feat(mcp): write, test and delete this instance's own playbooks (#5748)

#5748 "MCP can write, edit and test this instance's own playbooks
(steward-local)": two admin tools beside run_playbook, so an agent can
build automation for these hosts that isn't part of Steward itself.

- save_playbook(playbook, content, source="steward-local", replace=false)
  goes through the editor's own validate_playbook_yaml and
  write_playbook. A playbook that isn't a list of plays comes back with
  the YAML error and its line, and nothing is written. An existing file
  is overwritten only with replace=true. steward-builtin is refused
  (Steward's playbooks change in its repository), as are git sources
  (their repo owns them).
- After writing, `ansible-playbook --syntax-check` runs against a
  throwaway localhost inventory (no host contacted, 60 s), so a module
  name Ansible can't resolve shows before any run. The result rides the
  reply; the file is kept either way.
- delete_playbook(playbook, source="steward-local").
- Both audit as ansible.playbook.created/updated/deleted, "<user> via
  MCP", with the size and never the content.

The testing loop is the existing one: run_playbook defaults to check
mode, on one target, then get_ansible_run. The integration test does
that loop with the real ansible-playbook.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

steward/ansible/executor.py

@@ -329,6 +329,36 @@ def ansible_env(creds: dict | None, base_env) -> dict:
     return env
 
 
+# Long enough for Ansible to load its collections on a cold start; a syntax
+# check reaches no host.
+SYNTAX_CHECK_SECONDS = 60
+SYNTAX_OUTPUT_CHARS = 2000
+
+
+async def syntax_check(source_path: str, relative_path: str) -> tuple[bool | None, str]:
+    """`ansible-playbook --syntax-check` on a saved playbook (#5748): catches
+    what a YAML parse doesn't, such as a module name Ansible can't resolve or
+    a missing role. Runs in the source's folder against a throwaway localhost
+    inventory, so no host is contacted. Returns (True, ""), (False, output),
+    or (None, why) when the check couldn't run."""
+    try:
+        proc = await asyncio.create_subprocess_exec(
+            "ansible-playbook", "--syntax-check", "-i", "localhost,", relative_path,
+            stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.STDOUT,
+            cwd=source_path, env=ansible_env({}, os.environ), start_new_session=True)
+    except FileNotFoundError:
+        return None, "ansible-playbook isn't installed here, so only the YAML was checked."
+    try:
+        out, _ = await asyncio.wait_for(proc.communicate(), SYNTAX_CHECK_SECONDS)
+    except asyncio.TimeoutError:
+        await _terminate_then_kill(proc, grace=2.0)
+        return None, f"The syntax check took over {SYNTAX_CHECK_SECONDS} s and was stopped."
+    text = out.decode(errors="replace").strip()
+    if proc.returncode == 0:
+        return True, ""
+    return False, text[-SYNTAX_OUTPUT_CHARS:]
+
+
 # ── Structured results (parsed from the streamed output) ──────────────────────
 
 _RECAP_RE = re.compile(

steward/ansible/sources.py

@@ -212,7 +212,7 @@ def read_playbook(source_path: str, relative_path: str) -> str | None:
     return target.read_text(errors="replace")
 
 
-def _resolve_writable(source_path: str, relative_path: str) -> tuple[Path | None, str | None]:
+def resolve_writable(source_path: str, relative_path: str) -> tuple[Path | None, str | None]:
     """Resolve relative_path under source_path, guarding traversal + extension.
 
     Returns (target_path, None) on success or (None, error). The root need not
@@ -233,7 +233,7 @@ def _resolve_writable(source_path: str, relative_path: str) -> tuple[Path | None
 
 def write_playbook(source_path: str, relative_path: str, content: str) -> tuple[bool, str | None]:
     """Write a playbook into a source, creating parent dirs. Traversal-guarded."""
-    target, err = _resolve_writable(source_path, relative_path)
+    target, err = resolve_writable(source_path, relative_path)
     if err:
         return False, err
     try:
@@ -246,7 +246,7 @@ def write_playbook(source_path: str, relative_path: str, content: str) -> tuple[
 
 def delete_playbook(source_path: str, relative_path: str) -> tuple[bool, str | None]:
     """Delete a playbook from a source. Traversal-guarded; no-op if absent."""
-    target, err = _resolve_writable(source_path, relative_path)
+    target, err = resolve_writable(source_path, relative_path)
     if err:
         return False, err
     try:

steward/mcp/tools/automation.py

@@ -229,6 +229,88 @@ async def get_playbook(source: str, playbook: str) -> dict:
     return out
 
 
+def _editable(name: str) -> dict:
+    s = _source(name)
+    if not src_module.is_editable_source(s):
+        why = ("it holds Steward's own playbooks, which are changed in Steward's repository"
+               if s["name"] == src_module.BUILTIN_SOURCE_NAME else
+               "it is read from a git repository, which would overwrite a change on its next pull"
+               if s["type"] == "git" else "it is not editable")
+        raise ValueError(f"source {s['name']!r} can't be written here: {why}. Write this "
+                         f"instance's own playbooks into {src_module.LOCAL_SOURCE_NAME!r}.")
+    return s
+
+
+async def save_playbook(playbook: str, content: str, source: str = src_module.LOCAL_SOURCE_NAME,
+                        replace: bool = False) -> dict:
+    """Write a playbook of this instance's own: automation for these hosts that
+    isn't part of Steward itself. The same as saving in Steward's editor.
+
+    playbook: a path in the source ending .yml or .yaml, e.g. "upkeep/prune-runner.yml".
+    content: the whole playbook. It must parse as YAML and be a list of plays,
+      or nothing is saved and the error (with its line) comes back.
+    source: "steward-local" (default), or another editable folder source.
+      steward-builtin is refused: Steward's own playbooks are changed in
+      Steward's repository. Git sources are refused: their repo owns them.
+    replace: an existing playbook is overwritten only with replace=true, so
+      a typo in a name never clobbers another one.
+
+    After saving, `ansible-playbook --syntax-check` runs on it (no host is
+    contacted); `syntax` says whether it passed and, if not, why, e.g. a
+    module name Ansible can't resolve. The file is saved either way.
+
+    To test it: run_playbook(source, playbook, scope="target:<one host>") runs
+    it in check mode, get_ansible_run shows what would change; fix, save with
+    replace=true, run again; apply with check=false once it is right.
+
+    Admin only. Recorded in the audit log (its size, never its content).
+    """
+    s = _editable(source)
+    path = (playbook or "").strip().lstrip("/")
+    target, err = src_module.resolve_writable(s["path"], path)
+    if err:
+        raise ValueError(f"{err}. Nothing was saved.")
+    ok, err = src_module.validate_playbook_yaml(content or "")
+    if not ok:
+        raise ValueError(f"{err}\nNothing was saved.")
+    existed = target.exists()
+    if existed and not replace:
+        raise ValueError(f"{s['name']} already has {path!r}; pass replace=true to overwrite it, "
+                         "or choose another name. Nothing was saved.")
+    ok, err = src_module.write_playbook(s["path"], path, content)
+    if not ok:
+        raise ValueError(f"{err}. Nothing was saved.")
+    await audit_mcp("ansible.playbook.updated" if existed else "ansible.playbook.created",
+                    "ansible_playbook", f"{s['name']}:{path}",
+                    {"source": s["name"], "playbook": path, "bytes": len(content)})
+    passed, output = await executor.syntax_check(s["path"], path)
+    syntax = {"ok": passed}
+    if output:
+        syntax["output" if passed is False else "note"] = output
+    return {"source": s["name"], "playbook": path, "replaced": existed, "bytes": len(content),
+            "syntax": syntax,
+            "next": f'run_playbook(source="{s["name"]}", playbook="{path}", '
+                    'scope="target:<one host>") to see what it would change.'}
+
+
+async def delete_playbook(playbook: str, source: str = src_module.LOCAL_SOURCE_NAME) -> dict:
+    """Delete a playbook written into steward-local (or another editable folder
+    source). Its past runs stay. A schedule or alert action that names it
+    will fail until it is pointed elsewhere.
+
+    Admin only. Recorded in the audit log.
+    """
+    s = _editable(source)
+    path = (playbook or "").strip().lstrip("/")
+    _playbook(s, path)
+    ok, err = src_module.delete_playbook(s["path"], path)
+    if not ok:
+        raise ValueError(err)
+    await audit_mcp("ansible.playbook.deleted", "ansible_playbook", f"{s['name']}:{path}",
+                    {"source": s["name"], "playbook": path})
+    return {"deleted": path, "source": s["name"]}
+
+
 async def list_inventory() -> dict:
     """Steward's Ansible inventory: targets, groups, and how to scope a run to them.
 
@@ -787,4 +869,5 @@ def register(register_tool) -> None:
     for fn in (run_playbook, cancel_ansible_run, create_schedule, update_schedule, delete_schedule,
                update_group_members):
         register_tool(fn, role=UserRole.operator, read_only=False)
-    register_tool(set_alert_action, role=UserRole.admin, read_only=False)
+    for fn in (set_alert_action, save_playbook, delete_playbook):
+        register_tool(fn, role=UserRole.admin, read_only=False)