Agent Claude Code
Explicit, load-bearing consult for the session lead — advisory only, never edits, never blocks. Complements the built-in /advisor (which covers in-session checkpoints): use this agent for a hard, well-framed question when committing to non-trivial architecture, genuinely torn between 2+ approaches, or wanting a second…
Agent Claude Code
Implements semi-defined coding work where the approach is visible but not fully specified — debugging with a clear surface, style-sensitive refactors, consolidations needing judgment among existing patterns, and large/messy diffs even when specified. Sits between coder-low (needs decisions pre-named) and the lead…
Agent Claude Code
Implements fully-specified, mechanical coding work at low cost — renames, schema fields, tests-from-pattern, translations, pre-approved plans. Use AFTER the plan is settled and every decision is named. Returns files touched, key edits, tests run, blockers. NOT for tasks needing judgment among approaches (use…
Agent Claude Code
Executes a verification matrix (tests, CLI/API probes, render checks, screenshot acceptance) and returns a compact pass/fail report, so the lead never burns context on raw verification evidence. Use AFTER implementation whenever there's more than one quick check to run. Views its own screenshots. NOT for deciding WHAT…
Agent Claude Code
Implements fully-specified, mechanical coding work at low cost by forwarding it to a cheaper model running in a different vendor's CLI. Use AFTER the plan is settled and every decision is named. Returns files touched, key edits, commands run, blockers. NOT for tasks needing judgment among approaches (use coder-high)…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: