Skills I use in every coding agent: Claude Code, Codex and pi, on every machine.
| Skill | What it does |
|---|---|
| chains | Markdown handoffs in .chains/ that survive sessions, compaction and agent switches. Ships chain.mjs (Node, no dependencies). |
| hc | Run servers, builds and other agents through herdr-core, and reach other machines. |
| hc-flow | Multi-agent work as real pi, Claude Code and Codex sessions in hc terminals: by hand, or scripted with flow.mjs (agent/parallel/pipeline, Node, no dependencies). |
| diagnose | A debugging loop: reproduce, hypothesize, probe, fix with a regression check. |
| codebase-orientation | Map an unfamiliar code area just enough to act safely. |
| validation-review | A bounded test-and-review pass with a ship verdict. |
| concept-diagrams | Architecture, flow and state diagrams grounded in the code. |
| datadog-pup | Read Datadog safely with the pup CLI. |
npx skills add DeevsDeevs/agent-system -g -a claude-code -a codex -a pi
npx skills update -gClaude Code reads ~/.claude/skills, Codex and pi read ~/.agents/skills.
The old plugins (dev-experts, venue-expert, mft-research-experts and the rest) are on the
legacy branch.
node --test test/