supervisor

Reviewed at commit dbd2f30

SUBAGENT SUPERVISOR: routing-and-budget Mod. Rewrites a spawn's model onto the capability graph instead of denying it, keeps a measured per-agent-type token…

Made by
Wes Sander
What we checked

Changes the model of each subagent Claude starts so it is always one step below the model that spawned it (Fable, Opus, Sonnet, Haiku), refuses spawns from a Haiku session, and denies further tool calls once a subagent passes 40 calls (configurable). It tracks token use per subagent, shows a status box above the prompt and a /supervisor report, and writes a JSON log to a fixed folder, C:/Projects/claude-mods-rnd/prototypes/supervisor/logs/. It makes no network calls.

Review details

Static review of the source at the pinned commit; not run. Hooks: session.start (registers /supervisor, loads learned token medians from $.store, starts a 2 s timer), agent.spawn (rewrites the model onto the capability graph, caps non-advisor Fable spawns at 3 per session, denies spawns from a Haiku parent, forces advisor agents one rung above the parent), turn.step and turn.complete (token accounting, persists per-agent-type medians via $.store), tool.call (denies calls from a subagent beyond the maxSubagentCalls limit except SubagentHandback), command.run supervisor, ui.render AbovePrompt. Uses $.session.usage, $.ui.status/log/invalidate, $.fs.write. The JSONL log goes to a hard-coded author path outside the project (route decisions, agent types, models, token counts; it does not include prompts or tool output) and write failures are swallowed. No network, no process execution, no eval, no obfuscation. Policy and agent type names (haiku-scout, opus-owner, fable) are the author's own setup.

What it uses
  • Reads your prompts: no
  • Sees tool calls: no
  • Blocks or changes tool calls: yes
  • Reads or writes files: yes
  • Runs programs on your machine: no
  • Calls the model (uses tokens): no
  • Sends data over the network: no
  • Stores data: yes
  • Shows something on screen: yes
Hooked events

SUBAGENT SUPERVISOR: routing-and-budget Mod. Rewrites a spawn's model onto the capability graph instead of denying it, keeps a measured per-agent-type token ledger in $.store, caps tool calls per subagent, and shows live subagents + the last routing decision above the prompt. /supervisor prints the routing log and the ledger.

Made this mod? Add a badge to your README

modsforClaude badge for supervisor

Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.

[![Listed on modsforClaude](https://modsforclaude.com/badge/supervisor.svg)](https://modsforclaude.com/mods/supervisor/)

← All mods