원클릭으로 Manus에서 모든 스킬 실행

agent-ecosystem-watch

스타6

포크1

업데이트2026년 7월 24일 20:13

Agent-Ecosystem Watch — a daily autonomous intelligence pass over the Claude and AI-agent ecosystem on X. Reads a topic watchlist via last30days/ScrapeCreators, extracts CONCRETE improvements that make Claude / Claude Code / OmegaOS better (new features, prompting techniques, MCP servers, subagent/orchestration tricks, hooks, eval methods, model updates), scores each for integratability into OmegaOS, writes an HTML report + Telegram alert for the operator to decide, and (only when ARMED, behind an adversarial quality gate + daily cap + dedup) auto-publishes vetted best-practice tweets to @Agentik_os via omega-zernio. Use when the user says '/omg-ecosystem-watch', '/ecosystem-watch', 'veille Claude', 'veille agents IA', 'watch the Claude ecosystem', 'run the ecosystem watch', or wants to inspect/arm/disarm the daily watch. NOT a live web research primitive for one question (that is last30days / deep-research) and NOT a general social scheduler (that is the marketing suite): this is the standing daily radar tha

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

agentik-os

agentik-os/OmegaOS

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

관련 직업SOC

SOC 직업 분류 기준

기타 컴퓨터 관련 직업컴퓨터 및 수학직·SOC 15-1299

파일 탐색기

10 개 파일

SKILL.md

readonly

이 저장소의 다른 Skills

같은 저장소

changelog-adopt

agentik-os/OmegaOS

Claude Changelog Adopt — a daily autonomous self-improvement pass that reads the OFFICIAL Claude Code changelog (the authoritative GitHub CHANGELOG for anthropics/claude-code, with @claudecodelog as a secondary mirror), diffs it against the last version OmegaOS already absorbed, and for each NEW entry decides whether it implies a concrete OmegaOS or system-agent improvement — a new primitive, a doctrine change, an agent-behavior shift, a deprecation to unwind. It classifies each entry (opus), proposes a surgical adoption (what to change, in which file, why — like the manual /loop -> R-LOOP native-loop doctrine move), runs an adversarial anti-fabrication gate (opus, default-reject), writes a self-contained HTML report + Telegram alert for the operator to decide, and — only when ARMED — dispatches each vetted proposal to an oracle behind the full quality gate (In-Review handoff, never auto-Done, never force-push). Auto-adoption is scoped to doctrine (rules.rs), agent identities (agents/*.md), skills, and instal

2026-07-246

duo

agentik-os/OmegaOS

Binome Claude (stratège) ⇄ Codex/Sol (coder). Claude planifie, Codex challenge la stratégie puis implémente, Claude relit le diff et rend un verdict — avec bascule automatique sur Claude si le quota Codex est épuisé. Use when the user types /duo, /omg-duo, or says: build/implement/refactor/fix with the binome, code this with Codex, delegate to Codex/Sol, pair Claude and Codex, or FR 'code ça avec Codex', 'délègue à Sol', 'binome Claude Codex', 'fais coder Codex et relis'. NOT for pure questions, reads, or one-liners (Claude handles those directly — no need to burn the Codex quota).

2026-07-246

growth-engine

agentik-os/OmegaOS

Growth Engine — a bounded, gated X growth system for @Agentik_os aimed at real follower growth (north star 10k) by being genuinely useful in the right conversations, never by spam automation. A daily Reply Radar finds high-leverage AI-agents/Claude threads via last30days/ScrapeCreators, drafts genuine on-voice replies, passes them through an adversarial anti-spam gate, then (only when ARMED with a valid X session) posts a LOW-CAPPED number of replies + likes via Playwright driving the operator's logged-in session, with human pacing, dedup, and a kill-switch. Use when the user says '/omg-growth-engine', '/growth-engine', 'growth X', 'gagner des followers', 'reply radar', 'engager sur X', 'arm/disarm the growth engine'. NOT a mass like/follow/DM bot (that gets the account suspended and is refused), NOT the content publisher (that is agent-ecosystem-watch): this is the engagement layer, deliberately small-volume and quality-gated.

2026-07-246

llm-council

agentik-os/OmegaOS

Convene a multi-model Claude COUNCIL for a high-stakes, ambiguous, or irreversible decision — or whenever you want a genuine second opinion. Four different Claude models (Opus 5, Sonnet 4.6, Haiku 4.5, Fable 5) answer the SAME question independently and in parallel, then peer-review each other ANONYMOUSLY (answers shown as Response A/B/C, model identity hidden, so no model is favored by name), and an Opus 5 president synthesizes the final verdict WHILE surfacing the dissent. Inspired by Karpathy's LLM Council, re-implemented on the OmegaOS Workflow primitive. 100% Claude Code-native — NO API KEYS, no OpenRouter, no external provider, no extra cost: the members run as in-process Workflow sub-agents through your EXISTING Claude Code session. Use when the user says "llm council", "convene the council", "ask the council", "@council", "/llm-council", "second opinion", "deliberate", "multiple models", or faces a "high-stakes decision" / architecture choice / contradictory verification verdicts / a pre-commit design

2026-07-246

stax

agentik-os/OmegaOS

Convert ANY project fully to Stax — the panels-inside-panels dashboard grammar (Miller columns, evolved): one mechanic (open-right), one action zone (the footer), one way back (close / breadcrumb). No pages, no modals, no tabs — the whole UI derives from a single serializable panel-workspace state (URL-synced, pinnable references, per-type sizing). Use when the user says "/stax", "/omg-stax", "convert to stax", "panelize this app", "make this a panel stack", "stax layout", "miller columns", "panels inside panels", "dashboard vision structure", or in French "convertis en stax", "passe en panneaux", "structure de dashboard". This is the OmegaOS dashboard vision structure — tracks github.com/agentik-os/stax main.

2026-07-236

linear-setup

agentik-os/OmegaOS

Complete Linear feedback system setup wizard (v2.1). Installs widget with priority selector + page-aware Scope labels, API routes with hardcoded label IDs and default CTO assignee, Linear label GROUPS (FEEDBACK / SOURCE / optional AGENTS), public feedback page, and MCP integration. Auto-detects dashboard pages from sidebar to generate Scope labels. Use when user says "/linear-setup", "setup linear", "add feedback system", or "install Linear widget". One-time setup. After setup, use /linear-fix to resolve tickets.

2026-07-206

name

agent-ecosystem-watch

description

Agent-Ecosystem Watch

A standing daily radar on the thing that compounds OmegaOS's edge: how fast you absorb new Claude and agent techniques. It watches X for concrete improvements, tells you what is worth wiring into OmegaOS, and shares the best of it on @Agentik_os on autopilot, behind a gate so your public account never ships a weak or wrong tweet.

Same input feeds three outputs: (1) a decision report for you, (2) a curated best-practice tweet feed, (3) a growing memory of what the ecosystem shipped. Marginal cost is near zero once built; the read is a few dollars a month of ScrapeCreators.

Pipeline

cron (daily) → [1] READ  watchlist topics → last30days --search x (ScrapeCreators)
             → [2] ANALYZE (claude, sonnet)  extract + score integratable_to_omega 0-10
             → [3] GATE (claude, opus)  adversarial: try to REJECT each candidate tweet
             → [4] PUBLISH (if ARMED, not mock, under cap, unseen)  omega-zernio → @Agentik_os
             → [5] REPORT  self-contained HTML in ~/.omega/artifacts/
             → [6] ALERT   Telegram (Alerts topic) via omega-alert-send.sh

Deterministic control (daily cap, dedup persistence, publish side-effects) lives in shell; judgment (extract, score, draft, refute) lives in the model. The gate is an INDEPENDENT adversarial pass (R-VERIFY): its only job is to reject, default-reject on doubt.

Publishing safety rails (non-negotiable)

Publishing to the public @Agentik_os account requires ALL of:

armed flag present (~/.omega/state/ecosystem-watch/armed) — the kill-switch. Absent = tweets are drafts in the report only. rm the flag to stop all publishing instantly.
a real ScrapeCreators key (mock runs never publish).
the adversarial gate returned keep=true for that tweet.
the tweet fingerprint is not already in the seen-store (never tweet the same idea twice).
the daily cap (EW_DAILY_TWEET_CAP, default 5) is not exhausted.
no em/en dash, <= 280 chars, carries its source URL (enforced in shell too, R-NODASH).

Integration of a finding into OmegaOS itself is always a human call: the report proposes with a 0-10 integratability score, the operator decides.

Run

bash scripts/watch-run.sh                    # full run (reads config/topics.txt)
EW_MOCK=1 bash scripts/watch-run.sh          # fixtures, never publishes (smoke test)
EW_ZERNIO_DRYRUN=1 bash scripts/watch-run.sh # validate zernio post without publishing

Env knobs: EW_DAILY_TWEET_CAP (5), EW_DAYS (2), EW_ANALYZE_MODEL (claude-sonnet-5), EW_GATE_MODEL (claude-opus-5), EW_ZERNIO_PROJECT (agentik-os), EW_TOPICS_FILE.

Arm / disarm

touch ~/.omega/state/ecosystem-watch/armed    # go fully auto (auto-publish)
rm    ~/.omega/state/ecosystem-watch/armed     # kill-switch: drafts only

Config

config/topics.txt — one X query per line (broad AI-agent ecosystem, Claude-first).
config/accounts.txt — reference builders, an authority hint for scoring.

State (outside repo, R-ENV)

~/.omega/state/ecosystem-watch/seen.jsonl — published fingerprints (dedup memory).
~/.omega/state/ecosystem-watch/runs/<date>/ — per-run raw pulls, analysis, report.
~/.omega/artifacts/ecosystem-watch-<date>.html — the daily report.
~/.omega/logs/ecosystem-watch.log — run log.

Cron

# 08:00 daily — disarmed by default until the operator arms it
0 8 * * * OMEGA_DIR=$HOME/.omega bash $HOME/.omega/skills/agent-ecosystem-watch/scripts/watch-run.sh

Requires

~/.omega/secrets/integrations.env with SCRAPECREATORS_API_KEY (X read) and ZERNIO_API_KEY (publish). Both already present in this deployment.
omega-zernio with an agentik-os profile whose twitter account is connected.
last30days skill checkout at ~/.omega/repos/last30days-skill/.
claude CLI on PATH.