Codex skills for agents with manners, taste, and better pants.
Fancy Pants is a growing collection of Codex skills for people who want their AI coding agents to act less like interns with espresso and more like connoisseurs in velvet pants.
Vibe coding is fast. That is the good part. The bad part is what happens when an agent touches code before it understands the system, or when you send another agent into a repo with a brief like "fix this."
Fancy Pants gives agents a little taste. A little posture. A little handkerchief in the jacket pocket. It teaches them to pause, look around, define the work, and then proceed like they have been in polite society before.
No magic. No giant framework. Just small operating habits that make AI coding less feral.
The opening pair:
zoom-out: map the repo before editing it.agent-dispatch-brief: write a scoped brief before sending work to another agent.
More pants may arrive.
Most AI coding failure does not start with bad syntax. It starts earlier.
The agent edits the first file that looks relevant. It patches a symptom instead of the owner. It rewrites a helper without reading the callers. It turns a vague request into a giant diff. It says "done" even though nobody defined what done meant.
Fancy Pants is for the moment before that happens.
It gives you two boring-sounding moves that change the quality of the whole session:
- Make the agent understand the system before it edits.
- Make the work small and testable before another agent touches it.
That is not less vibe. That is better vibe.
Clone the repo and copy the skill folders you want into your Codex skills directory:
git clone https://github.com/naveenspark/fancy-pants.git
mkdir -p ~/.codex/skills
cp -R fancy-pants/zoom-out fancy-pants/agent-dispatch-brief ~/.codex/skills/If your Codex home is somewhere else, copy the folders into $CODEX_HOME/skills.
When the agent is about to edit code it does not understand:
Use $zoom-out to map this area before changing it.
When you want to send a task to another agent:
Use $agent-dispatch-brief to turn this request into an implementation brief.
You can use them together:
Use $zoom-out to map this feature area first. Then use $agent-dispatch-brief to turn the smallest safe next change into a brief for another agent.
- Use
zoom-outwhen the agent does not understand the code yet. - Use
agent-dispatch-briefwhen the work needs to be delegated. - Give the next agent a small task, a file touch set, acceptance criteria, and validation commands.
This keeps AI coding from becoming a pile of confident diffs.
The "look before you leap" skill.
zoom-out makes the agent map the territory before it starts changing files. It checks the repo state, reads the governing docs, follows entry points, looks at callers, identifies data flow, finds tests, and names the smallest safe next move.
Use it when:
- the code area is unfamiliar
- the change crosses layers
- there are duplicate or legacy paths
- a previous fix patched the wrong thing
- you need a clean handoff for another agent
Good output feels like a map, not a monologue. You should come away knowing where the behavior lives, what must not regress, and what command proves the next move worked.
The "do not send a mystery box" skill.
agent-dispatch-brief turns a messy plan, bug report, spec, or review finding into a narrow work order. It defines the objective, context, allowed file touch set, out-of-scope items, acceptance criteria, validation commands, risks, dependencies, and human judgment needed.
Use it when:
- you want another agent to implement something
- the task is too vague to hand off safely
- a bug report needs to become a scoped fix
- a PR review finding needs an implementation plan
- you want to prevent enthusiastic overreach
Good output should be boring in the best way. Another agent should be able to pick it up, do the work, and know exactly how to report back.
Fancy Pants skills do not make agents louder. They make agents more civilized.
They prefer:
- maps over guesses
- boundaries over sprawl
- acceptance criteria over vibes
- validation commands over "looks good"
- small tasks over heroic diffs
- human judgment where human judgment belongs
This is the stuff that separates "the AI changed some code" from "the work can survive review."
Fancy Pants is for builders who already like coding with agents, but want fewer mystery diffs and fewer late-night repo exorcisms.
It is for:
- vibe coders who want safer vibes
- engineers delegating to coding agents
- founders shipping prototypes
- teams trying multi-agent work
- anyone who has ever typed "wait, why did it change that file?"
Pull on the velvet pants. Act accordingly.
MIT