history-explorer

name	history-explorer
description	Query the per-turn DecisionEntry log for skill co-occurrence patterns and the router fixture corpus. Returns a JSON summary suitable for downstream LLM consumption. Useful for 'which skills did I use most this week?'
provenance	{"origin":"agentos-original","license":"MIT"}
metadata	{"requires":{"anyBins":["python","python3"]}}
entrypoint	{"command":"python {baseDir}/scripts/explore.py","args":["--query","{{ with.query | truncate(512) }}","--window-days","{{ with.window_days | default('30') }}","--include","{{ with.include | join(',') if with.include is sequence and with.include is not string else with.include | default('co_occurrences,router_fixtures') }}","--top-k","10"],"parse":"json","timeout":30}

History Explorer

Lightweight read-only view over ~/.agentos/logs/decisions-*.jsonl. Aggregates DecisionEntry.skills_invoked (SCHEMA_VERSION 10) into co-occurrence frequencies and surfaces the router fixture corpus (when present).

Usage

uv run python {baseDir}/scripts/explore.py \
  --log-dir ~/.agentos/logs \
  --query "Co-occurring chains for PDF workflows" \
  --window-days 30 \
  --include co_occurrences,router_fixtures \
  --top-k 10

Output

JSON to stdout with keys co_occurrences, router_fixtures, and a placeholder string when the log is empty.

Fallback

If no decision-log exists, return an empty result with a placeholder string explaining "no history; downstream should rely on user intent only".