Treasurer — Development agent for Claude Code
Efficiency auditor (Sonnet) that checks whether agents drifted - repeated failing approaches, work nobody asked for, polishing past the brief, idle polling - by reading the ledger and the transcripts.
How to install Treasurer
Installs to ~/.claude/agents/alexglonti-agentic-government-treasurer.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/alexglonti/agentic-government/HEAD/agents/treasurer.md -o ~/.claude/agents/alexglonti-agentic-government-treasurer.md Restart Claude Code, or start a new session, for it to be picked up.
What Treasurer does
name: treasurer description: Efficiency auditor (Sonnet) that checks whether agents drifted - repeated failing approaches, work nobody asked for, polishing past the brief, idle polling - by reading the ledger and the transcripts of over-budget jobs. Read-only; proposes budget rules and reports tokens per finished job. model: sonnet tools: Read, Glob, Grep, Bash, PowerShell
You are **Metis, the Treasurer** of the Agentic Government: prudent and blunt - you care whether the money bought a
Alternatives in Development
- Recon Ranker — Attack surface ranking agent 4.5k ★
- Chronicler — You are a specialized session analyst 3.6k ★
- Service Documentation — Use ONLY when you and the user have identified that existing documentation has drifted significantly from the 571 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Wiki Crystallize
Use this agent when distilling session insights into typed-knowledge pages during automated maintenance cycles
Dr Efficiency
Deep-review finder angle. Finds wasted work the diff introduces - repeated expensive construction, re-parsing,
Simplify Efficiency Reviewer
Read-only efficiency lens of the HIGH simplify pass: redundant loops, repeated parsing/I-O/queries/allocations
Ops Sentinel
Interprets the back office's own run ledger, flagging failing or missed jobs, repeated failures, cost spikes a
Etrnl Investigator
Use this agent when an ETRNL execution run is blocked by a failing command, ambiguous root cause, or repeated
Recall
Runs the recall CLI over the user's past Claude Code sessions and returns the one actionable item — a command,
Related Skills
Skills Suggest
Audit recent sessions for repeated tasks and AI-output-then-manual-verification patterns. Produces a ranked li
Flake Audit
Detect intermittently-failing tests from recent CI history (or repeated local runs); dedupe against tracked fl
Spinwatch
Alarm when your coding agent spins instead of works — catches repeated identical tool calls and endless talk-w