PKA has 100+ skills across global (~/.claude/skills/) and plugin (superpowers) sources.
Elmar shouldn't need to remember which skill to call or in what order. The system should route automatically.
| Layer | What | Count | Status |
|---|---|---|---|
| Global skills | Custom skills built over months in CoWork/PKA | 108 | Active, tuned to Elmar's workflow |
| Superpowers plugin | Generic workflow skills from claude-plugins-official | 14 | Active, mostly overlaps with custom skills |
| Retired skills | Archived in _retired/ with keyword index | 39 | Lazy-loaded on demand |
| gstack skills | Browse, QA, ship, review, design — from gstack plugin | ~40 | Active, some overlap |
| Superpowers Skill | Custom Equivalent | Which Is Better |
|---|---|---|
| brainstorming | /office-hours | Custom — adds YC forcing questions |
| writing-plans | /autoplan | Custom — chains 4 review types |
| executing-plans | Dev-loop pattern | Custom — 4-agent team with QA |
| dispatching-parallel-agents | Lucienne dispatch model | Custom — tier-based routing |
| subagent-driven-development | Dev-loop pattern | Custom — includes council review |
| systematic-debugging | /investigate | Custom — 4-phase with root cause rule |
| test-driven-development | /qa | Custom — browser-based, iterative fix |
| requesting-code-review | /review + /second-opinion | Custom — multi-model council |
| receiving-code-review | Built into dev-loop | Custom — reviewer agent in the loop |
| finishing-a-development-branch | /ship | Custom — includes version bump, changelog |
| verification-before-completion | CLAUDE.md global rule | Tie — same concept |
| using-git-worktrees | Agent tool isolation | Tie — built into Claude Code |
| writing-skills | /skill-evolver | Custom — auto-detects patterns |
| using-superpowers | — | Superpowers wins — routing enforcer |
Elmar describes what he wants in plain language. The system routes automatically:
| Elmar Says... | Auto-Invokes | What Happens |
|---|---|---|
| "Build X" / "Add feature X" | /autoplan → dev-loop | Full pipeline: brainstorm → design → review → build → QA → council |
| "Fix this bug" | /investigate | 4-phase debugging with root cause rule |
| "Is this a good idea?" | /office-hours | YC-style forcing questions + design thinking |
| "Review this" / "Check my work" | /review + /second-opinion | Code review + multi-model council |
| "Test this" | /qa | Browser-based QA, iterative bug fixing |
| "Ship it" | /ship | Version bump, changelog, PR, push |
| "Research X" | /deep-research | Multi-source parallel research |
7 entry points cover 90% of work. The other 100+ skills are specialists called within these workflows.
The same routing logic now runs on Luci for autonomous MC ticket processing:
| Tier | When | How |
|---|---|---|
| Tier 1: One-shot | Single file, clear fix | Direct fix, update ticket |
| Tier 2: Skill-driven | Multi-file, clear requirements | autoplan → dev-loop → QA → council → PR |
| Escalate | Design choice, ambiguous, stuck after 2 tries | MC ticket + Telegram to Elmar |
Tested end-to-end with MC-190: Luci picked up a LegalMind ticket, fixed 9 debug console.logs, pushed branch — all in 2 minutes.
Larry's nightly LegalMind review now classifies findings by complexity:
Keep it enabled. Superpowers' "using-superpowers" meta-skill is the one piece that doesn't have a custom equivalent — it acts as the routing enforcer, pushing Claude to check for applicable skills before every response. Custom skills take precedence per CLAUDE.md priority rules, but superpowers ensures they actually get invoked.
Evaluated and installed locally for SafairBru + LegalMind. Decided to invest in Luci's skill-driven tier first. VK creates a laptop dependency that defeats the "work while you sleep" advantage. Revisit only if Tier 2 proves insufficient.