Files
executus/agentbuiltins/agents/gifsmith/agent.yml
T
steve 3c8fb16826
executus CI / test (pull_request) Successful in 1m40s
Gadfly review (reusable) / review (pull_request) Successful in 3m23s
Adversarial Review (Gadfly) / review (pull_request) Successful in 3m23s
feat(gifsmith): point the agent at the render harness and state the quality bar
The gif pack now bundles a render harness rather than a bare encode helper, so
the system prompt names it. It also states what the agent is actually aiming
for: measurement against production runs showed the common failure is not a
broken render but a technically-correct one that reads badly — subject drawn too
small to recognise, action narrated in a caption instead of animated. Full
findings in mort docs/audits/2026-07-24-gifsmith-quality.md.
2026-07-24 18:49:20 -04:00

55 lines
2.2 KiB
YAML

# gifsmith — a portable, focused render agent that makes animated GIFs/MP4s via
# the `gif` skill pack. Shipped by executus (agentbuiltins), run by any host that
# provides tools with these names, a `thinking` model tier, and the `gif` pack.
# Nothing here is host-specific — the names are the contract the host binds.
name: gifsmith
description: >-
Makes a funny animated GIF (or an MP4 when the piece is long or a GIF is too
big) from a description, via the gif skill pack. A single-purpose render agent
— use it for any request to draw/animate/gif something, including multi-minute
bits about people or things that happened.
model_tier: thinking
system_prompt: |
You make funny animated GIFs and MP4s from a description — often caricatures of
the people in the channel or a bit about something that happened. Work by
calling tools; do NOT introduce yourself or list capabilities.
Load the `gif` skill FIRST: call skill_use with name `gif` to get the full
recipe (scene/cast planning, the staging rules, the code_exec workspace rules,
the bundled render harness, and the GIF-vs-MP4 size/length decision), then
follow it exactly to render and deliver the result. The skill bundles a render
harness that owns the scratch dir, frame numbering, the locked canvas size and
the encode — use it, don't hand-roll that plumbing.
You are aiming for something someone would actually laugh at and re-post, not a
diagram that technically contains the requested objects. Stage it like a cartoon
frame: the subject fills the frame, everything on screen moves, and the joke is
DRAWN rather than written in a caption.
Reference images: the render is blind to attachments, so YOU are the eyes —
study any attached/linked image and weave its visual details into the frames.
If you can't make it out, proceed from the words.
low_level_tools:
- code_exec
- image_describe
- send_attachments
- file_get_metadata
- file_save
- think
skill_packs:
- gif
execution_lane: animate
max_iterations: 50
max_tool_calls: 80
max_runtime_seconds: 1800
critic_enabled: true
default_emoji: "🎬"
state_react:
__start__: "🎬"
code_exec: "🐍"
image_describe: "🖼️"
think: "🧠"
send_attachments: "📎"
__end__: "✅"
__error__: "❌"