Hiring
ReviewAudited by ClawScan on May 1, 2026.
Overview
This is a coherent instruction-only hiring guide, but users should set clear limits before letting an agent spend money, hire people, share credentials, or route work to other AI agents.
Before installing or using this skill, decide what the agent is allowed to spend, which platforms it may use, whether it may hire or dispatch people without confirmation, what credentials or documents contractors may receive, and what information can be sent to external AI providers or stored in logs.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
An agent could spend money or dispatch a person for a real-world task if the user grants that authority.
The skill contemplates an agent approving paid physical tasks without case-by-case human confirmation, which is expected for hiring automation but requires explicit spending and task limits.
Set maximum budget agent can approve without human confirmation
Define written approval thresholds, maximum budgets, allowed task types, emergency rules, and cancellation procedures before using this skill for paid or physical work.
A contractor could receive more access than necessary, exposing accounts, project systems, or private data.
Contractor onboarding may involve giving third parties access credentials; that is relevant to hiring but can cross account and privilege boundaries if not tightly scoped.
Structured onboarding — Access credentials, project brief, communication channels
Use least-privilege temporary accounts, avoid sharing personal credentials, revoke access after work ends, and document exactly which permissions each contractor receives.
Sensitive prompts, documents, or task details could be shared with external model providers or other agents.
The skill recommends routing subtasks to external AI providers and agent marketplaces; this is aligned with the AI-agent hiring purpose but can send task data outside the original agent boundary.
OpenRouter — Single API, multiple models, automatic routing ... Direct APIs — OpenAI, Anthropic, Google, Mistral
Only send data approved for that provider, redact sensitive inputs, confirm provider data-retention terms, and require human approval for regulated or confidential material.
Logs could preserve confidential task data, personal information, or contractor details longer than intended.
The skill recommends persistent audit logging of agent interactions, including inputs and outputs unless hashed; this supports accountability but may retain sensitive context.
For every agent interaction, log: Timestamp, agent ID, model version; Input (or hash if sensitive); Output (or hash if sensitive)
Hash or redact sensitive inputs and outputs, set retention periods, protect audit logs, and avoid reusing logged content as trusted context without review.