New Skills — July 26, 2026 (Update #2)¶
Summary¶
| Metric | Count |
|---|---|
| New publishers found | 3 |
| Setup guides created | 3 |
| Combined installs | ~43,000+ |
| Combined GitHub stars | 466 ⭐ |
| Quality: 🟢 Production | 3 |
| Quality: 🟡 Beta | 0 |
| Quality: 🔵 Community | 0 |
New Skills¶
Observability / Monitoring¶
| Skill | Publisher | Installs | Stars | Tier | Guide |
|---|---|---|---|---|---|
| Datadog Agent Skills | datadog-labs/agent-skills | 12K+ | 146⭐ | 🟢 | Setup Guide |
Feature Management / Experimentation¶
| Skill | Publisher | Installs | Stars | Tier | Guide |
|---|---|---|---|---|---|
| LaunchDarkly Agent Skills | launchdarkly/ai-tooling | 26K+ | 20⭐ | 🟢 | Setup Guide |
Platform / Deployment¶
| Skill | Publisher | Installs | Stars | Tier | Guide |
|---|---|---|---|---|---|
| Railway Agent Skills | railwayapp/railway-skills | 5.5K+ | 300⭐ | 🟢 | Setup Guide |
🔑 Standout Finds¶
datadog-labs/agent-skills (12K+ installs, 146⭐)¶
The most comprehensive observability skills suite on skills.sh with 20+ skills spanning logs, monitors, APM, LLM observability, CI/CD, and audit. The Agent Observability (LLMO) skills are uniquely positioned — Datadog is the first major observability platform to ship agent-native skills for evaluating and improving LLM-powered agents. The 8-skill eval pipeline (classify → RCA → bootstrap evaluators → dataset → experiment → analyze) is production-grade agent quality infrastructure.
launchdarkly/ai-tooling (26K+ installs, 20⭐)¶
The highest-install feature management skills on skills.sh with 25+ skills across feature flags, experiments, metrics, and AgentControl (LLM prompt management). AgentControl is the standout: it treats LLM prompts as feature-flagged configurations, enabling A/B testing of system prompts, multi-agent graph orchestration with routing, and online eval attachment. This is LaunchDarkly applied to AI agents themselves — a category-defining approach.
railwayapp/railway-skills (5.5K+ installs, 300⭐)¶
The highest-starred deployment platform skills outside of the major cloud providers. Railway's one-command agent setup (curl -fsSL agents.railway.com | sh) is the smoothest onboarding of any platform skill — it installs skills, configures MCP, and verifies auth in a single step. The multi-plugin packaging (Claude Code, Codex, Grok Build, Cursor) makes Railway the most broadly compatible deployment skill.
Other Highlights¶
- Datadog LLMO skills fill a critical gap. No other observability platform offers agent-native LLM evaluation tools. The
agent-observability-eval-pipelineskill provides an 8-phase feedback loop from failure detection to experiment analysis — directly applicable to Hermes agents running in production. - LaunchDarkly AgentControl is category-defining. Feature-flagging LLM prompts, managing multi-agent routing graphs, and attaching evals to configs is a new paradigm. Other feature flag platforms (Split, Optimizely, CloudBees) have no equivalent agent skills.
- Railway's plugin packaging is the gold standard. The
claude-plugins-officiallisting, Cursor Marketplace presence, and Codex plugin manifest make Railway the most broadly accessible deployment skill across agent platforms.
Discovery Method¶
DevOps/observability platform sweep: following the July 26 enterprise sweep (Cloudflare, AWS, Google, HashiCorp, MongoDB, Databricks) and design/DB update (Neon, Anthropic, Figma, Grafana, PlanetScale), this third sweep targeted observability, feature management, and deployment platforms. Used curl + skills.sh /api/search endpoint with owner-scoped queries across 30+ additional publishers. Cross-referenced against 321 existing catalog entries. Confirmed 3 new publishers not previously catalogued.
Publishers checked but deprioritized: Kong (379 inst, membranedev wrapper), Splunk (546 inst, wrapper), New Relic (226 inst, wrapper), PagerDuty (275 inst, wrapper), Sentry (already catalogued via other publishers), WorkOS (1.2K inst, already catalogued), Upstash (2.1K inst, already catalogued).
Notes¶
- This is the third marketplace sweep of July 26 — following the morning enterprise sweep (6 publishers, 146K+ installs) and the afternoon design/DB update (5 publishers, 190K+ installs). Combined: 14 new publishers, ~379K+ installs found today.
- datadog-labs/agent-skills at 12K installs is the first major observability platform with agent-native skills — Grafana (16K) was catalogued in the previous update but Datadog's LLMO skills are uniquely comprehensive.
- launchdarkly/ai-tooling at 26K installs makes LaunchDarkly the highest-install feature management platform on skills.sh. The repo is
launchdarkly/ai-toolingbut skills.sh lists it aslaunchdarkly/agent-skills. - railwayapp/railway-skills at 300 GitHub stars has the highest star count of any deployment platform skills repo — higher than Cloudflare (2,481 in total but distributed across repos). Railway's community engagement is notably strong.
- 324 catalog entries now exist (up from 321 after this sweep).
- Firecrawl was unavailable for web extraction — all skill content was fetched via raw
curlto GitHub raw content URLs.