[RAGE-9][P0] Ship headless refinery commands and one resumable RAGE pipeline over AgentProlog #39

New issue

Open

opened 2026-09-20 01:40:01 +00:00 by nsaspy · 0 comments

nsaspy commented

2026-09-20 01:40:01 +00:00

Owner

Parent: #6
Depends on: RAGE-2, RAGE-4, RAGE-5, RAGE-6
Optional later stages: RAGE-7, RAGE-8

Goal

Expose the refinery as a first-class headless AgentProlog product workflow instead of a collection of library demos.

Target commands

Exact CLI spelling may follow current AgentProlog conventions, but the operations must cover:

observe
patterns
candidates
evaluate
evolve/revise
promote/reject/rollback
import-skills
status/explain

The command layer is a thin product facade over the canonical Prolog-RLM runtime and RAGE product records.

Requirements

resumable run identity;
deterministic status/explain output suitable for TUI/Emacs/automation clients;
bounded source scope and budget configuration;
dry-run for observation/classification/candidate/eval planning;
structured exit/failure states;
cancellation propagates to upstream Futures/subagents/evals;
no second scheduler, provider stack, authority layer, or generic agent loop.

Acceptance

one headless command can run observe -> classify -> synthesize -> evaluate end-to-end;
interruption/resume does not duplicate evidence or candidate identity;
operator can explain why a candidate exists and why it passed/failed evaluation;
headless behavior is frontend-neutral and consumable by future Workbench/Emacs clients.

Parent: #6 Depends on: RAGE-2, RAGE-4, RAGE-5, RAGE-6 Optional later stages: RAGE-7, RAGE-8 ## Goal Expose the refinery as a first-class headless AgentProlog product workflow instead of a collection of library demos. ## Target commands Exact CLI spelling may follow current AgentProlog conventions, but the operations must cover: ```text observe patterns candidates evaluate evolve/revise promote/reject/rollback import-skills status/explain ``` The command layer is a thin product facade over the canonical Prolog-RLM runtime and RAGE product records. ## Requirements - resumable run identity; - deterministic status/explain output suitable for TUI/Emacs/automation clients; - bounded source scope and budget configuration; - dry-run for observation/classification/candidate/eval planning; - structured exit/failure states; - cancellation propagates to upstream Futures/subagents/evals; - no second scheduler, provider stack, authority layer, or generic agent loop. ## Acceptance - [ ] one headless command can run observe -> classify -> synthesize -> evaluate end-to-end; - [ ] interruption/resume does not duplicate evidence or candidate identity; - [ ] operator can explain why a candidate exists and why it passed/failed evaluation; - [ ] headless behavior is frontend-neutral and consumable by future Workbench/Emacs clients.

nsaspy referenced this issue

2026-09-20 01:40:38 +00:00

[EPIC] Personal skill refinery: observe workflows, synthesize candidates, RAGE-evaluate, and import external skills #6