type: decision
status: active
timestamp: 2026-07-01
tags: [fleet, agents, cut, gocode, codeep, claurst, coddy]

Fleet cut 2026-07-01 — drop gocode, Codeep, Claurst, Coddy

Reduce coding-agent fleet from 10 → 6. Remove marginal agents (gocode, Codeep, Claurst, Coddy) after audit found no differentiating usage patterns.

Fleet cut 2026-07-01

Decision

Fleet drops from 10 → 6 supported coding agents. Removed:

AgentRepoReason
gocodeAlleyBo55/gocode (36⭐)Small ecosystem, low active usage, redundant with Codex/qwen-code
CodeepVladoIvankovic/CodeepSingle maintainer, small ecosystem, keytar+prebuild-install deprecation chain (see closed #6)
ClaurstKuberwastaken/claurst (9.8K⭐)ACP-first model diverges from MCP fleet parity; not worth the sync effort for one agent
Coddycoddy-project/coddy-agent (66⭐)Similar profile to gocode; recently added but never became daily-driver

Fleet after cut:

AgentTypeRole
Claude CodeCLI (primary)corp laptop, Bedrock-routed
ZCodeGUI IDEsecondary GUI
OpenCodeCLIpersonal-laptop primary CLI
Kilo CodeVS Code extin-editor agent
AntigravityStandalone IDEsecondary IDE
MiMoCodeCLIXiaomi ecosystem CLI

Why

Each agent added to the fleet costs:

  1. Sync tax — MCP config, AGENTS.md pointer, skills junctions, PATH management
  2. Cognitive tax — remembering which agent uses which config path
  3. Maintenance tax — every .mcp.json change requires re-running scripts/sync-mcp-configs.mjs
  4. Failure-mode tax — every added agent expands the surface where “why doesn’t this work” investigations start

The 4 removed agents individually cost more than they saved. Audit at 2026-07-01 (see oss-audit-2026-07-01) found:

All 11 closed with a neutral “we’re consolidating, not something wrong with the tool” note.

Executed cleanup

Why this supersedes fleet-cut-to-4-agents-2026-06-29

The 2026-06-29 decision cut Cline and locked at 4 agents (Claude Code, OpenCode, Kilo, Antigravity). Between then and 2026-07-01:

The 2026-06-29 file’s “fleet = 4” claim became stale within 48h. This decision replaces it and locks the new count at 6 explicitly — but with the caveat that any future fleet addition MUST come with a grill-me session per agent-fleet-parity.

What we still track upstream

The removals do NOT mean we stop caring about these projects. They just leave our default fleet. If a specific engagement needs one (e.g. Claurst’s ACP for a particular integration), it can be spun up ad-hoc without adding to the standing fleet.

Cross-refs


Edit on GitHub · Back to index