froggo-mission-control: Skill for Claude Code

Agent Evaluation

Purpose

Maintain a healthy, well-defined agent team. Catch stale definitions, identify capability gaps, surface underperforming agents, and ensure every agent has the skills and scope needed to do their job.

Trigger Conditions

Load this skill when:

• Asked to evaluate or audit a specific agent
• Reviewing the full agent roster for coverage gaps
• Onboarding a new agent (to check for overlap or missing skills)
• Reviewing agent task history for quality or performance
• Generating an HR report on team health

Procedure

Step 1 — Load Agent Context

For each agent under review:

1. Read their CLAUDE.md / soul file in .claude/agents/<name>.md
2. Read their memory file at ~/mission-control/agents/<name>/MEMORY.md if it exists
3. Check task history via mcp__mission-control-db__task_list { "assignedTo": "<agent-name>" }
4. Note: assigned skills, role description, tool access, and output paths

Step 2 — Evaluate Definition Quality

Score each agent definition across these dimensions:

Dimension	What to Check	Score (1-3)
Clarity	Is the role clearly defined? Can you tell what this agent does vs doesn't do?
Completeness	Does it have: character, responsibilities, workflow, memory protocol, library output?
Skills Coverage	Are the right skills listed? Are there obvious gaps?
Scope Boundaries	Is it clear what this agent escalates vs handles solo?
Collaboration Norms	Are the key collaborators named? Are handoff rules defined?

Scoring: 1 = needs revision, 2 = acceptable, 3 = strong

Step 3 — Review Task History

For the trailing 30 days:

• How many tasks were assigned?
• What percentage were completed vs stalled vs human-reviewed?
• Were any tasks escalated that this agent should have handled?
• Were tasks completed without logging activity? (anti-pattern — flag it)

Quality signals to check:

• Task output saved to correct library path
• Activity logged during execution (not just at end)
• Status transitions followed the correct pipeline