samuelgradientai-sys

brag

Project launch video via Hyperframes, upstream-maintained.

samuelgradientai-sys 2 1 Updated 6d ago
GitHub

Install

npx skillscat add samuelgradientai-sys/clawksis-agent/brag

Install via the SkillsCat registry.

SKILL.md

Brag (upstream-maintained)

Catalog stub. This entry is maintained upstream at
latent-spaces/brag: the project
ships the skill directory (skills/brag/) with its step references, tone
presets, bundled music and SFX library, and cue-analysis script. clawk skills install official/creative/brag pulls the current tree live from that
repo (quarantined and scanned like any hub install) — this directory holds
only the catalog metadata, so the vendored copy can never go stale.

/brag reads the project in the current directory, commits to an angle and a
tone (default, polished, yc-parody, chaotic, deadpan, cinematic,
app-store, or freeform direction), storyboards a 15–25 second video, hands a
composition brief to Hyperframes, and renders brag-output/brag.mp4 with a
best-frame poster and share copy. Options: --tone, --format landscape|vertical|square, --duration, --title, --no-music, --no-sfx,
and --voice (Kokoro narration, off by default).

On Claude Opus 5.5 it hands off to its bundled copy of brag-slim unless the
invocation asks for --full or --voice.

Prerequisites

  • Node.js 22+, FFmpeg on PATH, and the Hyperframes CLI (npx hyperframes doctor confirms the environment).
  • The Hyperframes path (--full, --voice, or any model other than Opus 5.5)
    loads HeyGen's domain skills by name: hyperframes-core,
    hyperframes-animation, hyperframes-creative, hyperframes-keyframes and
    hyperframes-cli, installed with clawk skills install heygen-com/hyperframes/skills/<name>. The skills guard currently blocks four
    of the five on HTML comments and credential docs in their examples
    (hyperframes-creative scores dangerous, which --force cannot override),
    so that path is not fully installable yet. brag-slim has no such
    dependency. The optional hyperframes skill in this catalog is a
    single-file port and does not provide those names.
  • uv runs scripts/analyze_music_cues.py when a custom track needs beat-sync
    cues; the bundled tracks ship precomputed cues.
  • Installs pull ~290 files (~16 MB, mostly the bundled music and SFX) through
    the GitHub contents API. Set GITHUB_TOKEN or sign in with the gh CLI
    first; the anonymous limit of 60 requests an hour is too low for this bundle.

Full documentation: https://github.com/latent-spaces/brag#readme