feat(#6788): add dummy-playback runtime for replaying canned results

Add DummyPlaybackRuntime that replays pre-recorded agent results from
an ordered playlist without LLM inference, enabling fast deterministic
e2e tests of the full dispatch pipeline. Each invocation reads the
current playlist entry's result.json, writes it to the sandbox output,
copies companion files, creates feature branches or commits fixes, and
advances the playlist index.

Includes security hardening: shell-quoted entry names in git commands,
path-traversal validation on playlist entries, CLI allowlist (gh/glab)
for comment tracking, and guarded SHA truncation.

Closes #6788

Signed-off-by: Greg Allen <gallen@redhat.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Greg Allen <gallen@redhat.com>

Author: ggallen

internal/cli/run.go

@@ -1972,6 +1972,7 @@ func runAgent(ctx context.Context, agentName, fullsendDir, outputBase, targetRep
 			Timeout:           timeout,
 			OutputPath:        filepath.Join(iterDir, "output.jsonl"),
 			Prompt:            agentPrompt,
+			Forge:             forgePlatform,
 			OnEvent:           contentEventHandler(agentruntime.NewEventRenderer(printer).Handle, collector),
 		}, printer, agentStart, &metrics)
 		close(heartbeatDone)

internal/config/config.go

@@ -299,7 +299,7 @@ func ValidProviders() []string {
 // ValidRuntimes returns the set of recognized agent runtimes. "pi" is
 // opt-in per org/repo (#6464); "dummy" is for behaviour test orgs only.
 func ValidRuntimes() []string {
-	return []string{"claude", "pi", "dummy"}
+	return []string{"claude", "pi", "dummy", "dummy-playback"}
 }
 
 // validModelRef matches a provider-qualified model reference: one or more