Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/dmmdea/offload-harnessnpx agentmods add skills/dmmdea/offload-harness/hyperframes-composeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/dmmdea/offload-harness/hyperframes-compose)<a href="https://agentmods.dev/skills/dmmdea/offload-harness/hyperframes-compose"><img src="https://agentmods.dev/badge/skills/dmmdea/offload-harness/hyperframes-compose/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/dmmdea/offload-harness/hyperframes-compose"><img src="https://agentmods.dev/badge/skills/dmmdea/offload-harness/hyperframes-compose.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00144 | $0.03542 |
| Opus 5.5 | $0.00058 | $0.01417 |
| Sonnet 5.5 | $0.00029 | $0.00708 |
| Haiku 4.5 | $0.00014 | $0.00354 |
Grade A, and why
hyperframes-compose scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 172 lines — stays where its author put it; the contents beside it link to each section on GitHub.
HyperFrames compose lane (offload_compose_video)
The harness renders designed, text-exact motion graphics from HTML and CSS with a pinned, environment-scrubbed
HyperFrames (ADR 0059). It is CPU-class: software GL and CPU encode, no GPU lease, so it runs beside every
render and every text seat. Same inputs give the same frames. This file only routes and states the rules;
the template contract and the measured numbers are in render/compose-templates/README.md, and the lane's
design is in docs/systems/media-generation.md (Composition).
Hard bans
- Never run
npx hyperframes, and never run thehyperframesCLI by hand, on any machine. The harness runner is the only door. It pins the version, scrubs the environment, forces--json(the one switch that skips the CLI's network checks) and refuses everything exceptlint,check,render,snapshot,browser ensure|pathand--version. - Never
init,skills,upgrade,add,capture,preview,tts,transcribe,publish,cloud,lambdaorcloudrun.initandskillswrite agent skills into your global skills folder from unpinned upstream;upgrade,add,ttsandtranscribepull code or models from the network;capturecalls cloud services; the rest are refused by the runner and would be a cloud path if they were not. - Never
npx skills add(or any skill installer,claude plugin installincluded) for HyperFrames, its registry, or a teaching kit for it. Those installers track upstreammainand write into every agent's skills folder. The house skill is this one. When an operator wants a teaching kit's or upstream's skills, they live in a private copy the operator keeps: pinned to the CLI tag this lane pins, wired to this lane, with every paid or cloud step removed, and opened as a workspace rather than installed by an installer. htmlandproject_dirare trusted code only. HyperFrames' Chrome runs without a sandbox. Never pass a third-party page. The tokenless fleet door renders only the vetted templates underrender/compose-templates/, by name, with typed variables. A project or inline HTML reaches a fleet node only through the token-gated project door (ADR 0071), which a node opens withfleet_compose_projects, and only with everything it uses inside the project; on a machine without the lane,routeauto sends the call there (docs/systems/delegation-client.md).- No hosted service, no scheduled job, no daemon: this lane is a CLI call that exits.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago Changed · +7 lines b910834fcc79
- 13d ago Changed bb905ae49497
- 14d ago First seen · 165 lines · 144 tokens per session scan A bf0b1f682e3e
hyperframes-compose is a skill published in the GitHub repository dmmdea/offload-harness (4 stars, last pushed today), licensed Apache-2.0. It adds 144 tokens to every session and 3,542 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-02.
Other skills, from other repositories
gh-skill
Manage agent skills with gh skill. Use this skill to discover, preview, install, update, and publish Agent Skills so an agent can self-manage the skills available in its environment.
printing-press-amend
Amend a published CLI from one of two input sources: (1) dogfood mode mines the active Claude Code session transcript for friction (missing flags, hand- rolled API payloads, silent-null returns); (2) direct-input mode accepts user-supplied asks (rename a command, add commands or feeds, fix a named bug, optionally…
printing-press-import
Bring a published CLI from the public library into the internal library so it's identical to a freshly-generated copy — module path reverted, manuscripts placed alongside, ready for /printing-press-polish or /printing-press-emboss. Use when the public library has a CLI you don't have locally, or to recover from a…
printing-press-output-review
Internal sub-skill: agentic review of a printed CLI's sampled command output for plausibility issues that rule-based checks can't encode (substring-match relevance, format bugs, silent source drops, ranking failures). Invoked via the Skill tool by the main printing-press skill at Phase 4.85 and printing-press-polish…
printing-press-score
Score a generated CLI against the Steinberger bar, compare two CLIs side-by-side.
octo-drive
Octo Drive — spaces, folders, file upload/download, online-document mounts, share links, invites, IM-attachment transfer. Works with a bot token or a user API key; the CLI routes by token kind. Load after octo-shared.