provision-secrets: ai) case -> provision-stack.py + normalized env block (branch rebuilt on current main)

Same two scoped changes as PR #15's rebase, now applied on a branch cut
from TODAY's main so the merge base is current (PR #15 became
unmergeable — its 09-03 merge base predated main's heavy hand-edits):

1. ai) case -> `python3 deploy/provision-stack.py ai` (one line). Retires
   main's heredoc with its live drift: AI_LITELLM_POSTGRES_PASSWORD
   printed twice (one copy from retired AI_LITELLM_DB_PASSWORD), stray
   legacy AI_WEBUI_SECRET_KEY, and the WEB_UI/WEBUI shell-var mismatch
   that renders an EMPTY OAuth client secret into ai.env.
2. env block ai section: normalized to exact template placeholder names,
   added AI_MCPO_API_KEY (NEW Woodpecker secret required pre-merge),
   pruned now-template-literal vars.

apk gains python3 py3-yaml. Everything else byte-matches current main.
Dollar-escape audit done per header lesson.

.woodpecker/deploy.yml

@@ -48,6 +48,17 @@ when:
 #   already unconditionally rsynced at the top of the deploy step regardless
 #   of which stacks changed, so it's safe to exclude it from the stack list
 #   everywhere folders are detected below.
+#
+#   2026-09-03/07 INCIDENT + REDESIGN: hand-maintained grep -v + printf
+#   line-surgery cases repeatedly drifted (missing '=', duplicated keys,
+#   mismatched env var names rendering EMPTY secrets) and broke live
+#   services. Root cause: no authoritative key list and shell heredocs
+#   hostile to hand-editing. Stacks migrate one at a time to a data-driven
+#   model: secrets/secrets-map.yaml (data only) + per-stack .env.template
+#   (authoritative FULL file) + deploy/provision-stack.py (whole-file
+#   render, hard failure naming any missing value). Migrated stacks call
+#   the script; unmigrated stacks keep legacy case entries until their own
+#   PR. See the "Secrets & Deployment Architecture — Global Direction" note.
 # ─────────────────────────────────────────────────────────────────────────────
 
 steps:
@@ -163,6 +174,11 @@ steps:
         from_secret: entertainment_sparky_encryption_key
       ENTERTAINMENT_BETTER_AUTH_SECRET:
         from_secret: entertainment_better_auth_secret
+      # ── ai stack (manifest-driven — secrets/secrets-map.yaml +
+      #    ai/ai.env.template + deploy/provision-stack.py). Env var names
+      #    below match the template placeholders EXACTLY; this block is the
+      #    only per-secret touchpoint left in this file for migrated stacks
+      #    (Woodpecker v3 requires explicit from_secret declarations). ──
       AI_AWS_ACCESS_KEY_ID:
         from_secret: ai_aws_access_key_id
       AI_AWS_SECRET_ACCESS_KEY:
@@ -171,16 +187,26 @@ steps:
         from_secret: ai_litellm_master_key
       AI_LITELLM_SALT_KEY:
         from_secret: ai_litellm_salt_key
-      AI_LITELLM_DB_PASSWORD:
-        from_secret: ai_litellm_db_password
-      AI_WEBUI_SECRET_KEY:
+      AI_LITELLM_DATABASE_URL:
+        from_secret: ai_litellm_database_url
+      AI_LITELLM_POSTGRES_PASSWORD:
+        from_secret: ai_litellm_postgres_password
+      AI_OPEN_WEBUI_SECRET_KEY:
         from_secret: ai_webui_secret_key
       AI_OPEN_WEBUI_DATABASE_URL:
         from_secret: ai_open_webui_database_url
-      AI_OAUTH_CLIENT_SECRET:
+      AI_OPEN_WEBUI_OAUTH_CLIENT_SECRET:
         from_secret: ai_oauth_client_secret
+      AI_MCPO_API_KEY:
+        from_secret: ai_mcpo_api_key
+      FLOWAGENT_AZURE_CLIENT_ID:
+        from_secret: flowagent_azure_client_id
+      FLOWAGENT_AZURE_TENANT_ID:
+        from_secret: flowagent_azure_tenant_id
+      FLOWAGENT_AZURE_CLIENT_SECRET:
+        from_secret: flowagent_azure_client_secret
     commands:
-      - apk add --no-cache openssh-client
+      - apk add --no-cache openssh-client python3 py3-yaml
       - mkdir -p ~/.ssh
       - echo "$SSH_KEY" | base64 -d > ~/.ssh/id_rsa
       - chmod 600 ~/.ssh/id_rsa
@@ -221,10 +247,38 @@ steps:
               ssh -o StrictHostKeyChecking=no root@$${SWARM_MANAGER_IP} "source /tmp/cs.sh
                 create_or_update_secret 'gamma_auth_token' '$${GAMMA_AUTH_TOKEN}'";;
             git)
-              ssh -o StrictHostKeyChecking=no root@$${SWARM_MANAGER_IP} "source /tmp/cs.sh
-                create_or_update_secret 'git_db_password'       '$${GIT_DB_PASSWORD}'
-                create_or_update_secret 'git_runner_token'      '$${GIT_RUNNER_TOKEN}'
-                create_or_update_secret 'git_mcp_access_token'  '$${GIT_MCP_ACCESS_TOKEN}'";;
+              # PATTERN B, DELIBERATE (see decision notes) — git hosts the source
+              # of truth for every other stack's compose files, so it must be
+              # restorable from a flat git.yaml + git.env backup alone, with zero
+              # dependency on a running Swarm's Docker secret store. Native Docker
+              # secrets (Pattern C) can't satisfy that: they only exist inside an
+              # already-running Swarm, which is exactly the circular dependency
+              # this stack can't have. Mirrors the retired ai) case's grep -v +
+              # printf rewrite-in-place approach, never sed (values may contain
+              # slash, dollar sign, ampersand).
+              #
+              # TEST PHASE: target is git.env.pipelinetest, NOT the real git.env.
+              # The real file is never opened for writing by this step. First run
+              # seeds the test file from the real git.env (carries over all
+              # non-secret lines untouched); every push after that only refreshes
+              # the 3 secret lines below. Also strips the legacy GITEA_ACCESS_TOKEN
+              # key name so the test file converges on the git.env.example-
+              # documented GITEA_MCP_ACCESS_TOKEN key. Cutover to the real file —
+              # and pointing git.yaml/stack-deploy at it — is a deliberate,
+              # separate follow-up after manually diffing this render.
+              # (Candidate for the secrets-map.yaml/provision-stack.py migration
+              # in its own PR; kept legacy for now.)
+              ssh -o StrictHostKeyChecking=no root@$${SWARM_MANAGER_IP} "FILE=/volume1/docker/compose-files/git.env.pipelinetest
+                TMP=\$FILE.tmp.\$\$
+                [ -f \$FILE ] || cp /volume1/docker/compose-files/git.env \$FILE
+                grep -vE '^(GITEA__database__PASSWD|GITEA_RUNNER_REGISTRATION_TOKEN|GITEA_MCP_ACCESS_TOKEN|GITEA_ACCESS_TOKEN)=' \$FILE > \$TMP 2>/dev/null || touch \$TMP
+                { cat \$TMP
+                  printf 'GITEA__database__PASSWD=%s\n' '$${GIT_DB_PASSWORD}'
+                  printf 'GITEA_RUNNER_REGISTRATION_TOKEN=%s\n' '$${GIT_RUNNER_TOKEN}'
+                  printf 'GITEA_MCP_ACCESS_TOKEN=%s\n' '$${GIT_MCP_ACCESS_TOKEN}'
+                } > \$FILE
+                rm -f \$TMP
+                echo '  [OK] git.env.pipelinetest updated - real git.env untouched'";;
             homeassistant)
               ssh -o StrictHostKeyChecking=no root@$${SWARM_MANAGER_IP} "source /tmp/cs.sh
                 create_or_update_secret 'homeassistant_frigate_rtsp_password' '$${HOMEASSISTANT_FRIGATE_RTSP_PASSWORD}'
@@ -261,40 +315,15 @@ steps:
                 create_or_update_secret 'vaultwarden_admin_token'     '$${VAULTWARDEN_ADMIN_TOKEN}'
                 create_or_update_secret 'vaultwarden_database_url_v2' '$${VAULTWARDEN_DATABASE_URL}'";;
             ai)
-              # NOTE: ai stack uses Pattern B (host .env, not native Docker secrets) —
-              # LiteLLM/boto3 and Open WebUI don't support the _FILE convention for these
-              # vars. Instead of Docker secrets, we regenerate ONLY the migrated lines in
-              # the remote ai/ai.env in place via grep -v + printf (never sed, since values
-              # may contain slash, dollar sign, ampersand). All other lines — including
-              # MCPO_API_KEY, OAUTH_CLIENT_ID, WEBUI_URL, and other non-secret config — are
-              # left completely untouched. MCPO_API_KEY migration is deferred to a
-              # follow-up; this step never reads or writes it.
-              #
-              # IMPORTANT: ai.yaml's litellm service references ${AI_AWS_ACCESS_KEY_ID} /
-              # ${AI_AWS_SECRET_ACCESS_KEY} (AI_-prefixed) and renders them into the
-              # container as plain AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY (see commit
-              # 37ed671a, "Change AWS keys to use Woodpecker Secrets"). So ai.env must be
-              # written with the AI_-prefixed key names, NOT the plain ones — writing
-              # plain AWS_ACCESS_KEY_ID here would leave ${AI_AWS_ACCESS_KEY_ID}
-              # unresolved at compose-render time (empty), silently breaking Bedrock auth.
-              # All other migrated vars in ai.yaml use plain (unprefixed) names, so only
-              # these two lines need the AI_ prefix.
-              ssh -o StrictHostKeyChecking=no root@$${SWARM_MANAGER_IP} "FILE=/volume1/docker/compose-files/ai/ai.env
-                TMP=\$FILE.tmp.\$\$
-                grep -vE '^(AI_AWS_ACCESS_KEY_ID|AI_AWS_SECRET_ACCESS_KEY|LITELLM_MASTER_KEY|LITELLM_SALT_KEY|POSTGRES_PASSWORD|DATABASE_URL|WEBUI_SECRET_KEY|OPEN_WEBUI_DATABASE_URL|OAUTH_CLIENT_SECRET)=' \$FILE > \$TMP 2>/dev/null || touch \$TMP
-                { cat \$TMP
-                  printf 'AI_AWS_ACCESS_KEY_ID=%s\n' '$${AI_AWS_ACCESS_KEY_ID}'
-                  printf 'AI_AWS_SECRET_ACCESS_KEY=%s\n' '$${AI_AWS_SECRET_ACCESS_KEY}'
-                  printf 'LITELLM_MASTER_KEY=%s\n' '$${AI_LITELLM_MASTER_KEY}'
-                  printf 'LITELLM_SALT_KEY=%s\n' '$${AI_LITELLM_SALT_KEY}'
-                  printf 'POSTGRES_PASSWORD=%s\n' '$${AI_LITELLM_DB_PASSWORD}'
-                  printf 'DATABASE_URL=postgresql://LiteLLM:%s@postgresql:5432/litellm\n' '$${AI_LITELLM_DB_PASSWORD}'
-                  printf 'WEBUI_SECRET_KEY=%s\n' '$${AI_WEBUI_SECRET_KEY}'
-                  printf 'OPEN_WEBUI_DATABASE_URL=%s\n' '$${AI_OPEN_WEBUI_DATABASE_URL}'
-                  printf 'OAUTH_CLIENT_SECRET=%s\n' '$${AI_OAUTH_CLIENT_SECRET}'
-                } > \$FILE
-                rm -f \$TMP
-                echo '  [OK] ai/ai.env secrets updated'";;
+              # MIGRATED (2026-09-07) to manifest-driven provisioning after the
+              # line-surgery approach repeatedly drifted (dropped keys 2026-09-03;
+              # duplicate AI_LITELLM_POSTGRES_PASSWORD + stray legacy keys +
+              # WEB_UI/WEBUI env-name mismatch rendering an EMPTY OAuth client
+              # secret, found 2026-09-07). All logic lives in
+              # deploy/provision-stack.py; the authoritative key list lives in
+              # ai/ai.env.template; the mapping lives in secrets/secrets-map.yaml.
+              # This case is intentionally one line.
+              python3 deploy/provision-stack.py ai;;
             entertainment)
               ssh -o StrictHostKeyChecking=no root@$${SWARM_MANAGER_IP} "source /tmp/cs.sh
                 create_or_update_secret 'entertainment_discord_token'          '$${ENTERTAINMENT_DISCORD_TOKEN}'
@@ -352,7 +381,7 @@ steps:
 
           # Bootstrap-tier guard: file synced to host, deploy is MANUAL
           case "$STACK" in
-            traefik|woodpecker|postgresql|secrets)
+            traefik|woodpecker|postgresql|secrets|git)
               echo "  [BOOTSTRAP] $STACK: file synced. Deploy is MANUAL."
               echo "  Run: bash /volume1/docker/compose-files/deploy/stack-deploy.sh $STACK"
               continue ;;

ai/ai.env.template

@@ -0,0 +1,55 @@
+# ─────────────────────────────────────────────────────────────────────────────
+# ai.env.template — AUTHORITATIVE template for ai/ai.env (rendered by
+# deploy/provision-stack.py per secrets/secrets-map.yaml).
+#
+#   - This file IS the complete key list for ai.env. The whole file is
+#     rendered on every provisioning run — no line surgery, so a key can
+#     never silently go missing again (root cause of the 2026-09-03 outage).
+#   - Key names match EXACTLY what ai/ai.yaml references (AI_<SERVICE>_*
+#     naming adopted on main 2026-09-06).
+#   - Non-secret config lives here as LITERAL values (visible, reviewable).
+#   - Secret values are dollar-brace placeholders resolved from the CI env
+#     (Woodpecker from_secret vars) at provisioning time. provision-stack.py
+#     FAILS HARD if any placeholder is missing/empty.
+#   - The rendered ai/ai.env exists only on the host (gitignored).
+#   - Rendered by provision-stack.py, NOT Woodpecker's yaml preprocessor —
+#     single-dollar placeholders are safe here (deploy.yml's double-dollar
+#     rule does NOT apply to this file).
+#
+# Consumed by ai/ai.yaml. DOMAIN_NAME comes from deploy/global.env, not here.
+# ─────────────────────────────────────────────────────────────────────────────
+
+# ── LiteLLM (non-secret config) ──────────────────────────────────────────────
+AI_AWS_REGION_NAME=us-east-2
+AI_LITELLM_MODIFY_PARAMS=False
+AI_LITELLM_DATABASE_MIGRATIONS=True
+
+# ── LiteLLM (secrets) ────────────────────────────────────────────────────────
+AI_AWS_ACCESS_KEY_ID=${AI_AWS_ACCESS_KEY_ID}
+AI_AWS_SECRET_ACCESS_KEY=${AI_AWS_SECRET_ACCESS_KEY}
+AI_LITELLM_MASTER_KEY=${AI_LITELLM_MASTER_KEY}
+AI_LITELLM_SALT_KEY=${AI_LITELLM_SALT_KEY}
+# Full connection URL is itself a secret (ai_litellm_database_url) — the
+# URL structure never appears in git.
+AI_LITELLM_DATABASE_URL=${AI_LITELLM_DATABASE_URL}
+AI_LITELLM_POSTGRES_PASSWORD=${AI_LITELLM_POSTGRES_PASSWORD}
+
+# ── Open WebUI (non-secret config) ───────────────────────────────────────────
+AI_OPEN_WEBUI_URL=https://ai.bryanmail.net
+AI_OPEN_WEBUI_ENABLE_OAUTH_SIGNUP=true
+AI_OPEN_WEBUI_OAUTH_MERGE_ACCOUNTS_BY_EMAIL=true
+AI_OPEN_WEBUI_OAUTH_PROVIDER_NAME=Authentik
+AI_OPEN_WEBUI_OPENID_PROVIDER_URL=https://auth.bryanmail.net/application/o/open-web-ui/.well-known/openid-configuration
+# OAuth client ID is a public identifier by OAuth2 design (it is sent to the
+# browser); the client SECRET below is the protected credential.
+AI_OPEN_WEBUI_OAUTH_CLIENT_ID=hVmhi1dS3TnG2cUw5QwLOx5FDLSWtnQUdZyeB5zK
+AI_OPEN_WEBUI_OAUTH_SCOPES=openid email profile
+AI_OPEN_WEBUI_OPENID_REDIRECT_URI=https://ai.bryanmail.net/oauth/oidc/callback
+
+# ── Open WebUI (secrets) ─────────────────────────────────────────────────────
+AI_OPEN_WEBUI_SECRET_KEY=${AI_OPEN_WEBUI_SECRET_KEY}
+AI_OPEN_WEBUI_DATABASE_URL=${AI_OPEN_WEBUI_DATABASE_URL}
+AI_OPEN_WEBUI_OAUTH_CLIENT_SECRET=${AI_OPEN_WEBUI_OAUTH_CLIENT_SECRET}
+
+# ── mcpo / mcpo-critical (secrets) ───────────────────────────────────────────
+MCPO_API_KEY=${AI_MCPO_API_KEY}

ai/ai.yaml

@@ -1,6 +1,6 @@
 services:
   open-webui:
-    image: ghcr.io/open-webui/open-webui:0.11.1
+    image: ghcr.io/open-webui/open-webui:0.11.3 #
     entrypoint:
     - /bin/bash
     - /app/tools/startup.sh
@@ -16,20 +16,20 @@ services:
       start_period: 60s
     environment:
     - OLLAMA_BASE_URL=http://ollama-intel-arc:11434
-    - WEBUI_SECRET_KEY=${WEBUI_SECRET_KEY}
+    - WEBUI_SECRET_KEY=${AI_OPEN_WEBUI_SECRET_KEY}
     - WEBUI_DB_HOST=postgresql
-    - DATABASE_URL=${OPEN_WEBUI_DATABASE_URL}
+    - DATABASE_URL=${AI_OPEN_WEBUI_DATABASE_URL}
     - ENABLE_TOOL_SERVER_CALLING=true
     - TOOL_SERVER_CALLING=true
-    - WEBUI_URL=${WEBUI_URL}
-    - ENABLE_OAUTH_SIGNUP=${ENABLE_OAUTH_SIGNUP}
-    - OAUTH_MERGE_ACCOUNTS_BY_EMAIL=${OAUTH_MERGE_ACCOUNTS_BY_EMAIL}
-    - OAUTH_PROVIDER_NAME=${OAUTH_PROVIDER_NAME}
-    - OPENID_PROVIDER_URL=${OPENID_PROVIDER_URL}
-    - OAUTH_CLIENT_ID=${OAUTH_CLIENT_ID}
-    - OAUTH_CLIENT_SECRET=${OAUTH_CLIENT_SECRET}
-    - OAUTH_SCOPES=${OAUTH_SCOPES}
-    - OPENID_REDIRECT_URI=${OPENID_REDIRECT_URI}
+    - WEBUI_URL=${AI_OPEN_WEBUI_URL}
+    - ENABLE_OAUTH_SIGNUP=${AI_OPEN_WEBUI_ENABLE_OAUTH_SIGNUP}
+    - OAUTH_MERGE_ACCOUNTS_BY_EMAIL=${AI_OPEN_WEBUI_OAUTH_MERGE_ACCOUNTS_BY_EMAIL}
+    - OAUTH_PROVIDER_NAME=${AI_OPEN_WEBUI_OAUTH_PROVIDER_NAME}
+    - OPENID_PROVIDER_URL=${AI_OPEN_WEBUI_OPENID_PROVIDER_URL}
+    - OAUTH_CLIENT_ID=${AI_OPEN_WEBUI_OAUTH_CLIENT_ID}
+    - OAUTH_CLIENT_SECRET=${AI_OPEN_WEBUI_OAUTH_CLIENT_SECRET}
+    - OAUTH_SCOPES=${AI_OPEN_WEBUI_OAUTH_SCOPES}
+    - OPENID_REDIRECT_URI=${AI_OPEN_WEBUI_OPENID_REDIRECT_URI}
     networks:
     - traefik_backend
     - postgresql_db-backend
@@ -62,13 +62,13 @@ services:
     environment:
     - AWS_ACCESS_KEY_ID=${AI_AWS_ACCESS_KEY_ID}
     - AWS_SECRET_ACCESS_KEY=${AI_AWS_SECRET_ACCESS_KEY}
-    - AWS_REGION_NAME=${AWS_REGION_NAME}
-    - LITELLM_MASTER_KEY=${LITELLM_MASTER_KEY}
-    - LITELLM_SALT_KEY=${LITELLM_SALT_KEY}
-    - DATABASE_URL=${DATABASE_URL}
-    - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
-    - LITELLM_MODIFY_PARAMS=${LITELLM_MODIFY_PARAMS}
-    - LITELLM_DATABASE_MIGRATIONS=${LITELLM_DATABASE_MIGRATIONS}
+    - AWS_REGION_NAME=${AI_AWS_REGION_NAME}
+    - LITELLM_MASTER_KEY=${AI_LITELLM_MASTER_KEY}
+    - LITELLM_SALT_KEY=${AI_LITELLM_SALT_KEY}
+    - DATABASE_URL=${AI_LITELLM_DATABASE_URL}
+    - POSTGRES_PASSWORD=${AI_LITELLM_POSTGRES_PASSWORD}
+    - LITELLM_MODIFY_PARAMS=${AI_LITELLM_MODIFY_PARAMS}
+    - LITELLM_DATABASE_MIGRATIONS=${AI_LITELLM_DATABASE_MIGRATIONS}
     networks:
     - traefik_backend
     - postgresql_db-backend
@@ -127,7 +127,7 @@ services:
       labels:
       - traefik.enable=false
   mcpo:
-    image: ghcr.io/open-webui/mcpo:main
+    image: git.bryanmail.net/homelab/flowagent-mcpo:c5b22618404a
     command:
     - --config
     - /app/config/config.json
@@ -141,6 +141,13 @@ services:
     - /volume1/docker/mcpo/data:/mcpo_data
     - /volume1/docker/cronicle/ssh_keys:/app/ssh_keys:ro
     - /volume1/docker/mcpo/uv-cache:/app/uv-cache
+    secrets:
+    - source: flowagent_azure_client_id
+      target: flowagent_azure_client_id
+    - source: flowagent_azure_tenant_id
+      target: flowagent_azure_tenant_id
+    - source: flowagent_azure_client_secret
+      target: flowagent_azure_client_secret
     networks:
     - traefik_backend
     deploy:
@@ -173,6 +180,13 @@ services:
       - traefik.http.middlewares.n8n.headers.STSPreload=true
       - traefik.http.routers.n8n.middlewares=forwardAuth-authentik@file, crowdsec@file
       - traefik.swarm.network=traefik_backend
+secrets:
+  flowagent_azure_client_id:
+    external: true
+  flowagent_azure_tenant_id:
+    external: true
+  flowagent_azure_client_secret:
+    external: true
 networks:
   traefik_backend:
     external: true

deploy/provision-stack.py

@@ -0,0 +1,198 @@
+#!/usr/bin/env python3
+"""provision-stack.py — manifest-driven env-file rendering + Docker secret
+provisioning for one stack.
+
+Usage (from the CI workspace root, inside the provision-secrets step):
+    python3 deploy/provision-stack.py <stack>
+
+Reads secrets/secrets-map.yaml (data only — no code, no values) and, for the
+named stack:
+
+  1. env_template -> renders the COMPLETE env file. Placeholders of the form
+     dollar-brace VARNAME are resolved from this process's environment (the
+     Woodpecker from_secret-backed vars). The whole file is rendered every
+     run; nothing is line-edited in place, so keys can never silently go
+     missing (root cause of the 2026-09-03 ai.env incident).
+  2. env_dest -> ships the rendered file to the Swarm manager over ssh stdin
+     (write-to-temp + atomic mv, mode 600). The rendered file never touches
+     the CI workspace disk under the repo (no chance of being committed) and
+     never appears on a command line.
+  3. docker_secrets -> for each swarm-secret-name -> ENV_VAR mapping, creates
+     or rotates the Docker secret. Values are passed via ssh stdin only.
+     Rotation uses the same sha256-checksum-label convention as
+     deploy/create-secrets.sh (unchanged secrets are skipped silently).
+
+Safety properties:
+  - FAILS HARD (non-zero) if any referenced env var is missing or empty, and
+    lists the missing NAMES. A partial/broken render can never ship.
+  - FAILS HARD if any unresolved placeholder remains after rendering.
+  - NEVER prints a secret value — names and counts only.
+  - Requires SWARM_MANAGER_IP in the environment and a usable ssh identity
+    (both already set up by the provision-secrets step).
+
+Stacks not present in the manifest exit 0 with a notice, so this script is
+safe to call unconditionally; legacy case-entries in deploy.yml keep handling
+unmigrated stacks.
+"""
+
+import hashlib
+import os
+import re
+import subprocess
+import sys
+
+REPO_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+MANIFEST_PATH = os.path.join(REPO_ROOT, "secrets", "secrets-map.yaml")
+REMOTE_BASE = "/volume1/docker/compose-files"
+PLACEHOLDER_RE = re.compile(r"\$\{([A-Za-z_][A-Za-z0-9_]*)\}")
+
+SSH_OPTS = ["-o", "StrictHostKeyChecking=no"]
+
+
+def die(msg: str) -> None:
+    print(f"ERROR: {msg}", file=sys.stderr)
+    sys.exit(1)
+
+
+def load_manifest() -> dict:
+    try:
+        import yaml  # py3-yaml, installed by the provision-secrets step
+    except ImportError:
+        die("PyYAML not available — provision-secrets step must apk add py3-yaml")
+    if not os.path.isfile(MANIFEST_PATH):
+        die(f"manifest not found: {MANIFEST_PATH}")
+    with open(MANIFEST_PATH, "r", encoding="utf-8") as fh:
+        data = yaml.safe_load(fh) or {}
+    stacks = data.get("stacks")
+    if not isinstance(stacks, dict):
+        die("manifest has no 'stacks:' mapping")
+    return stacks
+
+
+def ssh_target() -> str:
+    ip = os.environ.get("SWARM_MANAGER_IP", "").strip()
+    if not ip:
+        die("SWARM_MANAGER_IP is empty — check Woodpecker repo secrets")
+    return f"root@{ip}"
+
+
+def ssh_run(target: str, remote_cmd: str, stdin_data: bytes | None = None,
+            check: bool = True) -> subprocess.CompletedProcess:
+    proc = subprocess.run(
+        ["ssh", *SSH_OPTS, target, remote_cmd],
+        input=stdin_data, capture_output=True,
+    )
+    if check and proc.returncode != 0:
+        # stderr may be verbose but must never contain our secret values —
+        # we only ever send values via stdin, never embed them in remote_cmd.
+        die(f"remote command failed (rc={proc.returncode}): {remote_cmd}\n"
+            f"{proc.stderr.decode(errors='replace').strip()}")
+    return proc
+
+
+def render_template(template_path: str) -> str:
+    if not os.path.isfile(template_path):
+        die(f"env_template not found: {template_path}")
+    with open(template_path, "r", encoding="utf-8") as fh:
+        raw = fh.read()
+
+    referenced = sorted(set(PLACEHOLDER_RE.findall(raw)))
+    missing = [v for v in referenced
+               if not os.environ.get(v, "").strip()]
+    if missing:
+        die("template references vars that are MISSING or EMPTY in the CI "
+            "environment (add them via from_secret in "
+            ".woodpecker/deploy.yml provision-secrets, and as Woodpecker "
+            "secrets):\n  " + "\n  ".join(missing))
+
+    rendered = PLACEHOLDER_RE.sub(lambda m: os.environ[m.group(1)], raw)
+
+    # Belt-and-braces: nothing placeholder-shaped may survive the render.
+    leftover = sorted(set(PLACEHOLDER_RE.findall(rendered)))
+    if leftover:
+        die("unresolved placeholders remain after rendering: "
+            + ", ".join(leftover))
+
+    print(f"  [render] {template_path}: {len(referenced)} secret placeholder(s) "
+          f"resolved: {', '.join(referenced)}")
+    return rendered
+
+
+def ship_env_file(target: str, rendered: str, dest_rel: str) -> None:
+    dest = f"{REMOTE_BASE}/{dest_rel}"
+    tmp = f"{dest}.provision-tmp"
+    # Value travels over ssh stdin; never on a command line; atomic mv.
+    ssh_run(target,
+            f"umask 077 && cat > {tmp} && chmod 600 {tmp} && mv {tmp} {dest}",
+            stdin_data=rendered.encode())
+    keys = [ln.split("=", 1)[0] for ln in rendered.splitlines()
+            if "=" in ln and not ln.lstrip().startswith("#") and ln.strip()]
+    print(f"  [env] shipped {dest} ({len(keys)} keys): {', '.join(keys)}")
+
+
+def provision_docker_secret(target: str, name: str, env_var: str) -> None:
+    value = os.environ.get(env_var, "")
+    if not value.strip():
+        die(f"docker secret '{name}': env var {env_var} is missing/empty")
+
+    new_hash = hashlib.sha256(value.encode()).hexdigest()
+    probe = ssh_run(
+        target,
+        f"docker secret inspect {name} "
+        "--format '{{index .Spec.Labels \"checksum\"}}' 2>/dev/null || true",
+        check=True)
+    old_hash = probe.stdout.decode().strip()
+
+    if old_hash == new_hash:
+        print(f"  [skip]   docker secret {name} (unchanged)")
+        return
+
+    if old_hash:
+        rm = ssh_run(target, f"docker secret rm {name}", check=False)
+        if rm.returncode != 0:
+            die(f"docker secret {name}: value changed but removal failed — "
+                "it is probably referenced by a running service. Provision "
+                "under a versioned name (see vaultwarden_database_url_v2 "
+                "precedent) or scale the service down first.")
+        action = "update"
+    else:
+        action = "create"
+
+    ssh_run(target,
+            f"docker secret create --label checksum={new_hash} "
+            f"--label managed-by=woodpecker {name} -",
+            stdin_data=value.encode())
+    print(f"  [{action}] docker secret {name} (value via stdin)")
+
+
+def main() -> None:
+    if len(sys.argv) != 2:
+        die("usage: provision-stack.py <stack>")
+    stack = sys.argv[1]
+
+    stacks = load_manifest()
+    cfg = stacks.get(stack)
+    if cfg is None:
+        print(f"  [info] stack '{stack}' not in secrets-map.yaml — "
+              "legacy provisioning (deploy.yml case-entry) applies. Nothing to do.")
+        return
+
+    target = ssh_target()
+    print(f"==> provision-stack: {stack}")
+
+    template_rel = cfg.get("env_template")
+    dest_rel = cfg.get("env_dest")
+    if template_rel and not dest_rel:
+        die("env_template set but env_dest missing in manifest")
+    if template_rel:
+        rendered = render_template(os.path.join(REPO_ROOT, template_rel))
+        ship_env_file(target, rendered, dest_rel)
+
+    for name, env_var in (cfg.get("docker_secrets") or {}).items():
+        provision_docker_secret(target, name, env_var)
+
+    print(f"==> provision-stack: {stack} done")
+
+
+if __name__ == "__main__":
+    main()

secrets/secrets-map.yaml

@@ -0,0 +1,38 @@
+# ─────────────────────────────────────────────────────────────────────────────
+# secrets-map.yaml — DATA-ONLY manifest for deploy/provision-stack.py
+#
+# RULES:
+#   - This file contains NO code, NO shell, NO secret values — only names.
+#   - Each stack entry declares:
+#       env_template:   repo path of the FULL env-file template (tracked).
+#                       The template is authoritative: the COMPLETE env file is
+#                       rendered from it on every provisioning run. Nothing is
+#                       line-edited in place, so keys can never silently go
+#                       missing.
+#       env_dest:       host path (relative to /volume1/docker/compose-files/)
+#                       the rendered env file is shipped to. Rendered file
+#                       exists ONLY on the host — never committed to git.
+#       docker_secrets: map of docker-swarm-secret-name -> CI ENV VAR NAME
+#                       (Pattern C). The env var must be declared via
+#                       from_secret: in .woodpecker/deploy.yml's
+#                       provision-secrets step (Woodpecker v3 requires explicit
+#                       per-secret declaration; there is no expose-all).
+#
+# ADDING A NEW SECRET (3 small steps, no shell edits):
+#   1. Add the secret value in Woodpecker UI (repo Settings -> Secrets).
+#   2. Declare it in .woodpecker/deploy.yml provision-secrets environment:
+#      block (from_secret) — mechanical two-line addition.
+#   3. Reference it here (docker_secrets:) and/or in the stack's
+#      .env.template as a dollar-brace placeholder.
+#
+# Stacks not listed here fall through to deploy.yml's legacy case-entries
+# untouched. Migration is deliberately one stack per PR.
+# ─────────────────────────────────────────────────────────────────────────────
+stacks:
+  ai:
+    env_template: ai/ai.env.template
+    env_dest: ai/ai.env
+    docker_secrets:
+      flowagent_azure_client_id: FLOWAGENT_AZURE_CLIENT_ID
+      flowagent_azure_tenant_id: FLOWAGENT_AZURE_TENANT_ID
+      flowagent_azure_client_secret: FLOWAGENT_AZURE_CLIENT_SECRET