chore(opencode): switch small model to duo-chat-gpt-5-4-nano

Update opencode configuration and agent docs to use
gitlab/duo-chat-gpt-5-4-nano as the small model. This keeps
costs down while preserving the intended model separation for faster
runs.

Author: Pedro Pombeiro

.agents/skills/opencode-refine/SKILL.md

@@ -36,7 +36,7 @@ Runs a single non-interactive agent session. The agent executes, prints its resp
 
 | Flag | Purpose |
 |------|---------|
-| `-m <model>` | Override model (e.g. `-m gitlab/duo-chat-haiku-4-5` for fast/cheap iterations) |
+| `-m <model>` | Override model (e.g. `-m gitlab/duo-chat-gpt-5-4-nano` for fast/cheap iterations) |
 | `--print-logs` | Show debug logs on stderr (permission checks, plugin hooks, config resolution) |
 | `-f <file> --` | Attach a file as context — **requires `--` separator before the prompt** |
 | `--format json` | Machine-readable event stream for scripting |
@@ -67,7 +67,7 @@ opencode run 'run ls -al and tell me what you see'
 opencode run 'your test prompt'
 
 # Use a cheap model for faster iterations:
-opencode run -m gitlab/duo-chat-haiku-4-5 'your test prompt'
+opencode run -m gitlab/duo-chat-gpt-5-4-nano 'your test prompt'
 ```
 
 ### 3. Use --print-logs to debug infrastructure
@@ -213,7 +213,7 @@ Without the `--`, the CLI parser treats the prompt as a second file path and err
 
 - **Keep prompts self-contained**: Avoid vague prompts that might cause the agent to ask clarifying questions — that will hang the process (see Known Limitations).
 - **Config changes are instant**: No restart needed. Each `opencode run` invocation reads fresh config.
-- **Cheap models for iteration**: Use `-m gitlab/duo-chat-haiku-4-5` when testing infrastructure (permissions, plugins). Switch to your primary model for testing prompt/tone quality.
+- **Cheap models for iteration**: Use `-m gitlab/duo-chat-gpt-5-4-nano` when testing infrastructure (permissions, plugins). Switch to your primary model for testing prompt/tone quality.
 - **Stderr for logs, stdout for output**: `--print-logs` writes to stderr, so you can `2>debug.log` and still see the agent's response on stdout.
 - **Stdin piping**: `echo "prompt" | opencode run` works — useful for multi-line prompts or scripted input.
 

.config/opencode/agents/dotfiles.md

@@ -1,7 +1,7 @@
 ---
 description: "Use when the user mentions their dotfiles, dotfiles repo, dotfiles config, yadm, or references commits/changes in their personal configuration. Explores yadm-managed dotfiles in the home directory: find config files under ~/, search dotfile contents, look up agent docs, and inspect yadm git history (log, show, diff)."
 mode: subagent
-model: gitlab/duo-chat-haiku-4-5
+model: gitlab/duo-chat-gpt-5-4-nano
 hidden: true
 temperature: 0
 steps: 10

.config/opencode/commands/push.md

@@ -1,6 +1,6 @@
 ---
 description: Commit changes in logical commits and push
-model: gitlab/duo-chat-haiku-4-5
+model: gitlab/duo-chat-gpt-5-4-nano
 ---
 
 Commit all changes in discrete, logical commits and push to the remote.

.config/opencode/opencode.json##class.Work

@@ -27,7 +27,7 @@
     }
   },
   "enabled_providers": ["gitlab", "anthropic", "google"],
-  "small_model": "gitlab/duo-chat-haiku-4-5",
+  "small_model": "gitlab/duo-chat-gpt-5-4-nano",
   "provider": {
     "gitlab": {
       "models": {
@@ -66,7 +66,7 @@
             "input": 15, "output": 75, "cache_read": 1.50, "cache_write": 18.75
           }
         },
-        "gitlab/duo-workflow-haiku-4-5": {
+        "duo-workflow-haiku-4-5": {
           "cost": { "input": 1, "output": 5, "cache_read": 0.10, "cache_write": 1.25 }
         },
         "duo-chat-haiku-4-5": {

.config/opencode/opencode.json##default

@@ -10,7 +10,7 @@
     "opencode-wakatime@latest"
   ],
   "default_agent": "plan",
-  "small_model": "gitlab/duo-workflow-haiku-4-5",
+  "small_model": "gitlab/duo-chat-gpt-5-4-nano",
   "provider": {
     "anthropic": {
       "options": {