Minimum ceremony. Maximum certainty.
A skill for the moment when the decision has already been made.
Stop replanning. Stop inventing process. Execute the smallest safe path to the requested repository state — and prove that it landed.
Yeet is an execution-mode skill for agentic coding systems. It activates when the human has already settled the change, the destination, and the scope. Its job is to:
- carry forward that explicit intent,
- choose the shortest compliant transport,
- execute cleanly,
- verify the resulting remote state.
It is not permission to bypass branch protection, required review, credentials, security controls, or repository policy.
Yeet is an execution mode, not permission escalation.
Yeet is not merely “move fast.” It is situational recognition followed by decisive execution.
Think of a free safety looking at a trips formation. The call says one thing. The formation looks familiar. But something is off: a split, a stance, a protection tell, a tendency the safety recognizes from film. He has enough context to understand what is probably about to happen.
So he goes.
He does not call timeout so the defense can hold another meeting. He does not ignore the scheme, either. He makes a bounded judgment inside the system, commits to it, and lives with the result.
That is Yeet mode.
The same principle shows up in Robert Horry energy. Horry was not defined by taking every shot. He was defined by recognizing the moment when the shot was already his to take — then taking it without adding hesitation, choreography, or drama.
In both cases, the important work happened before the decisive action:
- preparation
- pattern recognition
- situational awareness
- understanding the system
- knowing which constraints are real
- knowing when the decision has effectively already been made
A good agent does not confuse decisiveness with recklessness. It does not bypass branch protection, ignore required review, or invent authority it does not have. It reads the field, preserves the user's intent, respects the rules that actually bind the repository, and then — when the path is clear — commits.
In other words:
- Not: rush everything.
- Not: bypass safeguards.
- Not: freelance without context.
- Not: turn every change into a hero play.
- Yes: recognize the tell.
- Yes: understand the leverage point.
- Yes: choose the shortest compliant path.
- Yes: act before unnecessary process destroys the advantage.
- Yes: verify that the play actually landed.
The spirit is Big Shot Rob with free-safety instincts: preparation first, recognition second, hesitation last.
- “yeet it”
- “ship it”
- “do it then” (when the preceding context already defined the change and destination)
- “commit this directly to main”
- “patch it and land it”
- any clear signal that the decision is settled and the agent should stop ceremony and ship
It does not trigger merely because the word “yeet” appears in prose or documentation.
Yeet deliberately removes unnecessary process after the decision is made. It never removes the safety rails that the repository or platform actually requires.
- Direct-to-main is used only when explicitly requested and the repository permits it.
- Branch protection always wins.
- Connector-only environments use SHA-guarded writes and re-verification.
- Scope is locked before any write.
- Validation is proportional to risk.
- Hard failure is preferred over papering over uncertainty.
- Remote state is always re-read or otherwise proven.
From the canonical SKILL.md:
- Carry forward explicit intent — reuse the repository, branch, destination, and constraints already established.
- Execution over ceremony — if the task is settled, act; do not invent a draft-PR ritual when direct write was requested and allowed.
- Safety over speed when they conflict.
- Verify, don’t assume — a local edit or a connector write is not completion until the remote state is confirmed.
- Scope lock — do not opportunistically expand the change.
- Proportional validation — match the checks to the risk.
- Hard failure instead of silent degradation.
See references/transport-matrix.md for the decision table used when multiple write paths exist.
Compatible with Agent Skills environments that load skills from a repository or zip.
# example — clone or download the skill
git clone https://github.com/mcc0nnell/yeet.git
# or download the release archive and place the skill where your agent discovers skillsPoint your agent at the SKILL.md (or the skill package). Once loaded, phrases such as “yeet it” or a contextually clear “do it then” engage the execution contract.
yeet/
├── README.md
├── SKILL.md # canonical behavioral specification
├── CHANGELOG.md
├── LICENSE # MIT
├── references/
│ └── transport-matrix.md
└── scripts/
└── preflight.sh # non-mutating local preflight helper
v2.0.0 — Minimum ceremony, maximum certainty
v2 formalizes contextual “do it then / ship it” semantics, explicit user destination over generic publishing ceremony, connector-only GitHub workflows, SHA-guarded writes, direct-to-main rules, scope locks, proportional validation, diff gates, hard-failure semantics, CI verification, and proof that the requested remote state actually landed.
MIT © Robert McConnell