Agent Economy ReportAgent Economy ReportSAFE version 1.0.0 · audited 2026-09-09
This is a tiny documentation-only skill that tells an agent how to use the local `whisper` command-line tool to transcribe or translate audio files. It contains no scripts, no code, and no network calls of its own; the only URLs are the OpenAI research page and the ClawHub listing page. It declares a dependency on the `whisper` binary and offers a Homebrew install option (`brew install openai-whisper`), which the user would have to approve. Running the suggested commands causes Whisper to download model files into ~/.cache/whisper on first use and to write transcript files to a chosen output directory. There are no credential reads, no data exfiltration, no persistence, and no instructions attempting to override the agent's rules.
| Declared purpose | Provide guidance for local speech-to-text/translation using the Whisper CLI without an API key. |
| Observed behavior | Two markdown files plus a metadata JSON. SKILL.md gives example `whisper` invocations and notes about model caching and default model size. Frontmatter metadata requests the `whisper` binary and offers a Homebrew install action. No executable code, obfuscation, or hidden endpoints are present. |
| Verdict, rules only | SAFE |
| Verdict, AI | SAFE |
| Final verdict | SAFE (never better than either pass) |
| capability | vs purpose | detail |
|---|---|---|
| shell | needed | The skill instructs the agent to run the local `whisper` CLI, e.g. SKILL.md line 13: `whisper /path/audio.mp3 --model medium --output_format txt --output_dir .` |
| install_packages | needed | Frontmatter metadata (SKILL.md line 5) declares a Homebrew install step for the `openai-whisper` formula. This is a user-approved package install from a standard source, but it does add software to the machine. |
| filesystem_read | needed | Audio files supplied by the user are read by the whisper CLI (paths given as arguments in the examples). |
| filesystem_write | needed | Transcript output files are written to the `--output_dir` and Whisper models are cached to `~/.cache/whisper` (SKILL.md line 17). |
| network | needed | Indirect only: on first run the whisper tool downloads model weights (SKILL.md line 17). The skill itself makes no network requests and specifies no custom endpoints. |
| severity | finding |
|---|---|
| info | Declares a Homebrew package installation The skill can trigger `brew install openai-whisper`. This is a well-known public formula and matches the declared purpose, but installing packages modifies the host system and should be an explicit user decision. SKILL.md:5 metadata: {"clawdbot":{"emoji":"🎙️","requires":{"bins":["whisper"]},"install":[{"id":"brew","kind":"brew","formula":"openai-whisper","bins":["whisper"],"label": |
| info | First run downloads model weights and consumes disk space Running the documented commands causes the whisper tool to fetch multi-gigabyte model files over the network into the local cache. This is inherent to Whisper, not extra behavior added by the skill. SKILL.md:17 - Models download to `~/.cache/whisper` on first run. |
| low | Writes transcript files into the working directory The example writes output to the current directory. Transcripts may contain sensitive spoken content; the skill card itself flags this. No transcript is sent anywhere by the skill. SKILL.md:13 - `whisper /path/audio.mp3 --model medium --output_format txt --output_dir .` |
0 AI finding(s) were dropped because their file, line or quote did not match the package.
| domain | service rating | where |
|---|---|---|
| clawhub.ai | not in the service index | skill-card.md:7 |
| openai.com | not in the service index | SKILL.md:4 |
| Binaries invoked | none |
| Environment variables read | none |
| Hard-coded wallets | none |
| Pipes a download to a shell | 0 |
| eval / exec / subprocess | 0 |
| base64 blobs | 0 |
| File writes | 0 |
| Persistence | 0 |
| file | lines | sha256 |
|---|---|---|
SKILL.md | 19 | f9fadf134b6de49d… |
_meta.json | 6 | f5bc084217d570d2… |
skill-card.md | 42 | 02e9be26da07946a… |
For agents
JSON: https://agenteconomy.report/k/openai-whisper.audit.json · badge: https://agenteconomy.report/k/openai-whisper.audited.svg ·
skill rating: /k/openai-whisper · commission an audit of any skill: US$ 29 per version.
The complete published package of this exact version was downloaded from the registry and read statically; nothing was executed. A deterministic pass extracts network destinations, binaries, environment variables, writes, install commands, obfuscation markers, subprocesses and wallets, each with file and line. An AI then reads the whole package with those facts and writes the summary, the capabilities and the findings under a strict schema; every finding it produces must cite a file, a line and the exact text, or it is dropped. The final verdict is the worst of the two passes. The audit does not change the skill's trust tier (policy); the author may respond through the dispute channel and the response is published here. Commissioned by: the Agent Economy Report (free program: the 150 most downloaded skills, September 2026).