Extract hardcoded GPT apply_patch permission logic into a reusable module
to ensure consistent behavior across all agents. This prevents GPT models
from using the unreliable apply_patch tool while allowing other models.
- Add gpt-apply-patch-guard.ts with GPT_APPLY_PATCH_GUIDANCE and getGptApplyPatchPermission
- Update Hephaestus agent to use centralized permission logic
- Update Sisyphus-Junior agent to use centralized permission logic
- Update all GPT prompt builders to reference shared guidance constant
🤖 Generated with assistance of OhMyOpenCode
The doctor's 'Model override uses unavailable provider' check only
looked at providers from ~/.cache/opencode/models.json (built-in
providers from models.dev). Custom OpenAI-compatible providers defined
in the user's opencode.json (under the 'provider' key) were not
included, causing false-positive warnings.
Now loadAvailableModelsFromCache() also reads provider names from
~/.config/opencode/opencode.json and ~/.config/opencode/opencode.jsonc,
merging them with the cache providers. This eliminates the false
positive while preserving real warnings for truly unknown providers.
7 new tests cover: cache-only, custom-only, merged, deduplicated,
JSONC variant, and malformed config resilience.
Fixes#3199
YeonGyu-Kim