[1.0] Add explicit instruction Skills to AgentBuilder

Reusable native Skills are designed in feat/skills but are not implemented at the inspected branch tip. The first slice should attach ordinary SKILL.md instructions to existing typed agents.

Scope

• Implement explicit paths/globs and configured roots on AgentBuilder.with_skills(..., runtime=False, token_budget=...).
• Safely parse optional metadata; reject ambiguous names and invalid bindings before execution. Do not import code or create publishing schemas from metadata.
• Bind immutable content per agent with source/revision and an explicit budget decision; combine Skill text with the instructions the engine actually uses.

Acceptance criteria

• A deterministic signature/engine test sees a plain prose Skill in the actual prompt even with custom engine instructions.
• The agent still emits its declared artifact and triggers its normal downstream subscription.
• Malformed metadata, name collisions, unsupported engines and budget overflow fail clearly before execution; shared content does not create shared mutable bindings.

Boundaries

Context providers continue to shape artifact context. Lazy reference tools are the next slice; Skill scripts, demo injection and optimization are deferred.

References

• Implementation dependencies: #442.
• src/flock/core/agent.py
• src/flock/engines/dspy_engine.py
• docs/plans/2026-04-17-001-feat-agent-skills-internal-flock-plan.md

Start with AgentBuilder and the entry points in src/flock/core/agent.py, then inspect src/flock/engines/dspy_engine.py and the referenced implementation plan. Use the deterministic signature/engine test described in the acceptance criteria to verify prompt content, artifact emission, subscriptions, validation failures, and isolated bindings.