Agentic Almanac Scout

Read-only mining, scoring and SKILL.md extraction for open-source agents.

• tools/almanac_scout.py — the scout: ingests GitHub/Hugging Face repos or local checkouts (never executes their code), scores them on scope/utility, free-tier token efficiency, and loop-risk, then writes vetted skills/drafts/<id>/SKILL.md proposals. Aggressive directives ("ignore user pushback", unauthorized auto-commits, unbounded loops, ...) are rejected by the safety filter before anything hits disk.
• docs/AGENT_ALMANAC.json — live registry consumed by Whorl's src/skill/scout.py.
• docs/AGENT_ALMANAC_SCHEMA.json — full registry schema.
• schema/agent_entry_schema.json — per-entry compatibility schema.

Quickstart

python3 tools/almanac_scout.py ingest https://github.com/huggingface/smolagents --name smolagents-code-agent
python3 tools/almanac_scout.py score          # ranked utility table
python3 tools/almanac_scout.py drafts         # live safety verdict per draft
python3 tools/almanac_scout.py check FILE...  # run the safety filter standalone
python3 tools/almanac_scout.py validate       # schema-validate the registry
python3 tools/almanac_scout.py console        # interactive REPL (dynamic views only)

Scoring: utility = 0.4*scope_utility + 0.25*free_tier_efficiency + 0.35*loop_risk (loop_risk is inverse risk: 10 = single-check bounded execution).