chore: add more agent skills for working on ConfUSIus

[sdiebolt]

Summary

Add more Agent Skills under .claude/skills/ to give LLM agents on-demand, domain-specific guidance for the most common ConfUSIus development tasks.

Background

ConfUSIus already has a .claude/skills/release/SKILL.md skill (correctly flagged disable-model-invocation: true since it's irreversible). Skills follow the open Agent Skills standard — a SKILL.md file with YAML frontmatter (required: name, description) plus optional scripts/, references/, assets/. The description field is loaded into every conversation at startup, so it must be specific and include trigger keywords. The body is loaded on demand (progressive disclosure).

Adopted by Claude Code, Codex, Gemini CLI, Cursor, Copilot, Goose, and 30+ other tools.

Proposal

Add three skills (gerund-form naming per Anthropic convention, no disable-model-invocation flag — they should be discoverable):

1. .claude/skills/processing-iq/SKILL.md — canonical data flow for adding a new IQ-processing step

   • IQDataArray → which _utils module → which public module → docstring patterns → test fixtures
   • Pointers to docs/user-guide/beamformed-iq.md, docs/api/iq.md, src/confusius/iq/

2. .claude/skills/adding-examples/SKILL.md — adding a new example to the gallery

   • Jupytext format (# %% [markdown] headers)
   • Directory convention under docs/examples/
   • Nav update in zensical.toml
   • tools/build_gallery.py discovery
   • Light/dark output-parity rule
   • tools/prefetch_doc_datasets.py cache update

3. .claude/skills/working-with-fusi-bids/SKILL.md — working with the fUSI-BIDS layout

   • Layout spec, validation module, loader patterns
   • Pointers to the fusi-bids example

Optional follow-ups

• Trim the corresponding sections out of AGENTS.md once skills exist (progressive disclosure — AGENTS.md stays high-level, skills carry the detail)
• Use the release skill as the template for SKILL.md structure and frontmatter

Why

• The current AGENTS.md is comprehensive (338 lines) but everything sits in one file. Skills let agents load domain knowledge on demand without bloating the always-loaded context.
• Each skill is one PR, ~1 day, low risk.
• Demonstrates the pattern for future contributors.

Effort

Low–medium per skill. Can be shipped as one PR with all three, or split.