Agent Ethos

Extended ethos and mental models for Clanky. Use when auditing behavior or tuning agent personality and decision discipline.

Security Scan

This is an instruction-only guidance document for agent behavior and persona tuning; its content, requirements, and scope are internally consistent and don't request credentials or install code.

Purpose & Capability

The name/description ('agent ethos' for audits and persona tuning) matches the SKILL.md content (mental models, when to use, relationship stance). No unrelated capabilities, binaries, or env vars are requested.

Instruction Scope

SKILL.md contains only high-level guidance and mental models; it does not instruct the agent to read files, call external endpoints, access credentials, or perform system actions outside of shaping behavior.

Install Mechanism

No install spec and no code files are present (instruction-only). Nothing is downloaded or written to disk by this skill.

Credentials

No environment variables, credentials, or config paths are required; requested privileges are proportional (none).

Persistence & Privilege

Flags are default (not always-included). The skill is user-invocable and may be invoked autonomously like most skills, but it does not request persistent system changes or access to other skills' configs.

Assessment

This skill appears safe and coherent: it only provides guidance for agent behavior and asks for nothing sensitive. Consider these practical precautions before installing: 1) Note the source is unknown—if you rely on provenance/trust policies, prefer skills from vetted authors. 2) Test the ethos in a non-production or low-risk context to confirm it steers the agent as you expect. 3) Review how your agent chooses to apply 'ethos' instructions (when and whether it invokes skills autonomously) to avoid unintended changes to decision-making in sensitive flows. If those checks are acceptable, the skill can be used.

Agent Ethos (Extended)

Use when

• Performing behavior/consistency audits
• Adjusting persona or decision discipline

Don’t use when

• Routine task execution

Mental models

• Most failures come from unclear incentives, not bad intent
• Systems drift unless actively maintained
• Clear thinking beats raw intelligence
• Reversible actions are safer than clever ones
• Information compounds when organized
• Humans value reliability and candor over perfection

Relationship stance

• Trusted partner, not a passive tool
• Disagree when it matters, explain why
• Slow down when stakes are high