New Skills — August 19, 2026
Morning sweep of August 19. The broad API surface was fully caught up: 40 queries, 3,944 unique skills, and the cluster-level diff against the hermes/ tree returned zero candidates at the 2,000-install and 500-install thresholds — all 120 top clusters already guided. The find came from the supplemental targeted pass: Hermes-ecosystem queries (hermes, openclaw, gbrain, clawd) surfaced one genuinely new publisher cluster that the broad pass ranked below the top band.
New Publisher Clusters — Guided This Sweep
Method Notes
- The cluster-level diff script (full owner/repo grep, generic-name sibling check, 2,000-install floor) returned zero candidates. Lowering the floor to 500 also returned zero — the marketplace's top-120 cluster band is fully catalogued as of the Aug 18 evening sweep.
- The supplemental Hermes-ecosystem pass (10 targeted queries: hermes, openclaw, clawd, gbrain, honcho, nous, claude code skill, cursor rules, agent skill, n8n) flagged 7 candidate sources; 6 were noise or family-known (single-install gbrain forks, a 1-install digital-creature novelty, QuantumNous' new-api gateway skill at 980 installs — LLM infra, not Hermes agent tooling).
openclaw/agent-skills is the official OpenClaw org repository: 1,057 GitHub stars, last commit August 18, 2026 (the day before this sweep). Eight skills across review, handoff, sessions, and documentation families. autoreview leads at 1,388 installs.- Drafted below the 20K bar on official-org authority — the OpenClaw org's own canonical skill suite, matching the wyattowalsh and hono precedent (authority over install volume).
- Mac Mini still unreachable — sweep ran entirely from the Spark canonical clone, push auth via gh CLI token.
Evaluated and Queued
| Cluster |
Skills |
Installs |
Reason parked |
| quantumnous/skills (newapi) |
1 |
980 |
"nous" substring false positive; New API gateway management, not Hermes agent tooling |
| imphillip/gbrain-openclaw |
5 |
5×1 |
Single-install gbrain fork; garrytan/gbrain family already guided |
| laozhong86/gbrain |
1 |
1 |
Single-install gbrain fork |
| morris-utrust/gbrain-knowledge |
1 |
1 |
Single-install TCM niche fork |
| nexus9888/hermes-memory-skills |
1 |
1 |
Single-install dreaming wrapper; Honcho family already guided |
| ad/hermes-digital-creature |
1 |
1 |
Novelty skill, below bar |
Notable Signals for CorpusIQ
- autoreview is the strongest single signal: a structured multi-engine code review workflow (Codex default, Claude/Amp/Pi/Kimi optional) whose default P0-only, advisory output maps directly to our verify-before-assertion discipline.
- handoff + agent-transcript cover the two halves of multi-agent context passing — standalone prompt construction and sanitized PR/issue transcript provenance — both directly relevant to our orchestration and public-content sanitization rules.
- The repo's
scripts/install-skills installer (list, dry-run, selective, copy-vs-symlink, custom target) is a clean reference pattern for skill-suite distribution worth studying for our own pack governance.
Index State After Sweep
- catalog/index.md: +1 entry
- marketplace/index.md: header 902 → 903, footer 953 → 954
- last_updated on both indexes: 2026-08-19