Plumb ctx through classifier and Ollama HTTP path

OllamaClient.Generate/GenerateText/call now take ctx and build the HTTP
request via NewRequestWithContext, so an in-flight LLM call is aborted
when the parent context is cancelled (Ctrl-C). Classifier.Classify and
its tier helpers take ctx too. ProcessFunc gets a ctx parameter so the
poller can forward its cancellable context down to classification.

Explainer.summarize manages its own 60s context since reaction-driven
flow has no parent ctx to inherit.

internal/classifier/classifier.go

@@ -1,6 +1,7 @@
 package classifier
 
 import (
+	"context"
 	"encoding/json"
 	"fmt"
 	"log/slog"
@@ -38,7 +39,8 @@ func NewClassifier(ollama *OllamaClient, matrixClient *matrix.Client) *Classifie
 }
 
 // Classify runs the appropriate classification pipeline for a feed item.
-func (c *Classifier) Classify(item *ingestion.FeedItem) (*ClassifyResult, error) {
+// ctx is checked before any LLM call so shutdown aborts in-flight work.
+func (c *Classifier) Classify(ctx context.Context, item *ingestion.FeedItem) (*ClassifyResult, error) {
 	// Direct routing — no LLM call needed
 	if item.DirectRoute != nil {
 		slog.Debug("direct route", "guid", item.GUID, "channel", *item.DirectRoute)
@@ -60,16 +62,16 @@ func (c *Classifier) Classify(item *ingestion.FeedItem) (*ClassifyResult, error)
 	}
 
 	if item.Tier == 2 {
-		return c.classifyTier2(item, recent)
+		return c.classifyTier2(ctx, item, recent)
 	}
 
-	return c.classifyTier1(item, recent)
+	return c.classifyTier1(ctx, item, recent)
 }
 
-func (c *Classifier) classifyTier1(item *ingestion.FeedItem, recent []storage.RecentHeadline) (*ClassifyResult, error) {
+func (c *Classifier) classifyTier1(ctx context.Context, item *ingestion.FeedItem, recent []storage.RecentHeadline) (*ClassifyResult, error) {
 	prompt := BuildTier1Prompt(item, recent)
 
-	raw, err := c.ollama.Generate(prompt)
+	raw, err := c.ollama.Generate(ctx, prompt)
 	if err != nil {
 		c.handleOllamaFailure(err)
 		return nil, fmt.Errorf("ollama generate: %w", err)
@@ -94,7 +96,7 @@ func (c *Classifier) classifyTier1(item *ingestion.FeedItem, recent []storage.Re
 	}, nil
 }
 
-func (c *Classifier) classifyTier2(item *ingestion.FeedItem, recent []storage.RecentHeadline) (*ClassifyResult, error) {
+func (c *Classifier) classifyTier2(ctx context.Context, item *ingestion.FeedItem, recent []storage.RecentHeadline) (*ClassifyResult, error) {
 	// Run keyword gating first
 	gatingResult, reason := RunGating(item.Headline, item.Lede)
 
@@ -115,7 +117,7 @@ func (c *Classifier) classifyTier2(item *ingestion.FeedItem, recent []storage.Re
 	// Stage 3 (ambiguous) or Stage 1 (definite post needing routing) — send to LLM
 	prompt := BuildTier2Prompt(item, recent)
 
-	raw, err := c.ollama.Generate(prompt)
+	raw, err := c.ollama.Generate(ctx, prompt)
 	if err != nil {
 		c.handleOllamaFailure(err)
 		return nil, fmt.Errorf("ollama generate: %w", err)

internal/classifier/ollama.go

@@ -2,6 +2,7 @@ package classifier
 
 import (
 	"bytes"
+	"context"
 	"encoding/json"
 	"fmt"
 	"io"
@@ -49,9 +50,11 @@ func NewOllamaClient(cfg config.OllamaConfig) *OllamaClient {
 	}
 }
 
-// Generate sends a prompt to Ollama via the chat API and returns the raw response text.
-func (o *OllamaClient) Generate(prompt string) (string, error) {
-	return o.call(
+// Generate sends a prompt to Ollama via the chat API and returns the raw
+// response text. Honors ctx cancellation — callers can abort an in-flight
+// call instead of waiting for the configured timeout.
+func (o *OllamaClient) Generate(ctx context.Context, prompt string) (string, error) {
+	return o.call(ctx,
 		"You are a JSON-only news classification API. Respond with valid JSON only. No preamble, no explanation, no markdown.",
 		prompt,
 		"json",
@@ -60,11 +63,11 @@ func (o *OllamaClient) Generate(prompt string) (string, error) {
 
 // GenerateText sends a system + user prompt to Ollama and returns the raw
 // text response. Unlike Generate, no JSON mode is forced.
-func (o *OllamaClient) GenerateText(systemPrompt, userPrompt string) (string, error) {
-	return o.call(systemPrompt, userPrompt, "")
+func (o *OllamaClient) GenerateText(ctx context.Context, systemPrompt, userPrompt string) (string, error) {
+	return o.call(ctx, systemPrompt, userPrompt, "")
 }
 
-func (o *OllamaClient) call(systemPrompt, userPrompt, format string) (string, error) {
+func (o *OllamaClient) call(ctx context.Context, systemPrompt, userPrompt, format string) (string, error) {
 	reqBody := chatRequest{
 		Model: o.model,
 		Messages: []chatMessage{
@@ -83,7 +86,12 @@ func (o *OllamaClient) call(systemPrompt, userPrompt, format string) (string, er
 	url := o.baseURL + "/api/chat"
 	slog.Debug("ollama: calling", "url", url, "model", o.model)
 
-	resp, err := o.httpClient.Post(url, "application/json", bytes.NewReader(body))
+	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
+	if err != nil {
+		return "", fmt.Errorf("ollama request build: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/json")
+	resp, err := o.httpClient.Do(req)
 	if err != nil {
 		return "", fmt.Errorf("ollama request: %w", err)
 	}

internal/explainer/explainer.go

@@ -4,6 +4,7 @@
 package explainer
 
 import (
+	"context"
 	"fmt"
 	"html"
 	"log/slog"
@@ -125,8 +126,11 @@ Rules:
 - Stick strictly to facts in the article. Do not speculate.`
 
 func (e *Explainer) summarize(headline, body string) (string, error) {
+	// Reaction-driven flow has no parent ctx; bound the LLM call ourselves.
+	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	defer cancel()
 	prompt := fmt.Sprintf("Headline: %s\n\nArticle body:\n%s", headline, body)
-	out, err := e.ollama.GenerateText(summarySystem, prompt)
+	out, err := e.ollama.GenerateText(ctx, summarySystem, prompt)
 	if err != nil {
 		return "", err
 	}

internal/ingestion/poller.go

@@ -12,7 +12,9 @@ import (
 )
 
 // ProcessFunc is called for each new feed item that needs classification.
-type ProcessFunc func(item *FeedItem)
+// ctx is the poller's parent context; callbacks should pass it into any
+// downstream LLM/HTTP work so shutdown aborts in-flight calls.
+type ProcessFunc func(ctx context.Context, item *FeedItem)
 
 // Poller manages per-source RSS polling goroutines.
 type Poller struct {
@@ -182,7 +184,7 @@ func (p *Poller) pollOnceWithErr(ctx context.Context, src config.SourceConfig) e
 			continue
 		}
 
-		p.process(&items[i])
+		p.process(ctx, &items[i])
 		newCount++
 	}
 
@@ -212,7 +214,7 @@ func (p *Poller) pollOnceWithErr(ctx context.Context, src config.SourceConfig) e
 			continue
 		}
 
-		p.process(&FeedItem{
+		p.process(ctx, &FeedItem{
 			GUID:        s.GUID,
 			Headline:    s.Headline,
 			Lede:        s.Lede,

main.go

@@ -95,8 +95,8 @@ func main() {
 	slog.Info("post queue started")
 
 	// Build the pipeline callback: classify → enqueue
-	processItem := func(item *ingestion.FeedItem) {
-		result, err := cls.Classify(item)
+	processItem := func(ctx context.Context, item *ingestion.FeedItem) {
+		result, err := cls.Classify(ctx, item)
 		if err != nil {
 			slog.Error("classification failed, will retry next cycle",
 				"guid", item.GUID,