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.
git clone --depth 1 https://github.com/punt-labs/voxWrote 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/commands/punt-labs/vox/call)<a href="https://agentmods.dev/commands/punt-labs/vox/call"><img src="https://agentmods.dev/badge/commands/punt-labs/vox/call/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/commands/punt-labs/vox/call"><img src="https://agentmods.dev/badge/commands/punt-labs/vox/call.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 | $0.00013 | $0.01355 |
| Opus 5.5 | $0.00005 | $0.00542 |
| Sonnet 5.5 | $0.00003 | $0.00271 |
| Haiku 4.5 | $0.00001 | $0.00136 |
Grade C, and why
call scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured today.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Reaches for credential fileshighPrivilege escalation
SSH keys, cloud credentials, git-credentials, .npmrc, /etc/shadow: reading these is how a config file becomes a credential leak.
`.git-credentials`, and the secret-looking patterns in the repo's How it starts
The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/call command
Conversation Mode: a live voice call between the human and a call agent,
mediated by voxd. The call agent is a separate, read-only process. It cannot
see this session, so you hand it a context snapshot when the call starts, and
it hands you a written summary when the call ends. There is no mic:call MCP
tool yet -- every verb shells out to the vox call CLI.
Usage
/call start <topic>-- start a real call about<topic>: real microphone capture, transcribed by ElevenLabs./call start <topic> --script <path>-- dev/test path: scripted turns from a JSON Lines file ({"text": ..., "confidence": ...}per line) instead of the microphone. For demos and CI./call stop-- ask the running call to hang up.
Implementation
start
-
Write the context snapshot yourself. You are the primary session; the call agent starts with nothing but what you give it. In plain prose, cover:
- the topic (
<topic>from$ARGUMENTS) and what the human wants from the call; - the current task: what you are working on, the branch, any bead ID;
- the files that matter, by path;
- the gist of the recent conversation: decisions made, open questions.
Before writing it, search with the quarry MCP
findtool (a query drawn from the topic) and fold the relevant hits in: what each says, and its source path. Never put the query on a shell command line -- topic text in a shell string can run as a command. Look up external library docs as well if the topic needs them. Keep the snapshot to a briefing -- a page or two, never whole files, never a secret value. - the topic (
-
Start the call in the background, with the snapshot on stdin through a quoted heredoc so nothing in it is expanded by the shell. The heredoc delimiter is fresh for every call:
VOX_CONTEXT_followed by 8 hex digits you choose at random now -- never a value written in this file -- so no text folded into the snapshot -- a quarry hit, a file excerpt -- can contain the line that ends it early and runs what follows as shell. Check that no line of the snapshot is exactly the delimiter; if one is, pick another. Run this Bash command withrun_in_background: true--vox call startruns for the whole call:
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.
- today Changed · +61 lines · +3 tokens per session scan A → C 81ef87177cbc
- 2mo ago First seen · 49 lines · 10 tokens per session scan A 5c5d4dcf36c1
call is a command published in the GitHub repository punt-labs/vox (3 stars, last pushed today), licensed MIT. It adds 13 tokens to every session and 1,355 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it C with 1 finding (reaches for credential files). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
call compared with commands from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| callpunt-labs/vox | Command | 3 | today | MIT | C | 13 | Claude Code |
| feedback-to-uspunt-labs/prfaq | Command | 26 | 3d ago | MIT | A | 19 | Claude Code |
| badgepunt-labs/prfaq | Command | 26 | 3d ago | MIT | A | 11 | Claude Code |
| importpunt-labs/prfaq | Command | 26 | 3d ago | MIT | A | 15 | Claude Code |
| setuppunt-labs/z-spec | Command | 5 | yesterday | MIT | A | 11 | Claude Code |
| contracts-devpunt-labs/z-spec | Command | 5 | yesterday | MIT | A | 17 | Claude Code |
| code2model-devpunt-labs/z-spec | Command | 5 | yesterday | MIT | A | 11 | Claude Code |