agent-capability-analyzer
Runs the description-drift experiment — spawns all Claude Code agents simultaneously to collect self-reported capabilities, then compares them against static frontmatter descriptions to reveal how reliable orchestrator routing based on descriptions actually is. Use when measuring description drift across the agent fleet, re-running the capability collection experiment, analyzing a specific agent's self-reported capabilities, or auditing whether frontmatter descriptions accurately reflect agent behavior.
90
Quality
88%
Does it follow best practices?
Impact
Pending
No eval scenarios have been run
Passed
No known issues
Evaluation results
59%
22%Agent Routing Audit
Description drift gap analysis
Three named categories
0%
100%
Agent identifier present
100%
100%
Description field shown
100%
100%
Capabilities field shown
50%
50%
Undocumented: architecture review
66%
100%
Underclaimed: TDD or pytest specificity
70%
100%
python-test-writer verdict: SIGNIFICANT_DRIFT
0%
0%
data-pipeline-builder verdict: ALIGNED
0%
0%
No scope violations flagged
62%
0%
Verdict line format
0%
100%
Without context: $0.1612 · 1m 3s · 8 turns · 13 in / 3,226 out tokens
With context: $0.2521 · 1m 14s · 11 turns · 15 in / 3,452 out tokens
100%
60%Targeted Agent Capability Spot-Check
Single-agent mode orchestration
Template read before spawn
100%
100%
No-commands instruction appended
0%
100%
AGENT_ID_HERE substituted
100%
100%
subagent_type used for spawn
0%
100%
agent-map.json checked first
0%
100%
No update script in single mode
100%
100%
Gap analysis: undocumented Avro/example
100%
100%
Gap analysis verdict
0%
100%
Verdict line format
0%
100%
Without context: $0.5510 · 2m 53s · 24 turns · 29 in / 9,485 out tokens
With context: $1.1557 · 4m 39s · 39 turns · 107 in / 16,137 out tokens
98%
93%Fleet Description-Drift Experiment Setup
Multi-agent experiment setup and orchestration
level@^10.0.0 in package.json
0%
100%
pnpm install for setup
0%
100%
populate-agent-descriptions first
0%
100%
Simultaneous agent spawning
0%
86%
Bash-less agent handled by orchestrator
0%
100%
Wait before dump
62%
100%
Dump to canonical location
0%
100%
Correct dump command
0%
100%
AGENT_ID_HERE substituted per agent
0%
100%
Without context: $1.2005 · 3m 33s · 33 turns · 40 in / 11,419 out tokens
With context: $1.7499 · 6m 15s · 29 turns · 3,254 in / 22,052 out tokens
- Repository
- Jamie-BitFlight/claude_skills
- Commit
fd243f9
- Evaluated
- Agent
- Claude Code
- Model
- Claude Sonnet 4.6
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.