Never forget a /slash command again.
Auto-generated workflow cheatsheets that live inside Claude Code via the companyAnnouncements field. Claude sees your full toolkit at the start of every session — and starts suggesting the right command chain instead of defaulting to /plan for everything.
You installed 40+ commands. You use 3.
Every harness — Superpowers, ECC, OMCC, Matt Pocock's — ships with dozens of carefully designed workflows. But Claude doesn't see them up-front, so neither do you.
Without a cheatsheet
You type /plan reflexively. Claude doesn't know /orchestrate, /diagnose, or autopilot: exist until you remember to use them.
The slash menu shows names, not chains. The right tool for a bug is buried three levels deep. Workflow muscle memory never forms.
With companyAnnouncements
This skill scans your installed commands, categorises them into workflow chains, and writes them to companyAnnouncements.
Claude reads them at session start and proactively suggests the right chain — /orchestrate bugfix "login crash" → /e2e — because it now knows the full toolkit.
One JSON field. Every session. Forever.
Claude Code's companyAnnouncements field is designed for team broadcasts. This skill repurposes it as persistent workflow memory.
~/.claude/commands/ and project .claude/commands/, recognises harness signatures (Superpowers, ECC, OMCC, Matt Pocock).→, |, , as a tiny grammar.companyAnnouncements. Claude reads it at every session start, zero latency.{ "companyAnnouncements": [ "[Workflows] Dev: /orchestrate feature \"desc\" -> /e2e\n[Workflows] Quality: /code-review -> /refactor-clean -> /verify" ] }
Important —
the field is string[]. With multiple items Claude Code rotates one per session. Use a single item with \n separators to show all lines every session.
Bring your harness. We brought the cheatsheet.
Five ready-made presets. Pick one, paste, done. Or mix-and-match for hybrid setups.
{ "companyAnnouncements": [ "[Superpowers] Design: /brainstorming -> /writing-plans\n[Superpowers] Execute: /using-git-worktrees -> (/executing-plans or /subagent-driven-development)\n[Superpowers] Per Task: /test-driven-development -> /requesting-code-review -> /receiving-code-review\n[Superpowers] Finish: /verification-before-completion -> /finishing-a-development-branch\n[Superpowers] Tools: /systematic-debugging (bugs) | /dispatching-parallel-agents (parallel) | /writing-skills (meta)" ] }
{ "companyAnnouncements": [ "[Workflows] Dev: /orchestrate feature|bugfix \"desc\" -> /e2e | Manual: /plan -> /tdd -> /code-review -> /verify\n[Workflows] Reproduce: /e2e (as-is) -> /orchestrate bugfix -> /e2e (to-be) | Build: /build-fix -> /verify\n[Workflows] Quality: /code-review -> /refactor-clean -> /verify | /quality-gate | /eval define|check|report\n[Workflows] Multi: /multi-plan -> /multi-execute | /devfleet \"task\" | /model-route \"task\" --budget low|med|high\n[Workflows] Docs: /update-docs, /update-codemaps | Ref: /docs \"lib\" | Loop: /loop-start sequential|continuous-pr\n[Workflows] Learn: /learn -> /learn-eval -> /skill-create | Instincts: /instinct-status -> /evolve -> /promote\n[Workflows] Session: /save-session, /resume-session | Meta: /harness-audit, /skill-health, /context-budget" ] }
{ "companyAnnouncements": [ "[OMC] Auto: autopilot: desc | Persist: ralph: desc | Parallel: ulw tasks\n[OMC] Team: /oh-my-claudecode:team 3:executor \"task\" | Plan: ralplan | QA: /oh-my-claudecode:ultraqa\n[OMC] Clarify: deep-interview \"idea\" -> omc-plan --consensus -> autopilot: desc\n[OMC] Investigate: /oh-my-claudecode:trace | Tri-Model: ccg \"query\" | Visual: /oh-my-claudecode:visual-verdict\n[OMC] Cleanup: deslop | Skills: /oh-my-claudecode:skill list | Learn: /oh-my-claudecode:learner\n[OMC] Session: /oh-my-claudecode:psm | Release: /oh-my-claudecode:release | Diag: /oh-my-claudecode:omc-doctor" ] }
{ "companyAnnouncements": [ "[Matt] Plan: /grill-with-docs -> /to-prd -> /to-issues\n[Matt] Bug: /diagnose (6-phase: reproduce -> hypothesise -> instrument -> fix+regression -> cleanup)\n[Matt] Feature: /tdd (vertical tracer bullets) | /prototype (LOGIC or UI)\n[Matt] Triage: /triage — needs-info / ready-for-agent / ready-for-human / wontfix\n[Matt] Explore: /zoom-out | /improve-codebase-architecture" ] }
Install first: npx skills add https://github.com/mattpocock/skills, then run /setup-matt-pocock-skills once per repo.
{ "companyAnnouncements": [ "[Workflows] Dev: /plan -> /code-review | Bug: /build-fix\n[Workflows] Docs: /docs \"lib\" | Session: /save-session, /resume-session" ] }
Two paths. Pick one.
Both routes write to the same place — ~/.claude/settings.json. The skill route auto-detects what's installed; the manual route is fastest if you already know your harness.
As a Claude Code skill recommended
$ cp -r . ~/.claude/skills/company-announcements
Then inside Claude Code:
/setup-announcements # auto-detect /setup-announcements --harness superpowers # Superpowers /setup-announcements --harness ecc # ECC /setup-announcements --harness omcc # OMCC /setup-announcements --harness mattpocock # Matt Pocock
Manual paste 30 seconds
Open ~/.claude/settings.json, add the preset for your harness from the Presets section, save. Done — every new session picks it up.
$ open ~/.claude/settings.json # paste preset → save → restart Claude Code session # # Verify it loaded: $ claude -p "what workflows do you have available?"
The questions everyone asks first.
Does this slow down Claude Code?
No. companyAnnouncements is injected into the system prompt — a few hundred tokens at most, with zero added latency per request.
Can I use this without any harness installed?
Yes. Use the Vanilla preset for built-in commands only, or write your own preset for any team-specific workflow.
Why not just list every command and flag?
Claude's attention is finite. Keep it under ~10 lines and focus on chains, not flag references. The slash menu still surfaces individual commands — this layer is for workflow shape.
The field is an array — what if I add multiple items?
Claude Code rotates one item per session (treating the array as a banner rotation). To show the same cheatsheet every session, use a single string with \n separators — that's what every preset here does.
Does this work in the VS Code / JetBrains extension?
Yes. settings.json is shared across all Claude Code interfaces — CLI, desktop app, IDE plugins.
Can I mix presets? I run both ECC and OMCC.
Combine them into a single string. Keep the prefix tags ([Workflows], [OMC]) so Claude can tell them apart, and stay under ~10 lines total.