fix: cancel blocked tasks + auto-complete completion reports

Two fixes for BLOCKED task issues:

1. Allow BLOCKED → CANCELLED state transition so users can cancel tasks
   stuck waiting for input. Adds Cancel button to BLOCKED task cards in
   the UI alongside the question/answer controls.

2. Detect when agents write completion reports to $CLAUDOMATOR_QUESTION_FILE
   instead of real questions. If the question JSON has no options and no "?"
   in the text, treat it as a summary (stored on the execution) and fall
   through to normal completion + sandbox teardown rather than blocking.
   Also tightened the preamble to make the distinction explicit.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

1 files changed, 38 insertions, 4 deletions

diff --git a/internal/executor/claude.go b/internal/executor/claude.go
index a58f1ad..a3d304d 100644
--- a/internal/executor/claude.go
+++ b/internal/executor/claude.go
@@ -144,10 +144,18 @@ func (r *ClaudeRunner) Run(ctx context.Context, t *task.Task, e *storage.Executi
 	data, readErr := os.ReadFile(questionFile)
 	if readErr == nil {
 		os.Remove(questionFile) // consumed
-		// Preserve sandbox on BLOCKED — agent may have partial work and its
-		// Claude session files are stored under the sandbox's project slug.
-		// The resume execution must run in the same directory.
-		return &BlockedError{QuestionJSON: strings.TrimSpace(string(data)), SessionID: e.SessionID, SandboxDir: sandboxDir}
+		questionJSON := strings.TrimSpace(string(data))
+		// If the agent wrote a completion report instead of a real question,
+		// extract the text as the summary and fall through to normal completion.
+		if isCompletionReport(questionJSON) {
+			r.Logger.Info("treating question file as completion report", "taskID", e.TaskID)
+			e.Summary = extractQuestionText(questionJSON)
+		} else {
+			// Preserve sandbox on BLOCKED — agent may have partial work and its
+			// Claude session files are stored under the sandbox's project slug.
+			// The resume execution must run in the same directory.
+			return &BlockedError{QuestionJSON: questionJSON, SessionID: e.SessionID, SandboxDir: sandboxDir}
+		}
 	}
 
 	// Read agent summary if written.
@@ -166,6 +174,32 @@ func (r *ClaudeRunner) Run(ctx context.Context, t *task.Task, e *storage.Executi
 	return nil
 }
 
+// isCompletionReport returns true when a question-file JSON looks like a
+// completion report rather than a real user question. Heuristic: no options
+// (or empty options) and no "?" anywhere in the text.
+func isCompletionReport(questionJSON string) bool {
+	var q struct {
+		Text    string   `json:"text"`
+		Options []string `json:"options"`
+	}
+	if err := json.Unmarshal([]byte(questionJSON), &q); err != nil {
+		return false
+	}
+	return len(q.Options) == 0 && !strings.Contains(q.Text, "?")
+}
+
+// extractQuestionText returns the "text" field from a question-file JSON, or
+// the raw string if parsing fails.
+func extractQuestionText(questionJSON string) string {
+	var q struct {
+		Text string `json:"text"`
+	}
+	if err := json.Unmarshal([]byte(questionJSON), &q); err != nil {
+		return questionJSON
+	}
+	return strings.TrimSpace(q.Text)
+}
+
 // sandboxCloneSource returns the URL to clone the sandbox from. It prefers a
 // remote named "local" (a local bare repo that accepts pushes cleanly), then
 // falls back to "origin", then to the working copy path itself.