HeyEddi
The router. Invoke @heyeddi. @heyeddi-orchestrator is a deprecated alias for this skill.
Routes work to the right @skill, refreshes .heyeddi/, installs thin always-on AGENTS.md + .cursor/rules/heyeddi-always-on.mdc, and keeps product, UX, and design pillars in sync.
Every-chat policy is delivered via those agent/rule files (not by pasting full skill bodies). This skill remains the router + sync playbook. See reference/always-on.md and hub docs/always-on-skills.md.
Hard gate (first action — no exceptions)
When this skill is invoked (@heyeddi, session start, or “what should we run”), before answering, routing, or reading other skills:
Run one of these (project root = the workspace you are in):
python .agents/skills/heyeddi/scripts/ensure_always_on_agents.py --project-root . # preferred when index/workflow may also be stale: python .agents/skills/heyeddi/scripts/sync.py --project-root .Fallbacks if skills live under Cursor:
~/.agents/skills/heyeddi/scripts/…or.cursor/skills/heyeddi/scripts/….Confirm
AGENTS.mdand.cursor/rules/heyeddi-always-on.mdcexist (or--checkpasses).Only then continue:
verify_setup,suggest_skills/load_catalog, routing, design, etc.
Forbidden: Chat-only @heyeddi (explaining the skill, listing gates, or routing) without running a script that writes/verifies always-on files. Attaching the skill is not enough — a script must run.
When to use
- Every session start on a HeyEddi project (
.heyeddi/present or greenfield app request) - User asks "what skills do we have?" or "which skill should handle this?"
- Before a multi-step pipeline (intake → scaffold → design → handoff → QA)
- Design / look / polish:
@taste-penpotfirst. If no style/identity is pinned and this is not a refine of an existingdesign.mdsystem, taste runs the identity gate (references/identity-gate.md): designer thinking (project, peers, follow vs unique) written under.heyeddi/designs/, clarifying questions when needed, then 3–5 What/Why/How options — and waits for the user to choose before tokens or pixels. Scaffolding / MVP never waives that. Then one pass for tokens. Prefer@heyeddi/penpot-mcp-server— design system first (colours/type/fonts), then boards; Flex only if reflow is real (else absolute 8px); PNG Gate D required (taste-penpot/references/layout-hard-gates.mdA–G). Absolute-only MCP still needshifi.html. Then@heyeddi-designrecords product context and blocks handoff on text-walls / empty phones / missing PNG. Vue implementation follows@vue-design-system. Do not let@heyeddi-designinvent a second look. Never use unscopedpenpot-mcp-serveror@penpot/mcp. - Redesign / from scratch:
@taste-penpotloadsreferences/code-to-design.md, then the identity gate (unless the user pinned the new identity). Scope is the whole app, the named pages, or one page. The pass writes the audit and the proposal. It does not edit Vue until the user asks to build. - Any time one pillar runs: bookend with
load_workflow_context/append_pillar_opinion - After
@heyeddi-intake: confirmskill-routing.jsonbefore downstream work
Always-on siblings (every chat)
After routing, enforce:
@heyeddi-setupprefs gate (verify_setup --check) before git/CI/commit/push or agent autonomy — incomplete prefs fail those actions; route setup until pass (reference/setup-always-on.mdon that skill)@heyeddi-local-firstfor infra or account changes: edit locally, runclassify_command.py --check, thenterraform planand warn. Exit 1 means do not run the command.@engineering-excellenceplan gate before coding; change gate (audit_engineering --check) after edits — errors fail; warns advisory- Prose anti-slop on any prose you author (chat, books, docs, UI, PRs) via
context/PROSE_ANTI_SLOP.md; runverify_prose --check(add--globfor manuscripts/other paths) when files change. No "UI-only" or "not for books" exemption. - Clarify-before-act when product/design/stack intent is missing
- Host surfaces — prefer this session's native plan / data / visual tools when listed; degrade to
.heyeddi/ scripts otherwise (reference/host-surfaces.md)
Automatic .heyeddi/ + always-on upkeep
On @heyeddi itself: use the Hard gate above (must run ensure_always_on_agents or sync first).
Other HeyEddi skill tool calls still auto-sync first:
- Ensures
.heyeddi/and refreshes.heyeddi/skills-index.{json,md}when missing - Installs/updates project
AGENTS.md(managed block) and.cursor/rules/heyeddi-always-on.mdc(alwaysApply: true)
Reinstall skills (npx skills add) and keep working: the next tool that resolves the project root updates .heyeddi/ + always-on files. That does not excuse skipping the Hard gate when the user invoked @heyeddi alone.
python .agents/skills/heyeddi/scripts/ensure_always_on_agents.py --project-root .
python .agents/skills/heyeddi/scripts/ensure_always_on_agents.py --project-root . --check
python .agents/skills/heyeddi/scripts/sync.py --project-root .
Cross-pillar sync (mandatory for product · UX · design)
Read reference/cross-pillar-workflow.md.
init_workflow_sync (once per project)
load_workflow_context --route /path (start of pillar session)
… @heyeddi-product | @ux-flow-auditor | @heyeddi-design work …
append_pillar_opinion --pillar … (end: triggers sibling opinions)
Skill discovery pipeline
load_catalog / suggest_skills → auto-sync runs first
read one SKILL.md → follow that skill's pipeline
If .heyeddi/docs/intake/skill-routing.json exists, follow route order.
Tools
| Script | Purpose |
|---|---|
| (auto) | Every tool: .heyeddi/ + always-on AGENTS/rule when missing/stale |
sync.py |
Optional full sync + workflow scaffold + always-on files |
ensure_always_on_agents.py |
Install/update AGENTS.md + Cursor alwaysApply rule |
write_skills_index.py |
Scan → .heyeddi/skills-index.* |
load_catalog.py |
Read cached index |
suggest_skills.py |
Rank skills for a prompt |
suggest_next_skill.py |
Next @skill + command after any skill finishes |
init_workflow_sync.py |
Scaffold .heyeddi/docs/workflow/ |
load_workflow_context.py |
Sibling opinions + checklist for route |
append_pillar_opinion.py |
Log opinion; request UX/design/product response |
When the task is complete: suggest next skills
When you have finished the user's request for this skill (not after every tool call or subagent phase), suggest what to run next:
Run:
python .agents/skills/heyeddi/scripts/suggest_next_skill.py --current-skill heyeddi --project-root .Add
--route /pathif you worked a specific route.Include the script's
### Next stepblock in your final reply. The user copies the Prompt line into chat (e.g.@heyeddi-design craft /settings).
Pass --mode shape (or craft, audit, etc.) when you know which sub-command just finished.
See @heyeddi → reference/next-skill-handoff.md.
Related
@heyeddi-product·@ux-flow-auditor·@heyeddi-design: three pillars@heyeddi-setup: always-on prefs hard gate (verify_setup --check)@engineering-excellence: always-on plan + change gatesreference/reproduce-first.md: bug test first; frontend bugs use Playwright on local, then ask before test or production. Non-local: look only, no destructive action@heyeddi-intake: upstream intakereference/host-surfaces.md: prefer host plan/data/visual tools; portable degradereference/next-skill-handoff.md: next-skill block when a pipeline task completesdocs/cross-pillar-workflow.md: hub summarydocs/always-on-skills.md: hub always-on policy