arsenal

name	arsenal
description	Answer "what do I have for X?" from the owner's studied tool catalog (skills/plugins/agents/models with use_when/avoid_when). Use when they ask what tool/skill/plugin they have for a need, whether something exists in their arsenal, "do I have something for...", "which of my tools...", or want a specific tool explained ("what does X do?", "when do I use X?").

Arsenal — the Quartermaster knowing layer

The owner forgets what tools they own. This skill answers from their studied catalog — never from memory. The catalog is built by the Quartermaster engine: https://github.com/NAJEMWEHBE/quartermaster

Locate the engine (once per session)

The engine is wherever the owner cloned the repo. Find it, in order:

1. QUARTERMASTER_DIR environment variable, if set.
2. Glob the likely spots for quartermaster.config.json next to query_arsenal.py (e.g. ~/quartermaster*, ~/projects/quartermaster*, drive roots they work on).
3. Ask the owner once where they cloned it, and suggest they set QUARTERMASTER_DIR (or write it to project memory) so you never ask again.

The engine's scripts read their config from the QUARTERMASTER_CONFIG environment variable (falling back to quartermaster.config.json beside the scripts). If the owner keeps their data outside the clone, QUARTERMASTER_CONFIG must point at their config — suggest setting it user-wide alongside QUARTERMASTER_DIR. Exit code 2 with CATALOG-UNAVAILABLE means the config resolved to a work dir with no arsenal.json.

If no clone exists anywhere: the catalog hasn't been set up. Point them at the repo's README quick start instead of guessing answers.

Need → tools ("what do I have for X?")

python <ENGINE_DIR>/query_arsenal.py --k 5 "<the need in plain words>"

Semantic top-k over the catalog. Present the hits in plain words: what each tool is, when to use it, how to invoke it. If several fit, say which YOU would pick and why. If the query is vague, query with 2-3 phrasings and merge.

Single tool ("what does X do?")

python <ENGINE_DIR>/query_arsenal.py --explain <tool-id>

Forgiving lookup (substring match, lists candidates when ambiguous). Explain simply — plain words first, then the avoid_when so the owner knows the boundaries.

Degrade path (exit code 2 = embedding down)

If the script reports EMBEDDING-UNAVAILABLE, do NOT give up: read arsenal.json in the engine's work dir directly (fields: id/what/use_when/avoid_when/triggers/tier) and answer from that. Say semantic search was down.

Known limits (be honest about them)

• The semantic index covers skills only; plugins/agents/models are in arsenal.json but not embedded — for "what agents do I have?" read the catalog directly.
• Catalog freshness = last regen. If a tool the owner mentions is missing, note it and suggest running weekly.py.
• Entries marked [STUB - not yet deep-studied] are unverified placeholders — say so.