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:
| Agent | Repo | Reason |
|---|---|---|
| gocode | AlleyBo55/gocode (36⭐) | Small ecosystem, low active usage, redundant with Codex/qwen-code |
| Codeep | VladoIvankovic/Codeep | Single maintainer, small ecosystem, keytar+prebuild-install deprecation chain (see closed #6) |
| Claurst | Kuberwastaken/claurst (9.8K⭐) | ACP-first model diverges from MCP fleet parity; not worth the sync effort for one agent |
| Coddy | coddy-project/coddy-agent (66⭐) | Similar profile to gocode; recently added but never became daily-driver |
Fleet after cut:
| Agent | Type | Role |
|---|---|---|
| Claude Code | CLI (primary) | corp laptop, Bedrock-routed |
| ZCode | GUI IDE | secondary GUI |
| OpenCode | CLI | personal-laptop primary CLI |
| Kilo Code | VS Code ext | in-editor agent |
| Antigravity | Standalone IDE | secondary IDE |
| MiMoCode | CLI | Xiaomi ecosystem CLI |
Why
Each agent added to the fleet costs:
- Sync tax — MCP config, AGENTS.md pointer, skills junctions, PATH management
- Cognitive tax — remembering which agent uses which config path
- Maintenance tax — every
.mcp.jsonchange requires re-runningscripts/sync-mcp-configs.mjs - 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:
- gocode: MCP was undocumented in AGENTS.md but present in the tool; skills format required a bespoke
SKILL.mdvariant. Two issues filed (#31 README, #32 SKILL.md interop) closed on removal. - Codeep: 4 issues filed (#3 AGENTS.md discovery, #4 Windows hooks, #5
.mcp.jsonauto-discover, #6 keytar/prebuild-install deprecation chain). Closed on removal. - Claurst: ACP protocol, docs unclear on MCP support. 2 issues filed (#201 MCP docs, #202 AGENTS.md), closed on removal.
- Coddy: 3 issues filed (#41 AGENTS.md, #42 Windows PATH docs, #43 YAML schema), closed on removal.
All 11 closed with a neutral “we’re consolidating, not something wrong with the tool” note.
Executed cleanup
npm uninstall -g codeep claurst— done- Deleted
~/bin/gocode.exe,~/bin/coddy.exe— done - Removed
.agents/{gocode,coddy,codeep,claurst}/from workspace — done - Filed 11 upstream issues STAY OPEN. Initial close on removal was reversed 2026-07-01 — the issues describe real gaps in each tool and are useful signal to maintainers regardless of our own usage decisions.
- Updated
AGENTS.mdfleet table + counts — done in this commit
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:
- MiMoCode was added (Xiaomi ecosystem, personal machine)
- ZCode was added (GUI IDE, dual-machine)
- gocode + Coddy were added (from binary releases)
- Codeep + Claurst were added (from npm)
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
oss-audit-2026-07-01— the audit that surfaced the case for cuttingno-fork-divergence— pairs with this: don’t add complexity we can’t upstreamagent-fleet-parity— every fleet member needs identical sync