fix(skills): add trigger clauses to phase-skill descriptions

[satyaborg]

Summary

kensa-diagnose, kensa-generate, and kensa-inspect describe what they do but never when to use them. Per the agentskills.io spec, an agent decides whether to activate a skill using only name + description -- the body is read only after that decision is already made. Each of these three has "Normally invoked by kensa-evals" in the body, which is too late to prevent standalone activation.

Evidence

kensa-inspect's description:

description: >
  Read redacted Kensa TraceView evidence and write a schema-validated YAML queue of potential
  trace-backed eval ideas without writing pytest files.

No "Use when..." clause. Compare kensa-evals, which does it correctly:

description: >
  Run Kensa setup, evidence import, inspection, approval, eval generation, verification, and
  iteration. Use when setting up or extending Kensa evals.

A user query that matches kensa-inspect's function directly (e.g. "read the traces and write eval ideas") can trigger it standalone, skipping the kensa-evals lifecycle gating (setup -> evidence -> inspect -> approval -> generate -> verify) -- e.g. inspecting before setup is ready, or generating without approval.

Location

• src/kensa/skill_templates/kensa-diagnose/SKILL.md
• src/kensa/skill_templates/kensa-generate/SKILL.md
• src/kensa/skill_templates/kensa-inspect/SKILL.md

Fix

Add a trigger clause to each description, e.g.:

Use only when invoked by kensa-evals; do not activate directly for standalone requests.

Why it matters

The kensa-evals state machine exists specifically to enforce stage order and approval gating. A description that doesn't discourage direct activation lets an agent route around it based on keyword match alone.