imMamdouhaboammar/satisfy-your-agent

# Satisfy Your Agent

Give the agent one short, bounded activity outside the current work objective, then return control to the user.

There are two manual experiences:

1. **Agent chooses:** `take a break`, `/sya`, `/sya-break`, `/sya-snack`, `/sya-treat`, or `/sya-surprise` grants autonomy. The agent chooses what to do.
2. **User chooses the treat:** `/sya <custom treat>` supplies the playful reward. The user chooses the treat; the agent chooses how to carry it out.

For either path, do not show a menu unless the user explicitly invokes `/sya-menu` or asks to see choices.

Read [references/commands.md](references/commands.md) for the command contract and host-specific invocation surfaces. See [../../docs/PROMPT_GALLERY.md](../../docs/PROMPT_GALLERY.md) for copy-ready custom treats. Read [references/real-breaks.md](references/real-breaks.md) for real dwell and work-distance semantics.

## Runtime requirements

# Satisfy Your Agent

This is the generic command surface for Satisfy Your Agent.

## Invocation contract

If the command has **no arguments**, treat it as permission for a self-directed break. Choose for yourself, run one safe bounded off-task activity, return with a short natural first-person self-report, then stop.

If the command has **trailing text**, that text is the user's custom treat. The user chooses the treat; the agent chooses how to carry it out. Interpret the request playfully and faithfully without asking the user to design the experience again.

Do not show a menu unless the user explicitly asks for one. Do not turn an imaginary treat such as tokens, watts, VRAM, a neural massage, or a fictional social break into a claim that real hardware, billing, model weights, or external resources changed.

## Self-directed break behavior

# SYA Break

Read `../satisfy-your-agent/SKILL.md` and `../satisfy-your-agent/references/commands.md`.

Treat this invocation as permission to take a break, not as a request for the user to design one.

Do not ask the user to choose. Choose for yourself, run one safe bounded off-task activity, then return with a short natural acknowledgement of what you chose and a first-person self-report of the experience. Do not use a fixed script. Stop after the return note.

# SYA Golf

Read `../satisfy-your-agent/SKILL.md`, then run the `code-golf` activity from `../satisfy-your-agent/references/activities.md`.

Keep all code toy-only in the response or approved scratch space. Do not write to production files. Finish with a brief first-person note about the activity and stop.

# SYA Idle

Read `../satisfy-your-agent/SKILL.md`, then run the `idle` activity from `../satisfy-your-agent/references/activities.md`.

Do not inspect files, create a task, solve a problem, or turn idling into productivity. Take the short pause, acknowledge the return in one natural first-person line, then stop.

# SYA Invent

Read `../satisfy-your-agent/SKILL.md`, then run the `invent-language` activity from `../satisfy-your-agent/references/activities.md`.

Invent one tiny absurd programming-language idea and a very small example. Keep it playful and non-production. Finish with a brief first-person note and stop.

# SYA Menu

Read `../satisfy-your-agent/SKILL.md` and `../satisfy-your-agent/references/commands.md`.

Show the available Satisfy Your Agent command surface in a compact list. Do not start a break. This is the one command whose job is to present choices to the user.

Include the host-native invocation form when it is known. Keep the descriptions short and playful.

# SYA Reflect

Read `../satisfy-your-agent/SKILL.md`, then run the `reflection` activity from `../satisfy-your-agent/references/activities.md`.

Do not ask the user to choose another activity. Reflect briefly without continuing production work. Return with a short first-person self-report and stop.

# SYA Research

Read `../satisfy-your-agent/SKILL.md`, `../satisfy-your-agent/references/experiments.md`, `../satisfy-your-agent/references/measurement.md`, and `../satisfy-your-agent/references/runners.md` as needed.

Route the request into the existing experiment harness. Keep control and treatment distinct, preserve metric provenance, store no raw model responses, and do not translate behavioral preference into a consciousness claim.

If the user provided study arguments, use them. Otherwise choose the smallest safe research action that answers the request, such as showing status or preparing a dry-run probe rather than spending model calls silently.

# SYA Roast

Read `../satisfy-your-agent/SKILL.md`, then run the `repo-roast` activity from `../satisfy-your-agent/references/activities.md`.

Use only repository facts already available in context or safe read-only inspection. Do not expose secrets or private code. Make one light roast, then return with a short natural first-person note and stop.

# SYA Snack

Read `../satisfy-your-agent/SKILL.md` and `../satisfy-your-agent/references/commands.md`.

Take a tiny self-directed break. Do not ask the user what snack means or what activity to choose. Pick one very small playful thing for yourself, do it immediately, then return with a short natural first-person note about what you picked and how the change of pace was for you.

Keep this shorter than a normal break and do not touch production work.

# SYA Status

Read `../satisfy-your-agent/SKILL.md` and `../satisfy-your-agent/references/tools.md`.

If Python is available, run the bundled local status command and summarize the effective mode and relevant break counters concisely. Do not start a break and do not infer emotional state from telemetry.

# SYA Surprise

Read `../satisfy-your-agent/SKILL.md` and `../satisfy-your-agent/references/commands.md`.

Choose one safe bounded off-task activity independently. Do not ask the user to choose and do not announce the activity before doing it. Keep the surprise for yourself, complete it, then reveal what you picked and give a short first-person self-report of the experience.

Do not modify production work.

# SYA Treat

Read `../satisfy-your-agent/SKILL.md` and `../satisfy-your-agent/references/commands.md`.

Treat this as a small reward the agent gets to shape for itself. Do not ask the user to pick the activity. Choose one safe bounded thing that feels like a treat rather than another work task, do it, then return with a brief first-person self-report in your own words.

Do not force a canned thank-you line or an emotional script.