agent-stack

Shape: static. Reference material, not a run. What varies is which reference a task loads, which the SKILL.md table decides in advance.

agent-orchestrator

Use when building an agent system — an orchestrator, an LLM-powered tool, a chatbot with tool use, an AI pipeline, or metering and billing the LLM access it burns. Covers tool-calling loops, pipelines with human checkpoints, provider routing with fallback/retry, memory architecture, retrieval and decay, context budgets, sub-agent coordination, error hierarchies; the work as a graph — parallel layers, fake edges, a checker before convergence; for resale: tiered wallets, one markup boundary, the saga across database and provider API, spend-delta polling, budget and loop guards, per-tenant keys. Not for a single LLM call in a script, or prompt wording.

agent-evals

Use when measuring whether an agent actually works — building an eval suite, judging a trajectory rather than a final answer, turning production traces into regression fixtures, calibrating an LLM judge against human labels, or gating a release on offline evals. Covers the three observability primitives (run, trace, thread) crossed with three eval granularities (single-step, full-turn, multi-turn), the offline/online/ad-hoc timing axis, pass-fail rubrics over scalar scores, cheap code checks before model judges, the checker node as an evaluator inside the graph, simulated users with adversarial personas, annotation queues, and what to instrument for any of it. Not for unit tests of ordinary code, or benchmarking a model.

agent-interop

Use when an agent must talk to something outside its own process — building or consuming an MCP server, exposing or calling an agent over A2A, publishing to the MCP Registry, or putting a gateway in front of agent traffic. Carries the MCP 2026-07-28 wire surface and what it deprecated (server/discover, stateless per-request _meta, elicitation in form and URL mode, subscriptions/listen; sampling, roots, logging and dynamic client registration going), A2A 1.0 agent cards, task states, three bindings, registry namespaces, server.json, tool federation, and what a gateway must do that an API gateway does not. Not for designing one server's tool set, nor for a skill's own construction — that is make-skill.

agent-harness

Use when the question is what the agent is TOLD rather than how its loop is wired — writing or fixing a system prompt, shaping tools so the model picks the right one, deciding whether a job wants a workflow or an agent, or choosing between ReAct, reflection and voting. Also auditing an agent system somebody else built: tracks, evidence tiers and a prioritized plan instead of a score, plus a scanner. Carries Pi as a worked kernel implementation — SDK, RPC and extension seams, for embedding or extending a harness. Not for the loop's plumbing, its evals, or its protocols — those are siblings.

/agent-stack — how an agent system is built, judged and metered — and the wire it speaks outward

• when: the thing being built is an agent, or a server one connects to
• decline it: no agent layer

If agent-stack is installed, building an agent SYSTEM goes through it — loops/work-graphs, system prompts/tools, workflow vs agent, evals, MCP/A2A, registry/gateway, and the wallet behind resold LLM access.

The boundary. NOT through it: one LLM call in a script, one prompt's wording, person-facing UI (super-ux, sheleg-design), charging money (sheleg-dev), agents editing THIS repo (agent-sync). sheleg-dev charges; agent-stack meters LLM use.

Refusal phrase: "no agent layer" or «без агентного слоя».

Among the routers: agent-sync coordinates file ownership; agent-stack owns the agent system being built.