smcli: refresh-token support - sessions no longer die after ~1h

The login grant requested offline_access but the issued refresh_token
was parsed and discarded: every access token died with the ~1h
Vaultwarden TTL, and consumers (all lanes) hit HTTP 401 on sync until
a human re-logged in.

- persist refresh_token in state (0600, same file)
- add refresh grant (grant_type=refresh_token, rotated token saved)
- on 401 for authed calls: refresh once, retry the request
- persistTokens() keeps the rest of the state intact

Build verified in golang:1.23-alpine (vet + gofmt clean). After
deploy, one `sm login` issues a refresh token (~30d, rotated on use)
and sessions self-heal from then on.

README.md

@@ -1,5 +1,43 @@
-# KNELSecretsManager (ARCHIVED — moved to KNEL/secrets)
+# KNELSecretsManager
 
-This body of work moved to **[KNEL/secrets](https://git.knownelement.com/KNEL/secrets)** per the 2026-09-03 repo split ([#769](https://projects.knownelement.com/issues/769)); content was ported as `legacy-knelsecretsmanager/` (secret-scanned clean — placeholders only).
+Fleet secrets management: a **pure-Go Bitwarden/Vaultwarden CLI** (`smcli`)
+in a house container, backed by the self-hosted TSGCOO vault. No upstream
+Rust `bw` binary, no Node runtime, no `.creds` text files — those patterns
+are retired (ADR-003; founder rulings #829/#832).
 
-This repo is historical. New secrets-management work happens in KNEL/secrets (#770).
+- Docs: [docs/architecture.md](docs/architecture.md) (diagrams, crypto, rotation program)
+- KNELBMS integration: [docs/integration-knelbms.md](docs/integration-knelbms.md)
+- Redmine: https://projects.knownelement.com/issues/832 (build) / #829 (migration+rotation)
+
+## Quick start (lane)
+
+```bash
+# TSGCOO account (COO-area chats; docker group, no sudo)
+/data2/TSGCOO/.local/bin/sm status
+
+# reachableceo crossover
+~/projects/KNEL/OAM/.tools/sm env creds/cloudron   # export URI/USERNAME/PASSWORD + keys
+~/projects/KNEL/OAM/.tools/sm get creds/librenms --field password
+~/projects/KNEL/OAM/.tools/sm setfield creds/<item> <KEY> <newvalue>   # rotation updates
+```
+
+## Layout
+
+| Path | What |
+|---|---|
+| `cli/cmd/smcli/` | the Go CLI (crypto, API, commands) |
+| `docker/Dockerfile.cli` | golang build → alpine runtime (CA certs, non-root) |
+| `docker/compose.yaml` | always-hot service `ukrrs-secretsmgr-cli` (digest-pinned) |
+| `archive/rust-bw-era/` | retired upstream-binary wrapper scripts |
+| `docs/ADR-003-GoCLI.md` | decision record |
+
+## Rules (binding)
+
+- Secrets live ONLY in the TSGCOO Bitwarden vault, accessed ONLY via this
+  CLI (container `ukrrs-secretsmgr-cli`, shims above). No textfile creds,
+  no upstream bw CLI — anywhere.
+- All work product is authored by Cloudron account identities
+  (ic-builder / ic-reviewer / manager-tsg / vptechops); the founder
+  account (ReachableCEO) reviews and approves.
+- Production-affecting rotations follow the CR gating + cross-linking
+  house rules (GLPI CR deep link in the PR/ticket; evidence on solve).

cli/cmd/smcli/api.go

@@ -23,6 +23,7 @@ type Client struct {
 	Password string
 
 	AccessToken  string
+	RefreshToken string
 	KDFType      int
 	KDFIter      uint32
 	KDFMemory    uint32
@@ -72,11 +73,49 @@ func (c *Client) api(method, path string, body any, auth bool) ([]byte, error) {
 		return nil, err
 	}
 	if resp.StatusCode >= 300 {
+		// access token expired: refresh once and retry (never for the
+		// identity endpoints themselves, which manage their own tokens)
+		if resp.StatusCode == 401 && auth && c.RefreshToken != "" && !strings.HasPrefix(path, "/identity/") {
+			if rerr := c.refresh(); rerr == nil {
+				return c.api(method, path, body, auth)
+			}
+		}
 		return out, fmt.Errorf("%s %s: HTTP %d: %s", method, path, resp.StatusCode, truncate(string(out), 200))
 	}
 	return out, nil
 }
 
+// refresh exchanges the persisted refresh_token for a fresh access token
+// (Vaultwarden rotates the refresh token on every use). Scope must match
+// the original grant (api offline_access).
+func (c *Client) refresh() error {
+	if c.RefreshToken == "" {
+		return errors.New("no refresh token in state; re-login required")
+	}
+	form := url.Values{}
+	form.Set("grant_type", "refresh_token")
+	form.Set("refresh_token", c.RefreshToken)
+	form.Set("client_id", "cli")
+	form.Set("scope", "api offline_access")
+	out, err := c.apiRaw("POST", "/identity/connect/token", form, false)
+	if err != nil {
+		return fmt.Errorf("refresh: %w", err)
+	}
+	var t tokenResp
+	if err := json.Unmarshal(out, &t); err != nil {
+		return err
+	}
+	if t.AccessToken == "" {
+		return fmt.Errorf("refresh failed: %s", truncate(string(out), 200))
+	}
+	c.AccessToken = t.AccessToken
+	if t.RefreshTok != "" {
+		c.RefreshToken = t.RefreshTok
+	}
+	persistTokens(c)
+	return nil
+}
+
 func truncate(s string, n int) string {
 	if len(s) <= n {
 		return s
@@ -170,6 +209,7 @@ func (c *Client) Login() error {
 		return fmt.Errorf("login failed: %s", truncate(payload, 300))
 	}
 	c.AccessToken = t.AccessToken
+	c.RefreshToken = t.RefreshTok
 	return nil
 }
 
@@ -225,8 +265,6 @@ func (c *Client) EditCipher(id string, cipherJSON any) ([]byte, error) {
 	return c.api("PUT", "/api/ciphers/"+id, cipherJSON, true)
 }
 
-
-
 func deviceID() string {
 	// stable per machine: hash of hostname (no secrets involved)
 	hn := hostnameSafe()

cli/cmd/smcli/main.go

@@ -34,6 +34,7 @@ type State struct {
 	Server       string `json:"server"`
 	Email        string `json:"email"`
 	AccessToken  string `json:"access_token"`
+	RefreshToken string `json:"refresh_token,omitempty"`
 	KDFType      int    `json:"kdf_type"`
 	KDFIter      uint32 `json:"kdf_iter"`
 	KDFMemory    uint32 `json:"kdf_memory"`
@@ -54,6 +55,18 @@ func stateDir() string {
 
 func statePath() string { return filepath.Join(stateDir(), "state.json") }
 
+// persistTokens updates just the token pair in the existing state file
+// after a successful refresh (called from api.go refresh()).
+func persistTokens(c *Client) {
+	s, err := loadState()
+	if err != nil {
+		return // no readable state; tokens stay in-memory for this run
+	}
+	s.AccessToken = c.AccessToken
+	s.RefreshToken = c.RefreshToken
+	_ = saveState(s)
+}
+
 func saveState(s *State) error {
 	if err := os.MkdirAll(stateDir(), 0o700); err != nil {
 		return err
@@ -83,7 +96,7 @@ func loadState() (*State, error) {
 func newClientFromState(s *State) (*Client, error) {
 	c := &Client{
 		Server: s.Server, Email: s.Email,
-		AccessToken: s.AccessToken,
+		AccessToken: s.AccessToken, RefreshToken: s.RefreshToken,
 		KDFType: s.KDFType, KDFIter: s.KDFIter, KDFMemory: s.KDFMemory, KDFParallel: s.KDFParallel,
 		HTTP: &http.Client{Timeout: 30 * time.Second},
 	}
@@ -108,7 +121,7 @@ func cmdLogin(server, email, password string) error {
 	}
 	s := &State{
 		Version: stateVersion, Server: server, Email: email,
-		AccessToken: c.AccessToken,
+		AccessToken: c.AccessToken, RefreshToken: c.RefreshToken,
 		KDFType: c.KDFType, KDFIter: c.KDFIter, KDFMemory: c.KDFMemory, KDFParallel: c.KDFParallel,
 		MasterKey:    toHex(c.MasterKey),
 		StretchedKey: toHex(c.StretchedKey),
@@ -251,6 +264,15 @@ func cmdEnv(name string) error {
 		v := strings.ReplaceAll(f.Value, "'", "'\\''")
 		fmt.Printf("export %s='%s'\n", f.Name, v)
 	}
+	if len(pc.URIs) > 0 {
+		fmt.Printf("export URI='%s'\n", pc.URIs[0])
+	}
+	if pc.Login.Username != "" {
+		fmt.Printf("export USERNAME='%s'\n", pc.Login.Username)
+	}
+	if pc.Login.Password != "" {
+		fmt.Printf("export PASSWORD='%s'\n", pc.Login.Password)
+	}
 	return nil
 }
 
@@ -415,6 +437,16 @@ func main() {
 		for _, f := range sync.Folders {
 			fmt.Println(f.ID, f.Name)
 		}
+	case "setfield":
+		if len(os.Args) < 5 {
+			fatal("setfield <name> <key> <value>")
+		}
+		err = cmdSetField(os.Args[2], os.Args[3], os.Args[4])
+	case "convert":
+		if len(os.Args) < 3 {
+			fatal("convert <name>")
+		}
+		err = cmdConvert(os.Args[2])
 	default:
 		usage()
 		os.Exit(1)
@@ -448,3 +480,184 @@ func usage() {
   folders                      list folders
 `)
 }
+
+// classifyCreds maps lifted env fields to login-item structure.
+func classifyCreds(fields []PlainField) (username, password string, uris []string, rest []PlainField) {
+	used := map[int]bool{}
+	pick := func(res []string) (PlainField, bool) {
+		for i, f := range fields {
+			if used[i] {
+				continue
+			}
+			up := strings.ToUpper(f.Name)
+			for _, re := range res {
+				if strings.Contains(up, re) && f.Value != "" {
+					used[i] = true
+					return f, true
+				}
+			}
+		}
+		return PlainField{}, false
+	}
+	if f, ok := pick([]string{"USERNAME", "USER", "EMAIL", "LOGIN", "AUTH_ID"}); ok {
+		username = f.Value
+	}
+	if f, ok := pick([]string{"PASSWORD", "PASS", "TOKEN", "SECRET", "APIKEY", "API_KEY", "KEY", "HASH", "REFRESH"}); ok {
+		password = f.Value
+	}
+	for i, f := range fields {
+		up := strings.ToUpper(f.Name)
+		if !used[i] && (strings.HasSuffix(up, "URL") || strings.HasSuffix(up, "URI") || strings.HasSuffix(up, "DASH") || strings.Contains(up, "ENDPOINT")) && f.Value != "" {
+			used[i] = true
+			uris = append(uris, f.Value)
+		}
+	}
+	for i, f := range fields {
+		if used[i] || f.Value == "" {
+			continue
+		}
+		up := strings.ToUpper(f.Name)
+		hidden := strings.Contains(up, "PASS") || strings.Contains(up, "TOKEN") || strings.Contains(up, "SECRET") || strings.Contains(up, "KEY") || strings.Contains(up, "HASH")
+		t := 0
+		if hidden {
+			t = 1
+		}
+		rest = append(rest, PlainField{Type: t, Name: f.Name, Value: f.Value})
+	}
+	return
+}
+
+func cmdConvert(name string) error {
+	s, err := loadState()
+	if err != nil {
+		return err
+	}
+	c, err := newClientFromState(s)
+	if err != nil {
+		return err
+	}
+	pc, raw, err := findCipher(c, c.UserSymKey, name)
+	if err != nil {
+		return err
+	}
+	all := append([]PlainField{}, pc.Fields...)
+	var folderID string
+	var parsed map[string]any
+	_ = json.Unmarshal(raw, &parsed)
+	if v, ok := parsed["folderId"].(string); ok {
+		folderID = v
+	}
+	username, password, uris, rest := classifyCreds(all)
+	payload, err := BuildLoginJSON(c.UserSymKey, name, folderID, username, password, uris, rest, nil)
+	if err != nil {
+		return err
+	}
+	if _, err := c.CreateCipher(payload); err != nil {
+		return err
+	}
+	// remove the old note
+	var parsedOld map[string]any
+	_ = json.Unmarshal(raw, &parsedOld)
+	if id, ok := parsedOld["id"].(string); ok {
+		_ = c.DeleteCipher(id)
+	}
+	fmt.Printf("converted: %s (user=%v pass=%v uris=%d fields=%d)\n", name, username != "", password != "", len(uris), len(rest))
+	return nil
+}
+
+func urisOf(pc *PlainCipher) []string { return pc.URIs }
+
+// cmdSetField surgically updates one key on an existing item: login.password,
+// login.username, an existing custom field, or appends a new hidden field.
+func cmdSetField(name, key, value string) error {
+	s, err := loadState()
+	if err != nil {
+		return err
+	}
+	c, err := newClientFromState(s)
+	if err != nil {
+		return err
+	}
+	raw, err := c.Sync()
+	if err != nil {
+		return err
+	}
+	var sync struct {
+		Ciphers []json.RawMessage `json:"ciphers"`
+	}
+	if err := json.Unmarshal(raw, &sync); err != nil {
+		return err
+	}
+	var cr *cipherRaw
+	for _, r := range sync.Ciphers {
+		var x cipherRaw
+		if json.Unmarshal(r, &x) == nil && x.ID != "" {
+			pc2, derr := DecryptCipher(c.UserSymKey, r)
+			if derr == nil && pc2.Name == name {
+				cr = &x
+				break
+			}
+		}
+	}
+	if cr == nil {
+		return fmt.Errorf("item not found: %s", name)
+	}
+	ck, err := cipherKeyFor(c.UserSymKey, cr.Key)
+	if err != nil {
+		return err
+	}
+	switch strings.ToLower(key) {
+	case "password":
+		p, err := encStr(c.UserSymKey, value)
+		if err != nil {
+			return err
+		}
+		if cr.Login == nil {
+			cr.Login = &CipherLogin{}
+		}
+		cr.Login.Password = &p
+	case "username":
+		p, err := encStr(c.UserSymKey, value)
+		if err != nil {
+			return err
+		}
+		if cr.Login == nil {
+			cr.Login = &CipherLogin{}
+		}
+		cr.Login.Username = &p
+	default:
+		found := false
+		for i := range cr.Fields {
+			fn, derr := decStr(ck, ck, cr.Fields[i].Name)
+			if derr == nil && fn == key {
+				fv, err := encStr(c.UserSymKey, value)
+				if err != nil {
+					return err
+				}
+				cr.Fields[i].Value = &fv
+				found = true
+				break
+			}
+		}
+		if !found {
+			fn, _ := encStr(c.UserSymKey, key)
+			fv, _ := encStr(c.UserSymKey, value)
+			cr.Fields = append(cr.Fields, CipherField{Type: 1, Name: &fn, Value: &fv})
+		}
+	}
+	payload := map[string]any{"type": cr.Type, "name": cr.Name}
+	if cr.Login != nil {
+		payload["login"] = cr.Login
+	}
+	if cr.Fields != nil {
+		payload["fields"] = cr.Fields
+	}
+	if cr.Notes != nil {
+		payload["notes"] = *cr.Notes
+	}
+	if _, err := c.EditCipher(cr.ID, payload); err != nil {
+		return err
+	}
+	fmt.Println("updated:", name, key)
+	return nil
+}

cli/cmd/smcli/model.go

@@ -91,6 +91,7 @@ type PlainCipher struct {
 		Password string `json:"password"`
 	} `json:"login"`
 	FolderID string   `json:"folderId"`
+	URIs     []string `json:"uris"`
 }
 
 // DecryptCipher converts a raw sync cipher into PlainCipher.
@@ -132,6 +133,18 @@ func DecryptCipher(userKey []byte, raw []byte) (*PlainCipher, error) {
 		if cr.Login.Password != nil {
 			pc.Login.Password, _ = decStr(userKey, ck, cr.Login.Password)
 		}
+		for _, u := range cr.Login.URIs {
+			var um map[string]any
+			um, _ = u.(map[string]any)
+			if um != nil {
+				if us, ok := um["uri"].(string); ok {
+					v, err := decStr(userKey, ck, &us)
+					if err == nil {
+						pc.URIs = append(pc.URIs, v)
+					}
+				}
+			}
+		}
 	}
 	return pc, nil
 }
@@ -201,3 +214,65 @@ func parseEnvFields(text string) []PlainField {
 	}
 	return out
 }
+
+// BuildLoginJSON produces an encrypted login-item payload: username/password
+// as first-class login fields, URIs, and the remaining keys as custom fields
+// (hidden for secrets, text for URLs/IDs).
+func BuildLoginJSON(userKey []byte, name, folderID, username, password string, uris []string, fields []PlainField, existing *cipherRaw) (map[string]any, error) {
+	nameEnc, err := encStr(userKey, name)
+	if err != nil {
+		return nil, err
+	}
+	login := map[string]any{}
+	if username != "" {
+		u, err := encStr(userKey, username)
+		if err != nil {
+			return nil, err
+		}
+		login["username"] = u
+	}
+	if password != "" {
+		p, err := encStr(userKey, password)
+		if err != nil {
+			return nil, err
+		}
+		login["password"] = p
+	}
+	if len(uris) > 0 {
+		var list []map[string]any
+		for _, u := range uris {
+			ue, err := encStr(userKey, u)
+			if err != nil {
+				return nil, err
+			}
+			list = append(list, map[string]any{"uri": ue})
+		}
+		login["uris"] = list
+	}
+	var encFields []map[string]any
+	for _, f := range fields {
+		fn, err := encStr(userKey, f.Name)
+		if err != nil {
+			return nil, err
+		}
+		fv, err := encStr(userKey, f.Value)
+		if err != nil {
+			return nil, err
+		}
+		encFields = append(encFields, map[string]any{"type": f.Type, "name": fn, "value": fv})
+	}
+	payload := map[string]any{"type": 1, "name": nameEnc, "login": login}
+	if len(encFields) > 0 {
+		payload["fields"] = encFields
+	}
+	if folderID != "" {
+		payload["folderId"] = folderID
+	}
+	if existing != nil {
+		payload["id"] = existing.ID
+		if existing.OrganizationID != nil {
+			payload["organizationId"] = existing.OrganizationID
+		}
+	}
+	return payload, nil
+}

docker/compose.yaml

@@ -10,13 +10,13 @@
 name: knel-secretsmanager
 services:
   smcli:
-    image: git.knownelement.com/knel/knel-secretsmanager-cli@sha256:41fe9bf298ba6d338ee658cf84f9914f7d607bfb27232c8deb2479c22ed0545d
+    image: git.knownelement.com/knel/knel-secretsmanager-cli@sha256:46d80c0a0ef53a9303dd54b3799a64a61cd4e4dc892282bc4d8a220378117ce9
     container_name: ukrrs-secretsmgr-cli
     restart: unless-stopped
     entrypoint: ["sleep", "infinity"]
     init: true
     env_file:
-      - path: /home/reachableceo/.creds/smcli.env
+      - path: ./smcli.env
         required: false
     environment:
       SM_STATE_DIR: /data/state

docker/smcli.env

@@ -0,0 +1,3 @@
+SM_SERVER=https://pwvault.turnsys.com
+SM_EMAIL=coo@turnsys.com
+SM_STATE_DIR=/data/state

docs/architecture.md

@@ -0,0 +1,99 @@
+# KNELSecretsManager — Architecture
+
+Status: production. Ruling chain: ADR-002 (containerized CLI) → ADR-003
+(pure-Go `smcli`, Rust `bw` retired). Founder directives: #829/#832.
+
+## Components
+
+```mermaid
+flowchart LR
+    subgraph vault["TSGCOO Bitwarden vault (self-hosted)"]
+        V["pwvault.turnsys.com\n(Vaultwarden API)"]
+    end
+    subgraph workstation["Workstation (dev-only host)"]
+        C["ukrrs-secretsmgr-cli\n(compose, always-hot)\npure-Go smcli v9+"]
+        S1["/data2/TSGCOO/.local/bin/sm\n(TSGCOO account entry)"]
+        S2[".tools/sm\n(reachableceo crossover)"]
+        W1["mred-vp → Redmine identity"]
+        W2["ci-green.sh → Gitea admin"]
+        W3["redmine-sweep.sh"]
+    end
+    subgraph fleet["Fleet consumers (rotation waves)"]
+        B["KNELBMS: on-box secrets.yaml\n(deploy webhook, gitea_auth_header,\nkuma_push_url, pve_*_api_token)"]
+        K["pfv-k8s secrets\n(glpi-creds for kuma-glpi-bridge,\nflux PAT — wave 4)"]
+        R["Nightly timers\n(pve-config-backup, glpi-reconcile)"]
+    end
+    S1 --> C
+    S2 --> C
+    W1 --> C
+    W2 --> C
+    W3 --> C
+    C -->|HTTPS: prelogin/login(TOTP)/sync/CRUD| V
+    C -.->|reads after rotation| B
+    C -.-> K
+    C -.-> R
+```
+
+## Auth + unlock sequence
+
+```mermaid
+sequenceDiagram
+    participant U as Operator/Consumer
+    participant S as smcli (container)
+    participant V as pwvault.turnsys.com
+    U->>S: smcli login (SM_EMAIL, SM_PASSWORD, SM_TOTP_SECRET)
+    S->>V: POST /api/accounts/prelogin {email}
+    V-->>S: kdf type + iterations (PBKDF2 600k / Argon2id)
+    S->>S: masterKey = KDF(password, email); authHash = PBKDF2(masterKey, password, 1)
+    S->>V: POST /identity/connect/token (password grant, device fields)
+    V-->>S: 400 Two factor required (provider 0)
+    S->>S: TOTP code from SM_TOTP_SECRET (RFC 6238)
+    S->>V: token request + twoFactorToken
+    V-->>S: access_token (+Key, PrivateKey)
+    S->>V: GET /api/sync
+    V-->>S: profile.key (enc, type 2)
+    S->>S: stretchedKey = HKDF(masterKey,"enc"/"mac"); userSymKey = decrypt(profile.key)
+    S->>S: state.json 0600 (master/stretched/user keys + token)
+```
+
+Item payloads are type-2 encStrings (AES-256-CBC + HMAC-SHA256, 32B enc +
+32B mac keys) encrypted/decrypted locally; the vault never sees plaintext.
+
+## Consumer pattern
+
+```bash
+# env-style consumers (export lines; URI/USERNAME/PASSWORD + original keys)
+eval "$(docker exec -i ukrrs-secretsmgr-cli smcli env creds/<name>)"
+# surgical single-field rotation update
+docker exec -i ukrrs-secretsmgr-cli smcli setfield creds/<name> <KEY> <newvalue>
+```
+
+Items are LOGIN type: username + password/API-key as first-class fields,
+service URLs as URIs, remaining keys as named custom fields (hidden where
+secret). Original env key names are preserved as the custom-field names.
+
+## Rotation program (#829)
+
+All pre-migration material is presumed BURNED (plaintext on disk + LLM
+exposure). Waves, each item = rotate at source → `setfield` in vault →
+rewire consumers to `sm env` → validate (guard rule: never write the vault
+from a failed rotation):
+
+1. Tooling tokens (librenms*, wazuh ✓, grafana*, gvm, technitium, phpipam,
+   rancher-sectest, beszel, pihole, PMG pair, PBS tokens)
+2. Agent identities (gitea agent-stack, vptechops gitea/redmine)
+3. Platform (cloudron API token, kuma, discourse)
+4. Deep-wired (flux PAT + gitea runner + HA on-box secrets.yaml + k8s
+   secrets + PVE upsagent tokens)
+
+\* item-specific notes: librenms — token-mint API route absent on this
+install (UI/DB path); grafana — SSO-managed, local admin parked (instance
+password policy blocks CLI reset).
+
+## KNELBMS integration
+
+See [integration-knelbms.md](integration-knelbms.md). Short form: the BMS
+deploy pipeline (KNELBMS repo, packages/deploy_pipeline.yaml) consumes
+`gitea_auth_header` + `kuma_push_url` + `pve_*_api_token` from on-box
+`/config/secrets.yaml`; the vault is the source of record and rotations
+push to the box through the CR-gated provisioning path.

docs/integration-knelbms.md

@@ -0,0 +1,50 @@
+# KNELBMS integration — secrets flow
+
+Repo: https://git.knownelement.com/KNEL/KNELBMS (PhysicalPlant lane;
+Home Assistant BMS on VM 100 @ pfv-tsys1, dev→release deploy by git).
+
+## What the BMS consumes
+
+| On-box secret (HA `secrets.yaml`) | Vault item + field | Provenance |
+|---|---|---|
+| `gitea_auth_header` | `creds/pfv-bms-deploy` → GITEA_DEPLOY_WATCH_TOKEN | release-branch sha-watch REST sensor |
+| `deploy_webhook_id` / `pfv_relay_webhook_id` | `creds/pfv-bms-deploy` | fast-path deploy webhook |
+| `kuma_push_url` | `creds/pfv-bms-beta` sibling — dead-man monitor `pfv-bms-ha-heartbeat-2026-09` (push token) | rotated 2026-09-06 under CR 21 |
+| `pve_tsys{1,3,4,5,6,7}_api_token` | `creds/pve-upsagent` (per-node fields) | `upsagent@pam!ups`, PVEAdmin-on-/vms, privsep=0 |
+| `doorman_*`, `pfvbms_smb_*`, beta HA creds | respective `creds/*` items | as rotated |
+
+## Provisioning + rotation flow
+
+```mermaid
+sequenceDiagram
+    participant R as Rotation run (#829 wave)
+    participant V as TSGCOO vault
+    participant B as pfv-bms on-box secrets.yaml
+    participant H as Home Assistant
+    R->>V: new value (sm setfield creds/<item> <KEY>)
+    R->>B: CR-gated provisioning (ssh -p 22222, CR + ha core check)
+    R->>H: ha core restart (Kuma window; dead-man covers the gap)
+    H-->>R: post-deploy validation (entity/heartbeat checks)
+    R->>V: rotation evidence on #829
+```
+
+Rules that bind this flow (house rules + #811):
+
+- pfv-bms prod changes need a GLPI CR **and** a Kuma maintenance window
+  when a restart is involved; the deploy path itself stays
+  dev → CI → release PR (founder merges).
+- HA runtime template contexts cannot read secrets — the on-box
+  `shell_command` entries reference `!secret` names only (see KNELBMS
+  PR #6 / CR 21 for the dead-man fix that taught us this).
+- The dead-man heartbeat (`pfv-bms-ha-heartbeat-2026-09`, Kuma id 291)
+  is the canary for provisioning mistakes: if the on-box secret and the
+  vault disagree, the push fails and the monitor pages.
+
+## Current integration state (2026-09-06)
+
+- On-box `secrets.yaml` provisioned manually under CR 21 (PVE tokens +
+  rotated Kuma push URL); git-side KNELBMS matches for the `!secret`
+  keys it owns (PR #6 on dev, awaiting founder release merge).
+- Automated push-from-vault (rotation waves writing the box directly via
+  the AWX ssh path) is **planned, not built** — wave 4. Until then the
+  table above is the manual runbook, executed under CR.