feat(vision): gate recognition on a configured provider

Expose ImageRecognitionRuntime.available() and gate the image-task pre-step
injection on it, so the capability is dormant until an endpoint is
configured instead of guiding the model to a call that can only fail.

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Pine
2026-08-14 12:08:30 +08:00
parent 2b81e38263
commit 08ac7c3da5
4 changed files with 46 additions and 0 deletions
@@ -102,6 +102,26 @@ export class ImageRecognitionRuntime extends Service {
return () => void dispose() return () => void dispose()
} }
/**
* Whether a provider can currently serve recognition. True when a configured
* provider is registered and usable, or exactly one usable provider is
* registered without a pin. Cheap and side-effect-free (no network): it reads
* only the registry and each provider's local `available()` check, so a
* consumer can gate its surface on configuration without attempting a call.
* @returns whether recognition is currently configured and usable.
*/
available(): boolean {
try {
resolveProvider({
providers: this.providers,
...this.providerId !== undefined ? { configuredId: this.providerId } : {},
})
return true
} catch {
return false
}
}
/** /**
* Run one recognition through the selected provider. Resolves the provider at * Run one recognition through the selected provider. Resolves the provider at
* call time with the selection rules above; throws {@link ImageRecognitionError} * call time with the selection rules above; throws {@link ImageRecognitionError}
@@ -98,3 +98,22 @@ describe('ImageRecognitionRuntime selection', () => {
.rejects.toThrow(expect.objectContaining({ code: 'IMAGE_RECOGNITION_PROVIDER_AMBIGUOUS' })) .rejects.toThrow(expect.objectContaining({ code: 'IMAGE_RECOGNITION_PROVIDER_AMBIGUOUS' }))
}) })
}) })
describe('ImageRecognitionRuntime.available', () => {
it('is false with no registered provider', async () => {
const { runtime } = await mountRuntime()
expect(runtime.available()).toBe(false)
})
it('is false when the configured provider is unavailable (not configured)', async () => {
const { runtime } = await mountRuntime({ provider: 'http' })
runtime.registerProvider(makeProvider('http', unavailable, () => Promise.resolve(recognizeResult('a'))))
expect(runtime.available()).toBe(false)
})
it('is true when a configured provider is available', async () => {
const { runtime } = await mountRuntime({ provider: 'http' })
runtime.registerProvider(makeProvider('http', available, () => Promise.resolve(recognizeResult('a'))))
expect(runtime.available()).toBe(true)
})
})
@@ -96,5 +96,9 @@ Prefix-stable while the guidance text is unchanged.
## Known Limitations and Deferred Work ## Known Limitations and Deferred Work
- **Dormant until configured:** the image-task injection is gated on a usable
provider, so without a configured endpoint the capability is inert and the
model is not guided to recognition. Direct `recognize_image` calls then throw
a clear `ImageRecognitionError`.
- Path/URL detection is heuristic and can be disabled; misdetections are possible. - Path/URL detection is heuristic and can be disabled; misdetections are possible.
- Requires a configured, available provider — see `dsh-image-recognition-http`. - Requires a configured, available provider — see `dsh-image-recognition-http`.
@@ -150,6 +150,9 @@ export function apply(ctx: Context, config: Config = {}): void {
const decision = await next() const decision = await next()
if (decision.kind === 'reject') return decision if (decision.kind === 'reject') return decision
if (!hasImageSignal(decision.messages, detectBlocks, detectPaths)) return decision if (!hasImageSignal(decision.messages, detectBlocks, detectPaths)) return decision
// Dormant until configured: without a usable provider the recognition
// surface is inert, so the model is not guided to a call that can only fail.
if (!ctx.imageRecognition.available()) return decision
signal.throwIfAborted() signal.throwIfAborted()
const lookup = { cwd: agent.session.header.cwd, signal, scope: agent } const lookup = { cwd: agent.session.header.cwd, signal, scope: agent }
const skill = await ctx.skills.get(SKILL_NAME, lookup) const skill = await ctx.skills.get(SKILL_NAME, lookup)