Agent Economy Report

Skill code audit

CAUTION version 1.0.0 · audited 2026-09-17

himalaya @lamelas

This skill is a set of instructions for using the Himalaya email tool from a terminal. It tells an agent how to install the tool, configure email accounts, list/read/send/forward/delete messages, manage folders and attachments, and compose rich emails. The instructions require shell access and access to the user's email credentials, and they also show how to download attachments. No hidden network destinations, obfuscated code, persistence mechanisms, or attempts to override the agent's rules were found. Because it deals with sensitive credentials and can send email or download files, a user should review it carefully before installation.

Declared purposeCLI to manage emails via IMAP/SMTP. Use `himalaya` to list, read, write, reply, forward, search, and organize emails from the terminal. Supports multiple accounts and message composition with MML (MIME Meta Language).
Observed behaviorThe package is documentation and examples. It declares a Homebrew install for himalaya and instructs shell commands such as `himalaya account configure`, `himalaya envelope list`, `himalaya template send`, and `himalaya attachment download`. Configuration examples include email account passwords, command-based secret retrieval, and OAuth2 token settings. It also warns against plaintext password use.
Verdict, rules onlySAFE
Verdict, AICAUTION
Final verdictCAUTION (never better than either pass)

Capabilities

capabilityvs purposedetail
shellneededThe skill repeatedly instructs running himalaya CLI commands and a bash heredoc, for example SKILL.md line 27 and line 124.
install_packagesneededSKILL.md metadata instructs installing himalaya via Homebrew using the formula himalaya.
filesystem_writeneededCommands can create or update configuration at ~/.config/himalaya/config.toml and download email attachments to disk.
filesystem_readneededThe skill requires reading the Himalaya config file and can attach local files to outgoing email through MML.
credentialsneededConfiguration examples include email passwords and OAuth secrets. The skill recommends pass/keyring but also documents a raw password field.
networkneededHimalaya operations use IMAP/SMTP/OAuth to user-configured mail servers. The only domains present are clawhub.ai, github.com, and an example placeholder provider.com.

Findings

severityfinding
infoDeclares Homebrew installation of himalaya
The skill metadata tells the agent/runtime to install the himalaya CLI via Homebrew. This is a third-party package install and is consistent with the skill's declared purpose.
SKILL.md:5 metadata: {"clawdbot":{"emoji":"📧","requires":{"bins":["himalaya"]},"install":[{"id":"brew","kind":"brew","formula":"himalaya","bins":["himalaya"],"label":"Inst
infoInstructs shell-based account setup
The skill instructs running an interactive Himalaya command to configure an email account. This requires shell access to the himalaya binary.
SKILL.md:27 himalaya account configure
lowCan send email from supplied text
The skill shows how to pipe a composed message into `himalaya template send`. An agent following this could send email from configured accounts; this is part of managing email but should be reviewed.
SKILL.md:124 cat << 'EOF' | himalaya template send
lowDownloads email attachments to disk
Attachment downloads write files to the local filesystem, and line 189 shows writing to a chosen directory. This is expected for email management, but attachments can carry malicious content.
SKILL.md:184 himalaya attachment download 42
mediumDocuments raw password storage
The example stores an email password directly in a TOML config file. If copied, this places credentials in plaintext; line 34 notes 'testing only, not recommended' and line 41 recommends a command instead.
references/configuration.md:20 backend.auth.raw = "your-password"
lowConfigures command-based credential retrieval
The skill instructs Himalaya to execute a command such as `pass` or `security find-generic-password` to retrieve email credentials. This is a shell-command based secret source and should be reviewed.
references/configuration.md:41 backend.auth.cmd = "pass show email/imap"
infoOAuth URL is a placeholder
The OAuth2 example points to provider.com, an example placeholder domain, not a hidden destination. No hard-coded credential exfiltration endpoint is present.
references/configuration.md:151 backend.auth.auth-url = "https://provider.com/oauth/authorize"
infoPublisher discloses credential exposure risk
The skill card explicitly warns about credential exposure and recommends command-based secret storage on line 26.
skill-card.md:24 Risk: Email credentials could be exposed if plaintext password examples are copied into shared, backed-up, or source-controlled configuration files.

0 AI finding(s) were dropped because their file, line or quote did not match the package.

Network destinations in the code

domainservice ratingwhere
clawhub.ainot in the service indexskill-card.md:9
github.comnot in the service indexSKILL.md:4
provider.comnot in the service indexreferences/configuration.md:151

Other facts

Binaries invokedbash
Environment variables readEDITOR
Hard-coded walletsnone
Pipes a download to a shell0
eval / exec / subprocess0
base64 blobs0
File writes0
Persistence0

Files audited

filelinessha256
SKILL.md2172265054da976a3d9…
_meta.json6da9754eb04a63644…
references/configuration.md174af8ce863062d00f6…
references/message-composition.md182ac84822e20489d4b…
skill-card.md505dfb2e7dfd9fe262…

For agents

JSON: https://agenteconomy.report/k/himalaya.audit.json · badge: https://agenteconomy.report/k/himalaya.audited.svg · skill rating: /k/himalaya · commission an audit of any skill: US$ 29 per version.

How this is computed

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).