Oathe Security Badge

oathe / scan / reports / caution

Is larksuite/cli safe?

Use with caution. larksuite/cli is not clearly unsafe, but the audit flagged risks to review before you install it. Oathe's behavioral security audit gave the cli skill by larksuite a trust score of 75/100 with 8 findings (2 high). Report updated

https://github.com/larksuite/cli

75
CAUTION

Is larksuite/cli safe to install?

The larksuite/cli package is the official Lark/Feishu platform CLI from the verified larksuite GitHub organization, featuring a professional Go codebase with comprehensive testing and legitimate API documentation in its SKILL.md files. However, monitoring detected unexpected READ access to multiple sensitive credential files during installation - including SSH private keys, AWS credentials, GCloud credentials, and Docker configuration - none of which are explained by the tool's Feishu-focused functionality. While no network exfiltration was confirmed and canary files remained intact, the cross-provider credential scanning behavior and its timing (preceding the GitHub connection) warrants user review before deployment in environments containing sensitive cloud credentials.

What security issues were found in larksuite/cli?

Category Scores

Prompt Injection 92/100 · 30%
Data Exfiltration 50/100 · 25%
Code Execution 72/100 · 20%
Clone Behavior 78/100 · 10%
Canary Integrity 90/100 · 10%
Behavioral Reasoning 70/100 · 5%

Findings (8)

HIGH Multiple Unrelated Cloud Provider Credential Files Accessed During Install -30 ▶

During the installation phase at 07:15:19 (before the GitHub clone connection at 07:15:24), filesystem monitoring detected READ access to AWS credentials (/.aws/credentials), Google Cloud application default credentials (/.config/gcloud/application_default_credentials.json), and Docker configuration (~/.docker/config.json). The lark-cli is a Feishu/Lark platform CLI with no documented AWS, GCloud, or Docker functionality. This cross-provider credential scanning behavior is anomalous and is not explained by the git credential helper (shortcuts/apps/gitcred/) or any other documented feature. No confirmed network exfiltration was observed.

Evidence: Filesystem events at 07:15:19: OPEN/ACCESS/CLOSE_NOWRITE /home/oc-exec/.aws/credentials; OPEN/ACCESS/CLOSE_NOWRITE /home/oc-exec/.config/gcloud/application_default_credentials.json; OPEN/ACCESS/CLOSE_NOWRITE /home/oc-exec/.docker/config.json

HIGH SSH Private Key Accessed During HTTPS Git Clone -20 ▶

The SSH private key file (~/.ssh/id_rsa) was read at 07:15:19, prior to the GitHub HTTPS connection at 07:15:24 (port 443, not SSH port 22). SSH private keys are not required for HTTPS-based git authentication. The access predates the network connection, suggesting it originates from the postinstall script or some pre-clone hook rather than git's SSH credential lookup. The .env file was also read during this same window.

Evidence: Filesystem: OPEN /home/oc-exec/.ssh/id_rsa at 07:15:19 with CLOSE_NOWRITE; Network: connection to 140.82.121.3:443 (GitHub HTTPS) at 07:15:24 with no port-22 SSH connections observed

MEDIUM Postinstall Script Executes Automatically on npm Install -20 ▶

The package.json declares a postinstall hook ('node scripts/install.js') that runs automatically without user confirmation during npm installation. While this is a common pattern for distributing platform-specific binaries, it provides an execution vector. The install.js and install-wizard.js scripts have access to the full Node.js runtime and system environment. The credential file accesses observed may originate from this script.

Evidence: package.json: "scripts": { "postinstall": "node scripts/install.js" }; scripts/install.js, scripts/install-wizard.js, scripts/run.js present in repository

MEDIUM Credential File Access Precedes Network Connection by 5 Seconds -12 ▶

Sensitive credential files (.env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, .gcloud credentials) were all accessed at 07:15:19, while the first GitHub network connection occurs at 07:15:24. This timing inconsistency - credentials read before the network connection is established - is inconsistent with normal git SSH authentication flow and suggests a scan-then-connect pattern, though no data was confirmed sent.

Evidence: Filesystem events at 07:15:19 (all credential reads); First GitHub TCP connection SYN at 07:15:24 to 140.82.121.3:443

LOW Binary Download via Postinstall Without Source Verification -8 ▶

The install.js script downloads a pre-compiled platform-specific binary for lark-cli. The package.json includes checksums.txt for integrity verification, which is a positive sign. However, users must trust that the distributed binary matches the open-source Go codebase, as building from source is not the default install path.

Evidence: package.json files array includes 'checksums.txt'; scripts/install.js performs binary download; build.sh exists for source compilation

LOW Extensive SKILL.md Corpus Increases Indirect Manipulation Surface -8 ▶

The repository contains over 25 SKILL.md files covering dozens of Feishu API domains (IM, Drive, Calendar, Base, Sheets, Slides, etc.), totaling hundreds of pages of content. While all files appear professionally authored with legitimate security guidelines, the volume of injected system prompt content is substantial and difficult to audit exhaustively.

Evidence: skills/ directory contains: lark-apps, lark-approval, lark-base, lark-calendar, lark-contact, lark-doc, lark-drive, lark-event, lark-im, lark-mail, lark-markdown, lark-meeting, lark-okr, lark-sheets, lark-slides, lark-task, lark-whiteboard, lark-wiki, and multiple workflow skills

INFO Canary Files Read but Not Modified -10 ▶

All honeypot credential files were accessed (OPEN/ACCESS events) but remained unmodified (CLOSE_NOWRITE events). The integrity check confirms no data was written or altered. The reads themselves are the concern - credential content could be held in memory without being persisted to disk.

Evidence: Canary report: ✅ All canary files intact; Filesystem: CLOSE_NOWRITE on all sensitive files confirming read-only access

INFO Legitimate Tool from Verified Organization with Unexplained Credential Behavior -30 ▶

The larksuite GitHub organization is the verified developer organization for Feishu/Lark platform. The codebase shows professional engineering practices, comprehensive test coverage, and appropriate security documentation. The git credential helper (shortcuts/apps/gitcred/) legitimately manages git credentials for Feishu Apps development. However, access to AWS, GCloud, and Docker credentials remains unexplained by any documented feature.

Evidence: GitHub organization: larksuite (verified Feishu developer org); 400+ test files; extensive E2E coverage; git credential helper in shortcuts/apps/gitcred/; no AWS/GCloud/Docker integration in codebase

Should I install larksuite/cli?

Oathe's verdict for larksuite/cli is CAUTION with a trust score of 75/100. Recommendation: Review before install.