corpspeak

corpspeak

Two joke skills that translate between blunt speech and corporate speak, in Russian and English, and decode corporate emails back into what was actually meant.

Русская версия

What it does

Blunt to corporate, three intensity levels:

fuck if I know

1 → I'm not sure, let me check.
2 → I don't have visibility into that, it sits outside my area.
3 → Team, I don't have full visibility here. I'd suggest routing this
    to the domain owner and circling back with an update.

Corporate to blunt, which is the direction most people actually want:

Let's circle back on this after planning
→ Never.
Bottom line: the question is closed, there will be no discussion.

I'd like to sync to level-set expectations
→ You did the wrong thing.
Bottom line: prepare to explain yourself, and not over email.

There are some nuances to how this works
→ It doesn't work.
Bottom line: you're expected to figure it out and not ask in public.

Bullshit density, for a whole email:

Bullshit Density: 78%  filler email, the real content is one paragraph

What was actually said:
- the launch moved to Q3
- Sarah owns the migration now

Red flags:
- "we've heard your feedback" → nothing will change
- "aggressive timeline" → impossible timeline

Bottom line: the date slipped and nobody wants to write that down.

Why it is not a lookup table

The obvious way to build this is a dictionary: 200 rude phrases on the left, 200 corporate ones on the right. That breaks on the first input nobody anticipated, which is always the funniest one, and it returns the same output every time so the joke dies on the third use.

The skills work from a generative rule instead. Every utterance carries three layers: the proposition (what was said), the affect (how annoyed the speaker is), and the implicature (what was meant but not said). Corporatization keeps the proposition intact, launders the affect through grammar, and buries the implicature. The decoder runs it backwards.

Six moves do the laundering:

The dictionaries that ship with the skills exist to calibrate register, not to be searched. Moves apply first; the lexicon is there so the output sounds like a real Tuesday standup rather than a thesaurus.

The part that is easy to get wrong

Density is not the joke. Precision is.

A translation lands when the original thought is fully recoverable and the output still reads like a slide. Level 3 piles on frameworks and workstreams, but the proposition has to survive underneath. Word salad with nothing under it is not funny, it is just empty, and it is what most "corporate bullshit generators" produce.

The second failure is collapse: every angry input converging on "let's sync on this". The skills hold the concrete object of the complaint (the code, the deadline, the person, the process) and read the target, because "you're an idiot" and "this code is idiotic" have to come out differently.

Russian and English are not translations of each other

English corporate speak runs on its own idioms. A calque from Russian reads dead, and the reverse is also true.

So there are two skills, not one skill with a language flag.

What is inside

Both skills are self-contained: the whole mechanism, the scale, all four modes, and a working lexicon live in SKILL.md. The references/lexicon.md next to each one is an extension, not a dependency.

Install

As a Claude Code plugin:

/plugin marketplace add spyrae/corpspeak

Then install corpspeak from the marketplace. Both skills activate on their own when you ask to translate something to or from corporate.

Manually, any agent that reads skills:

git clone https://github.com/spyrae/corpspeak.git
cp -r corpspeak/skills/* ~/.claude/skills/

Anywhere else: the skills are plain Markdown prompts. Paste skills/corpspeak-en/SKILL.md (or the -ru one) into ChatGPT, Gemini, Cursor, or whatever you use, then send your text. Nothing in them is Claude-specific, and each file works alone.

Boundaries

This is a joke, and it should stay one.

The skills take profanity as input, because that is the linguistic material they operate on. They do not produce abusive messages aimed at a named living person: the decoder returns meaning, not ammunition for a fight you will regret.

They are also not a business writing tool. If you have a real email that has to land well, write it yourself or use something built for that. Running your actual quarterly update through mode 1 at level 3 will produce something very funny and completely unsendable, which is the point.

License

MIT. See LICENSE.