Merge pull request #36 from xmindlab-heicode/fix/agent-token-cryptosecret-and-purge-guard

fix(agent,secret): enforce CRYPTO_SECRET for agent deploy (#31) + guard unscoped vault purge (#33)

heicode/controller/agent_template_handlers.go

@@ -3,6 +3,7 @@ package controller
 import (
 	"context"
 	"crypto/subtle"
+	"os"
 	"strconv"
 	"strings"
 	"sync"
@@ -38,6 +39,23 @@ func sealAgentToken(token string) string {
 	return sealed
 }
 
+// agentTokenSealKeyConfigured reports whether a stable, explicitly-configured
+// CRYPTO_SECRET is present so the per-agent access_token can be sealed with a
+// key that survives a container restart (#31).
+//
+// Subtle but important: common.CryptoSecret is NEVER literally "" — it defaults
+// to uuid.New() (constants.go) and, when CRYPTO_SECRET is unset, falls back to
+// SessionSecret (init.go). So the "CryptoSecret unset?" plaintext fallback in
+// sealAgentToken is effectively unreachable; the real production hazard is a
+// key that is not stable across restarts. If CRYPTO_SECRET is not explicitly
+// set, the seal key may be an ephemeral random UUID, so every previously sealed
+// agent token becomes undecryptable after the next restart. We therefore gate
+// agent deployment on CRYPTO_SECRET being explicitly configured. Pure helper so
+// the policy is unit-testable without mutating process env globals.
+func agentTokenSealKeyConfigured(cryptoSecretEnv string) bool {
+	return strings.TrimSpace(cryptoSecretEnv) != ""
+}
+
 // unsealAgentToken reverses sealAgentToken. If the stored value is not a sealed
 // blob (legacy plaintext / CryptoSecret unset), it is returned as-is.
 func unsealAgentToken(stored string) string {
@@ -117,6 +135,20 @@ func HeicodeDeployAgent(c *gin.Context) {
 		agentError(c, "POLICY_REJECTED", "authentication required")
 		return
 	}
+
+	// #31: refuse to deploy when the per-agent access_token cannot be sealed with
+	// a stable key. Without an explicit CRYPTO_SECRET the seal key falls back to
+	// an ephemeral random value, so the sealed token becomes undecryptable after
+	// the next restart — the agent would silently lose its credential. A dev-only
+	// override keeps local runs (no CRYPTO_SECRET) working; it must never be set
+	// in production.
+	if !agentTokenSealKeyConfigured(os.Getenv("CRYPTO_SECRET")) &&
+		!common.GetEnvOrDefaultBool("HEICODE_ALLOW_PLAINTEXT_AGENT_TOKEN_IN_DEV", false) {
+		common.SysLog("agent deploy rejected: CRYPTO_SECRET not set (per-agent access_token cannot be sealed with a restart-stable key)")
+		agentError(c, "POLICY_REJECTED", "agent deployment is disabled until CRYPTO_SECRET is configured, so the per-agent access token can be sealed with a key that survives restarts (dev-only override: HEICODE_ALLOW_PLAINTEXT_AGENT_TOKEN_IN_DEV=true)")
+		return
+	}
+
 	var req struct {
 		TemplateID string `json:"template_id"`
 		BindingIDs []int  `json:"binding_ids"`

heicode/controller/agent_template_test.go

@@ -264,3 +264,13 @@ func TestTemplateAgentResponse(t *testing.T) {
 	require.Equal(t, []int{1, 2}, resp["binding_ids"])
 	require.Equal(t, "running", resp["status"])
 }
+
+// #31: agent deployment requires an explicitly configured CRYPTO_SECRET so the
+// per-agent access_token is sealed with a key that survives a container restart.
+// common.CryptoSecret is never literally "" (it defaults to uuid/SessionSecret),
+// so the gate must key off the CRYPTO_SECRET env value, not the runtime var.
+func TestAgentTokenSealKeyConfigured(t *testing.T) {
+	require.False(t, agentTokenSealKeyConfigured(""))
+	require.False(t, agentTokenSealKeyConfigured("   "))
+	require.True(t, agentTokenSealKeyConfigured("a-real-secret"))
+}

heicode/controller/secret_lifecycle.go

@@ -214,6 +214,19 @@ func StartSecretPurgeTask() {
 		if strings.TrimSpace(os.Getenv("AZURE_KEY_VAULT_URL")) == "" {
 			return // no vault configured — nothing to purge
 		}
+		// #33: a destructive purge that scans the WHOLE vault is only safe when
+		// HM owns the vault exclusively. If no name prefix scopes purging to
+		// HM-managed secrets, require an explicit opt-in (HEICODE_SECRET_PURGE_
+		// VAULT_EXCLUSIVE=true) so HM never permanently purges another tenant's
+		// soft-deleted secrets that happen to live in a shared vault.
+		purgePrefix := strings.TrimSpace(common.GetEnvOrDefaultString("HEICODE_SECRET_PURGE_NAME_PREFIX", ""))
+		vaultExclusive := common.GetEnvOrDefaultBool("HEICODE_SECRET_PURGE_VAULT_EXCLUSIVE", false)
+		allowed, scopeDesc := secretPurgeScopeAllowed(purgePrefix, vaultExclusive)
+		if !allowed {
+			common.SysLog("secret purge task NOT started: " + scopeDesc)
+			return
+		}
+		common.SysLog("secret purge scope: " + scopeDesc)
 		intervalHours := common.GetEnvOrDefault("HEICODE_SECRET_PURGE_INTERVAL_HOURS", 24)
 		if intervalHours < 1 {
 			intervalHours = 24
@@ -231,6 +244,23 @@ func StartSecretPurgeTask() {
 	})
 }
 
+// secretPurgeScopeAllowed decides whether the destructive vault purge may run,
+// given the configured name prefix and the vault-exclusive opt-in (#33). A purge
+// that scans the WHOLE vault (empty prefix) is only safe when HM owns the vault
+// exclusively, so it must be explicitly opted in. Returns the decision plus a
+// human-readable scope/refusal description for the startup log. Pure helper for
+// unit testing without touching process env.
+func secretPurgeScopeAllowed(namePrefix string, vaultExclusive bool) (bool, string) {
+	prefix := strings.TrimSpace(namePrefix)
+	if prefix == "" {
+		if !vaultExclusive {
+			return false, "HEICODE_SECRET_PURGE_NAME_PREFIX is empty and HEICODE_SECRET_PURGE_VAULT_EXCLUSIVE!=true — refusing to purge an entire (possibly shared) vault. Set a name prefix to scope to HM-managed secrets, or set HEICODE_SECRET_PURGE_VAULT_EXCLUSIVE=true only if this vault is exclusive to HM."
+		}
+		return true, "ENTIRE vault (HEICODE_SECRET_PURGE_VAULT_EXCLUSIVE=true, no name prefix)"
+	}
+	return true, "secrets with name prefix '" + prefix + "'"
+}
+
 func secretPurgeRetentionDays() int {
 	d := common.GetEnvOrDefault("HEICODE_SECRET_PURGE_RETENTION_DAYS", 30)
 	if d < 1 {

heicode/controller/secret_lifecycle_test.go

@@ -7,6 +7,23 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+// #33: an unscoped (empty-prefix) purge must be refused unless the operator
+// explicitly declares the vault exclusive to HM, so HM never permanently purges
+// another tenant's soft-deleted secrets in a shared vault.
+func TestSecretPurgeScopeAllowed(t *testing.T) {
+	allowed, desc := secretPurgeScopeAllowed("", false)
+	require.False(t, allowed)
+	require.Contains(t, desc, "refusing to purge an entire")
+
+	allowed, desc = secretPurgeScopeAllowed("  ", true)
+	require.True(t, allowed)
+	require.Contains(t, desc, "ENTIRE vault")
+
+	allowed, desc = secretPurgeScopeAllowed("heicode-", false)
+	require.True(t, allowed)
+	require.Contains(t, desc, "heicode-")
+}
+
 func TestParseDeletedSecretsPage(t *testing.T) {
 	body := []byte(`{
 		"value": [