[P1] Add a minimal dumb expert-system agent as a downstream consumer #51

New issue

Closed

opened 2026-08-16 04:01:24 +00:00 by lost-rob0t · 1 comment

lost-rob0t commented

2026-08-16 04:01:24 +00:00

(Migrated from github.com)

Goal

Add a deliberately simple rule-based expert-system agent that demonstrates how downstream agents are built on top of prolog-rlm without becoming part of the runtime core.

This should be boring on purpose: facts, rules, a small query surface, optional loaded tools, and explicit outcomes.

Shape

The reference agent should support a tiny knowledge base such as:

fact(...).
rule(...).

or ordinary Prolog predicates, then expose a small agent-facing API for asking a goal and returning structured results.

The expert system may use prolog-rlm supervision/tool-loading infrastructure, but its domain rules remain separate from the runtime library.

Acceptance criteria

Agent implementation lives outside the core runtime modules.
It can run with zero LLM calls for ordinary expert-system queries.
Facts/rules are ordinary inspectable Prolog data/predicates rather than hidden prompt state.
Query results distinguish at least success, failure, and unknown/unsupported input where appropriate.
It can optionally load explicitly selected external tool libraries through the core tool contract.
It demonstrates capability scoping rather than receiving every loaded tool automatically.
It includes a tiny deterministic example knowledge base and tests.
It documents how another agent can replace the example rules/domain without modifying prolog-rlm.

Non-goals

No autonomous planner required.
No LLM required for inference.
No hard-coded dependency on the standard tool pack.
Do not turn the example into a second agent framework.

Architectural intent

prolog-rlm = reusable runtime/library.

External tool library = optional capabilities.

Expert-system agent = one small downstream application proving the library can support agents that are much simpler than an RLM planner.

## Goal Add a deliberately simple rule-based expert-system agent that demonstrates how downstream agents are built **on top of** `prolog-rlm` without becoming part of the runtime core. This should be boring on purpose: facts, rules, a small query surface, optional loaded tools, and explicit outcomes. ## Shape The reference agent should support a tiny knowledge base such as: ```prolog fact(...). rule(...). ``` or ordinary Prolog predicates, then expose a small agent-facing API for asking a goal and returning structured results. The expert system may use `prolog-rlm` supervision/tool-loading infrastructure, but its domain rules remain separate from the runtime library. ## Acceptance criteria - [ ] Agent implementation lives outside the core runtime modules. - [ ] It can run with **zero LLM calls** for ordinary expert-system queries. - [ ] Facts/rules are ordinary inspectable Prolog data/predicates rather than hidden prompt state. - [ ] Query results distinguish at least success, failure, and unknown/unsupported input where appropriate. - [ ] It can optionally load explicitly selected external tool libraries through the core tool contract. - [ ] It demonstrates capability scoping rather than receiving every loaded tool automatically. - [ ] It includes a tiny deterministic example knowledge base and tests. - [ ] It documents how another agent can replace the example rules/domain without modifying `prolog-rlm`. ## Non-goals - No autonomous planner required. - No LLM required for inference. - No hard-coded dependency on the standard tool pack. - Do not turn the example into a second agent framework. ## Architectural intent `prolog-rlm` = reusable runtime/library. External tool library = optional capabilities. Expert-system agent = one small downstream application proving the library can support agents that are much simpler than an RLM planner.

nsaspy referenced this issue

2026-08-27 20:02:05 +00:00

Reconcile backlog, roadmap, and GitHub issue state #89

nsaspy closed this issue

2026-09-08 01:29:00 +00:00

nsaspy commented

2026-09-08 01:31:53 +00:00

Owner

Superseded by #376/#377. The useful requirement from this issue — a deterministic zero-LLM expert proving that expert-system behavior is ordinary Prolog and not hidden prompt/model state — is now a required acceptance fixture of the canonical first-class expert runtime rather than a separate one-off downstream demo architecture.