Prioritize measurable skill quality and targeted security-review methods over installing another large skill pack.
Source-reviewed discovery. No candidate installed, executed, or benchmarked. Compatibility with Hermes is unverified unless stated. Public report contains generic findings only.
Treat skills as tested artifacts: compare with-skill and baseline runs, grade assertions, review outputs, and separately optimize trigger descriptions.
Verdict: Best method to borrow first. Source includes eval, description-improvement and benchmark scripts. Do not assume its Claude-specific runner works in Hermes.
Repository non-archived; latest default-branch commit 2026-10-09. No GitHub workflow directory found in API tree.Borrow differential-review for security-sensitive diffs and sharp-edges for dangerous APIs/configuration. The first asks for git-history evidence, blast radius and explicit coverage limits; the second probes zero/empty/null values and unsafe defaults.
Verdict: Strongest targeted skill source. Inspect supporting references and tool/agent assumptions before adapting; not a drop-in validated Hermes package. CC-BY-SA-4.0 license matters when redistributing adaptations.
Both SKILL.md bodies inspected. Multi-contributor project; latest default-branch commit 2026-10-23; lint and validation workflows present, results not evaluated.Assert on real agent sessions: skill reads, commands, tool calls, final outputs and token snapshots. Isolated fixtures and repeat runs can detect routing and behavior regressions.
Verdict: Experimental pilot candidate, not a production dependency recommendation. Listed runners cover Claude Code and Codex but not Hermes. Small contributor base and several months since default-branch commit.
Latest default-branch commit 2026-06-15; CI and release workflows present. README and GitHub metadata reviewed; package/runtime not tested.Build tasks as independent packages with environment, skill resources, oracle solution and deterministic verifier. Compare skill and no-skill conditions.
Verdict: Useful reference for evaluation design; heavier than a small personal regression suite. Default cloud execution and credential-dependent extra tasks need separate cost/auth review. No benchmark scores reproduced.
Multi-contributor project; latest default-branch commit 2026-10-13; CI workflows present. README documents local Docker alternative and task packages.Pair each skill with explicit acceptance criteria and scenario fixtures; use triggers and anti-triggers. Keep a routing layer separate from SDK-specific details.
Verdict: Borrow evaluation and organization patterns, not the Azure catalog wholesale. Source warns against loading all skills. Formatting scores do not prove task quality.
Latest default-branch commit 2026-10-23; skill-evaluation and test-harness workflows present. README includes scenarios and acceptance-criteria structure.Capture atomic evidence-backed lessons, keep them project-scoped, and promote selected repeated patterns into reusable skills.
Verdict: Architecture inspiration only. Do not add a parallel memory or auto-learning runtime by default. Confidence scores and absence of user correction are not proof; privacy, hook reliability and promotion behavior were not independently tested.
Skill body inspected. ECC repository latest default-branch commit 2026-10-26; many contributors and CI workflows. README reliability claims are unverified.Separate the execution core, tools/providers, agent roles and composable behavior bundles. A focused behavior need not replace the whole root configuration.
Verdict: Read for architectural ideas, not migration. Project explicitly calls itself an early research demonstrator with incomplete safety systems.
Latest default-branch commit 2026-09-19. README states preview status; no workflow directory found in main repository tree. Other ecosystem repositories not audited.Use a cross-agent discovery directory and inspect individual upstream skill sources. The CLI supports discovery, update checks and one-off use.
Verdict: Discovery source, not another sync owner. Existing Git/symlink management should remain authoritative. Registry presence is not quality endorsement; CLI telemetry is documented.
Latest default-branch commit 2026-10-19; CI/publish workflows present. README reviewed; CLI not run.Superpowers already appears in the available skill catalog; do not present it as new. Its archived predecessor obra/superpowers-skills is not the current repository.
Search plus first-party README/skill inspection; GitHub REST metadata and recursive trees for eight repositories. Workflow presence is not proof that CI passes. No installs, live eval runs, paid services or production dependency changes.