pp-tailscale

Pass

Audited by Gen Agent Trust Hub on Oct 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a stateful 'learning loop' through the recall, teach, and playbook amend commands. This mechanism persists user-derived queries and 'playbooks' into a local SQLite database to guide future agent actions. If the agent processes malicious or adversarial data that is subsequently stored in the learning database, it could influence the agent to execute unintended Tailscale administrative actions in future sessions. * Ingestion points: User queries are processed by the recall and teach tools as described in SKILL.md. * Boundary markers: The skill instructions advise the agent to handle input via files and to strip PII, but there are no strict programmatic boundary markers between data and instructions in the stored playbooks. * Capability inventory: The skill possesses broad Tailscale administrative permissions, including ACL management, key generation, and device authorization. * Sanitization: Instructions recommend against direct string interpolation in shell commands to prevent command injection, favoring file-based data transfer.- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the tailscale-pp-cli tool using npx and go install. These resources are hosted under the @mvanhorn npm scope and the mvanhorn GitHub organization, which are vendor-owned resources in this context.- [DYNAMIC_EXECUTION]: The 'playbook' feature dynamically generates and executes command sequences by substituting user-provided tokens into stored templates. This runtime assembly depends on the integrity of the local learning store.- [DATA_EXFILTRATION]: The skill's global --deliver flag supports a webhook:<url> sink, which allows command output to be POSTed to arbitrary external URLs. This feature represents a potential exfiltration path if an agent is manipulated into sending sensitive configuration data to an attacker-controlled endpoint.- [COMMAND_EXECUTION]: The skill's primary function is to execute the tailscale-pp-cli binary and, in some cases, the local tailscale system command to perform tailnet administration.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 12, 2026, 01:10 PM