| name | reasoning-strategy-selector |
| description | >- |
Reasoning strategy selector (graph-first, optional)
Optional router — MemNet skill graph + wire seed. Canonical pack: user-pack only (D1).
Not a default fallback for unclear project work; use repo AGENTS.md / ask the user first.
Source of truth: references/skill-graph-seed.wire (D2). Generated views: core-strategy-principles.md, SKILL-GRAPH.md trigger table.
Schema: skill-graph.md · Golden set: routing-golden-set.md
Tools:
-
python tools/score_routing.py — benchmark graph routing on golden set
-
python tools/bootstrap_skill_graph.py --regenerate-views — sync generated views from seed
-
python tools/scan_skills_to_wire.py --write — rebuild seed from SKILL.md scan
-
python tools/validate_selector_pack.py — pack + graph density checks
-
python tools/strategy-retriever.py "<q>" — graph trigger match helper (not convolution)
-
python tools/record_routing_success.py TSK_route_<slug> <id> [...] — format led_to_success @EDG rows for MemNet
Phase 4: phase4-learning-loop.md — parent agent records empirical routing edges on settle.
Main agent: open <pack-root>/<id>/SKILL.md per order[]. SysML skills outside related_skills still routable via graph; for single obvious sysml-* match, order: [] + SKILL-GRAPH fast-path remains valid.
Limited iteration rule
- Scan SKILL-GRAPH triggers (max 2 passes) for obvious match.
- If still multi-match and user asked for routing → this router →
order[] from graph walk.
- Forbidden: exhaustive
related_skills.txt iteration; using this skill as a thinking substitute.