feat: add context-collapse-detection skill for AI agent self-monitoring

[fcfsprojects]

Context Collapse Detection Skill

This PR introduces a new AI agent skill that detects when agents are losing coherence, drifting off-goal, stuck in loops, or missing critical context — before mistakes compound.

What This Skill Does

Four collapse signals detected:

• Coherence drift — outputs no longer connect to original goal
• Goal displacement — agent solving different problem than assigned
• Loop trap — agent repeating same actions without progress
• Memory gap — agent missing information needed to proceed

Key Features

✅ Non-blocking by design — runs as parallel evaluation
✅ Self-monitoring layer — lightweight checks at key agent loop moments
✅ Recovery protocol — structured response when collapse detected
✅ Universal compatibility — Claude Code, Copilot Agent, any MCP-compatible agent
✅ Zero dependencies — no external API calls

When It Triggers

• Before any irreversible action (delete, send, deploy, purchase)
• Every 3–5 steps in a long task
• When switching between subtasks
• Before reporting results to user

Files Added

• skills/context-collapse-detection/SKILL.md — Core implementation
• skills/context-collapse-detection/README.md — User documentation
• skills/context-collapse-detection/LICENSE — MIT License