove repetition, preserve useful reasoning, and update memories when old conclusions no longer fit.

Dataojitori/nocturne_memory · 37 tokens

add-connector

Onboard a deus connect connector — a way to route Claude Code sessions to non-Claude models (e.g. GPT via CLIProxyAPI) alongside normal Claude access. Generic orchestrator over whichever connector is selected.

sliamh11/Deus · 49 tokens

mengram-memory

Long-term memory with 3 types (facts, events, workflows). Remember user preferences, past conversations, and learned procedures across sessions. Use when recalling what the user said before, saving important info, getting user context, or tracking completed workflows.

alibaizhanov/mengram · 53 tokens