Orchestration agents for the PSDN-AI nexus ecosystem. Each agent chains multiple nexus-skills into end-to-end workflows.
This repository defines a shared AGENTS.md contract plus per-agent knowledge-only packages that define:
AGENTS.md— repo-level best practices and shared invariants for all agents.README.md— human-facing usage and invocation guide.AGENT.md— orchestration instructions: pipeline steps, input/output contracts, constraints, and error handling.config.yaml— declares which skills the agent depends on, with version metadata, refs, and source locations.
Agent packages themselves do not contain executable code. They are consumed by an AI runtime that reads the AGENT.md and follows its workflow, invoking the referenced skills at each step. (The .github/ directory holds repo-level governance workflows, not agent logic.)
| Agent | Description | Skills Used |
|---|---|---|
| prd‑planner | Decomposes a PRD into domain specs, then generates task graphs per domain | prd‑decompose, spec‑plan |
| gha‑planner | Scans a repo's tech stack and generates hardened GitHub Actions workflows | gha‑create |
nexus-skills (atomic capabilities)
|- prd-decompose
|- spec-plan
|- gha-create
|- repo-audit
|- agent-launcher
nexus-agents (orchestration)
|- prd-planner --> chains prd-decompose + spec-plan
|- gha-planner --> wraps gha-create
|- ...
Skills are standalone, single-purpose units. Agents compose skills into multi-step pipelines. During active development, an agent's config.yaml may track a branch ref such as main; for releases, pin each dependency to an immutable tag or commit SHA.
- Read
AGENTS.mdand follow the shared authoring conventions - Create
agents/{agent-name}/README.mdwith human-facing usage and invocation instructions - Create
agents/{agent-name}/AGENT.mdwith YAML frontmatter (name, description, license, metadata) - Create
agents/{agent-name}/config.yamllisting skill dependencies, including areffor each skill (prefer immutable refs for releases) - Add the agent to the table above
MIT