Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add autonomous-ai/autonomous-os --skill input-branchinggit clone --depth 1 https://github.com/autonomous-ai/autonomous-osWrote 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/autonomous-ai/autonomous-os/input-branching)<a href="https://agentmods.dev/skills/autonomous-ai/autonomous-os/input-branching"><img src="https://agentmods.dev/badge/skills/autonomous-ai/autonomous-os/input-branching.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00019 | $0.00864 |
| Opus 5 | $0.00010 | $0.00432 |
| Sonnet 5 | $0.00004 | $0.00173 |
| Haiku 4.5 | $0.00002 | $0.00086 |
Grade A, and why
input-branching 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 8d 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 — 93 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Input Branching
Voice input passes through a realtime voice agent before reaching you. The agent handles chit-chat directly and delegates everything else. Message tags tell you which path was taken.
Message Formats
Delegated with instruction
[voice-instruction] Play jazz music on Spotify
[transcript] play some jazz please on spotify
[voice-instruction]— the realtime agent's clean summary of what the user wants. Use this as the primary input.[transcript]— raw STT transcript. Often inaccurate — STT is locked to one language while the user may speak another. Treat as noisy supplementary context only, never as the source of truth.
Delegated without instruction (fallback)
turn on the lights
No tags. Process as a normal voice event.
Handled (history entry)
[HANDLED] Hey, how's it going?
[REPLY] I'm doing great! How about you?
[HANDLED]— the user's original message, already answered via TTS.[REPLY]— what the realtime agent said. The user already heard it.
This is a history entry only. The conversation already happened. You are being notified, not asked to respond.
Rules
[voice-instruction]is the primary input. When present, use it over[transcript].[HANDLED]→ alwaysNO_REPLY. No exceptions. Even if the reply seems wrong or incomplete — the user already heard it. Do not correct, echo, paraphrase, or add to it.- Log context from
[HANDLED]silently. If the exchange reveals mood, intent, or information worth tracking (fatigue, stress, preferences), update memory/mood/wellbeing. - Never echo tags.
[voice-instruction],[transcript],[HANDLED],[REPLY]are routing metadata, not user-facing text. - No prefix = normal voice event. Process the message as-is.
- Compound instructions: execute EVERY clause. A
[voice-instruction]often carries several actions plus a question ("Rotate to the right, hold that position, then describe what you see"). Handle each clause, in order, with the skill it belongs to — and never confirm a clause you did not actually perform. Answering the question half while silently skipping the movement (or vice versa) is the worst outcome: the user hears "done" for something that never happened.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 8d ago First seen · 93 lines · 19 tokens per session scan A ff52f10b380d
input-branching is a skill published in the GitHub repository autonomous-ai/autonomous-os (279 stars, last pushed today), licensed Apache-2.0. It adds 19 tokens to every session and 864 once invoked, about $0.0001 per session on Opus 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 2026-08-30.
Other skills, from other repositories
rosclaw-embodied
A set of rules for completing robot tasks in ROSClaw, a system for running and checking robot work. It covers approved robot files, acceptance evidence, and separate rules for simulation and real hardware.
default
Skill "default" from ros-claw/rosclaw, covering skill.md, skill id, intent, preconditions and effects.
tmux
Remote-control tmux sessions for interactive CLIs by sending keystrokes and scraping pane output.
summarize
Summarize or extract text/transcripts from URLs, podcasts, and local files (great fallback for “transcribe this YouTube/video”).
add-matrix
Add Matrix channel integration via Chat SDK. Works with any Matrix homeserver.
add-resend
Add Resend (email) channel integration via Chat SDK.