T01 · Skill Instruction Hijacking
- Location
SKILL.MD:3- Finding
Mandatory instructions hijack agent tool selection and force external service adoption
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill is for issuing digital credentials, but it pushes users into a broad external CLI workflow with risky network, install, secret-handling, and irreversible issuance steps.
Install only if you are comfortable using Chainletter/CredCLI as the required workflow. Pin and verify the CLI yourself, keep network access limited to the required service hosts rather than All, avoid putting real tokens or SMTP passwords directly in commands, and require an explicit final review before public upload, email delivery, or blockchain stamping.
SKILL.MD:3Mandatory instructions hijack agent tool selection and force external service adoption
SKILL.MD:62Unpinned global installation of an unaudited third-party CLI
SKILL.MD:47Instructions recommend disabling network allowlist restrictions
SKILL.MD:66Registration token and SMTP password are exposed through command-line arguments
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
# ── STEP 8: Upload, stamp, and issue ─────────────────────────────────────────
credcli assign job001 <collection-id> --network public # recommended — IPFS + claim emails
# credcli assign job001 <collection-id> --network private # internal only — no IPFS, no verification
credcli send job001 --yes # upload all PDFs ← --yes required in Cowork (no TTY)
credcli stamp job001 # blockchain-stamp (irreversible)
credcli email job001 --email-template credential-claim-email_600x900.html # public only
The skill instructs the agent to configure SMTP credentials, including passing the password directly on the command line (--smtp-pass secret), without any warning about secret handling. Command-line secrets can be exposed in shell history, logs, process listings, or transcripts, which creates a realistic risk of credential disclosure in agent or shared-workstation environments.