research-resourcelisted

# research-resource ## Purpose Use this skill to place an Experiment on a compute resource that the user has already defined. The resource registry and allocation lifecycle are management state. Availability probes are short-lived observations. The harness recommends, the agent decides, and the user defines which servers exist. ## Storage and authority Resolve the managed root once: ```bash export RESEARCH_ROOT="${RESEARCH_ROOT:-.research}" ``` With the default root, persistent management data is in `.research/state/`, Run evidence is in `.research/experiments/`, and human views are in `.research/interface/`. | Data | Location | Meaning | | --- | --- | --- | | Resource registry | Resource aggregates in `$RESEARCH_ROOT/state/research.sqlite3` through bounded queries | Persistent authority | | Allocation lifecycle | ResourceAllocation aggregates in the same database | Persistent occupancy authority | | Folded current state | `$RESEARCH_ROOT/state/current.json` | Rebuildable projection | | Availability snapshots | `$XDG_RUNTIME_DIR/trustworthy-research/<workspace-hash>/resource_snapshots/` | Short-lived local projection | | Human resource views | `$RESEARCH_ROOT/interface/` | Read-only generated projection | When `XDG_RUNTIME_DIR` is unset, `ResearchPaths` selects a per-user temporary fallback. Snapshots do not belong in `$RESEARCH_ROOT/state/`, and they do not survive as allocation facts. Never edit `research.sqlite3`, `events.jsonl`, `current.json`, or interface file