Projects as configuration entries (M1.4, D8)

- Config: `projects: {name: entry}` with repo, base_branch, workspace,
  build_host, push credential, labels (defaults added to every issue),
  secrets by ${VAR} reference with visibility (commands | harness), and
  egress (recorded, not enforced). Every problem in every entry is
  reported at once; an inline secret, a bad visibility, a repeated repo or
  a non-slug name is an error.
- The old loop runs exactly one project (only_project) and refuses more.
- The new loop is per project: each its own clone, workspace (with its
  commands-visible secrets, redacted on return), push credential and
  milestones; intake polls every project's repository; items are scoped
  to their project for every run, publish and PR watch. Issues are named
  project#14 when there is more than one project, and a bare #14 that
  matches two is asked about. Switching projects parks the one worked on
  last (all others after a boot).
- A restatement that ends in a question is sent back to the driver, not
  posted: the e2e driver kept asking "Is this the right interpretation?",
  which a plain yes can't answer.

Tests: config entries and errors, two-project assembly, two-project loop
tests (intake, own workspaces and repos, #14 in both, do-next, label
defaults, parking across projects), the property walk over two projects
with a new own_repository_only property (an item using the first
project's repo fails 15 of 16 seeds), and the restatement check.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

nisse-dev/src/nisse_dev/config.py

@@ -17,7 +17,7 @@ silently ignored would look exactly like a knob that has no effect.
 import os
 import re
 from collections.abc import Mapping
-from dataclasses import dataclass, field, fields
+from dataclasses import dataclass, field, fields, replace
 from pathlib import Path
 from typing import Any
 
@@ -30,8 +30,6 @@ DEFAULT_CONFIG_FILE = "nisse-dev.yaml"
 
 REQUIRED_KEYS = (
     "state_db",
-    "workspace",
-    "project",
     "llm.base_url",
     "llm.token",
     "roles.executor",
@@ -45,22 +43,31 @@ REQUIRED_KEYS = (
     "matrix.operators",
     "gitea.url",
     "gitea.token",
-    "gitea.repo",
     "gitea.operators",
-    "build_host.address",
-    "build_host.key",
 )
 OPTIONAL_KEYS = (
     "llm.max_output_tokens",
     "llm.request_timeout_s",
-    "gitea.base_branch",
-    "gitea.push_key",
-    "gitea.push_user",
-    "gitea.push_token",
-    "build_host.user",
-    "build_host.port",
     "loop",
 )
+# A project entry (D8), under `projects: {name: {...}}`.
+PROJECT_REQUIRED = ("repo", "workspace", "build_host.address", "build_host.key")
+PROJECT_OPTIONAL = (
+    "base_branch",
+    "push_key",
+    "push_user",
+    "push_token",
+    "build_host.user",
+    "build_host.port",
+    "labels",
+    "secrets",
+    "egress",
+)
+# Who may see a project secret (D7, D8): the model's commands get it as an
+# environment variable, redacted from what comes back; harness-only ones are
+# never in a command the model chose.
+VISIBILITY = ("commands", "harness")
+_PROJECT_NAME = re.compile(r"[a-z0-9][a-z0-9-]*")
 
 _REFERENCE = re.compile(r"\$\{([A-Za-z_][A-Za-z0-9_]*)\}")
 
@@ -225,6 +232,136 @@ def _read_file(path: Path, errors: list[str]) -> dict[str, Any]:
     return loaded
 
 
+def _parse_projects(
+    raw: Any, env: Mapping[str, str], errors: list[str]
+) -> tuple["Project", ...]:
+    """`projects:`, a mapping of name to entry (D8). Every problem in every
+    entry is reported, like the rest of the file."""
+    if not isinstance(raw, dict) or not raw:
+        errors.append("projects: must name at least one project")
+        return ()
+    projects: list[Project] = []
+    repos: dict[str, str] = {}
+    for name, entry in raw.items():
+        where = f"projects.{name}"
+        if not isinstance(name, str) or not _PROJECT_NAME.fullmatch(name):
+            errors.append(f"{where}: a project name is lowercase letters, digits, -")
+            continue
+        if not isinstance(entry, dict):
+            errors.append(f"{where}: must be a mapping")
+            continue
+        project = _parse_project(name, entry, env, errors)
+        if project is None:
+            continue
+        if project.repo in repos:
+            errors.append(
+                f"{where}.repo: {project.repo} is also project {repos[project.repo]}'s"
+            )
+        repos[project.repo] = name
+        projects.append(project)
+    return tuple(projects)
+
+
+def _parse_project(
+    name: str, entry: Mapping[str, Any], env: Mapping[str, str], errors: list[str]
+) -> "Project | None":
+    where = f"projects.{name}"
+    # labels, secrets and egress are structured; the rest flattens.
+    flat = _flatten({k: v for k, v in entry.items() if k not in ("secrets",)})
+    known = set(PROJECT_REQUIRED) | set(PROJECT_OPTIONAL)
+    for key in sorted(set(flat) - known):
+        errors.append(f"{where}.{key}: unknown setting")
+    values = {
+        key: _resolve(value, f"{where}.{key}", env, errors)
+        for key, value in flat.items()
+        if key in known
+    }
+    missing = [
+        key
+        for key in PROJECT_REQUIRED
+        if values.get(key) is None
+        or (isinstance(values[key], str) and not values[key].strip())
+    ]
+    if missing:
+        errors.append(f"{where}: unset or empty — " + ", ".join(missing))
+        return None
+    port = values.get("build_host.port", 22)
+    try:
+        port = int(port)
+        if not 0 < port < 65536:
+            raise ValueError
+    except (TypeError, ValueError):
+        errors.append(f"{where}.build_host.port: must be a TCP port number")
+    push_user = values.get("push_user")
+    push_token = values.get("push_token")
+    push_key = values.get("push_key")
+    if bool(push_user) != bool(push_token):
+        errors.append(f"{where}: push_user and push_token: set both or neither")
+    if push_token and push_key:
+        errors.append(f"{where}: set a push token or a push key, not both")
+    labels = _strings(values.get("labels", []), f"{where}.labels", errors)
+    egress = _strings(values.get("egress", []), f"{where}.egress", errors)
+    secrets = _parse_secrets(entry.get("secrets"), where, env, errors)
+    return Project(
+        name=name,
+        repo=str(values["repo"]),
+        workspace=Path(values["workspace"]).expanduser(),
+        build_host=BuildHost(
+            address=str(values["build_host.address"]),
+            key=Path(values["build_host.key"]).expanduser(),
+            user=str(values.get("build_host.user") or "root"),
+            port=port if isinstance(port, int) else 22,
+        ),
+        base_branch=str(values.get("base_branch") or "main"),
+        push_key=Path(push_key).expanduser() if push_key else None,
+        push_user=str(push_user) if push_user else None,
+        push_token=str(push_token) if push_token else None,
+        labels=labels,
+        secrets=secrets,
+        egress=egress,
+    )
+
+
+def _strings(raw: Any, where: str, errors: list[str]) -> tuple[str, ...]:
+    if not isinstance(raw, list) or not all(
+        isinstance(x, str) and x.strip() for x in raw
+    ):
+        errors.append(f"{where}: must be a list of names")
+        return ()
+    return tuple(x.strip() for x in raw)
+
+
+def _parse_secrets(
+    raw: Any, where: str, env: Mapping[str, str], errors: list[str]
+) -> dict[str, "Secret"]:
+    """name → {value: ${VAR}, visible: commands | harness}. A value must be
+    a reference: secrets live in secret stores, never in this file."""
+    if raw is None:
+        return {}
+    if not isinstance(raw, dict):
+        errors.append(f"{where}.secrets: must be a mapping of name to secret")
+        return {}
+    secrets: dict[str, Secret] = {}
+    for name, spec in raw.items():
+        at = f"{where}.secrets.{name}"
+        if not isinstance(name, str) or not re.fullmatch(r"[A-Z_][A-Z0-9_]*", name):
+            errors.append(f"{at}: a secret's name is an environment variable name")
+            continue
+        if not isinstance(spec, dict) or set(spec) - {"value", "visible"}:
+            errors.append(f"{at}: give value and visible, nothing else")
+            continue
+        visible = spec.get("visible")
+        if visible not in VISIBILITY:
+            errors.append(f"{at}.visible: must be one of {', '.join(VISIBILITY)}")
+            continue
+        value = spec.get("value")
+        if not isinstance(value, str) or not _REFERENCE.fullmatch(value.strip()):
+            errors.append(f"{at}.value: must be a ${{VAR}} reference, never inline")
+            continue
+        secrets[name] = Secret(_resolve(value.strip(), at, env, errors), visible)
+    return secrets
+
+
 @dataclass(frozen=True)
 class BuildHost:
     """The tool VM the workspace lives on (L13, D3). ``key`` is the path to
@@ -236,11 +373,42 @@ class BuildHost:
     port: int = 22
 
 
+@dataclass(frozen=True)
+class Secret:
+    value: str = field(repr=False)
+    visible: str = "harness"  # one of VISIBILITY
+
+
+@dataclass(frozen=True)
+class Project:
+    """A project entry (D8): what makes a project known to the agent, and
+    everything per-project the rest of the agent refers to."""
+
+    name: str
+    repo: str
+    workspace: Path  # on the build VM
+    build_host: BuildHost
+    base_branch: str = "main"
+    # What may push to the project repository, handed to each push on the
+    # build VM and to nothing else (D4): a push-only account's HTTPS token
+    # (push_user, push_token), or the path to a deploy key (push_key).
+    push_key: Path | None = None
+    push_user: str | None = None
+    push_token: str | None = field(default=None, repr=False)
+    labels: tuple[str, ...] = ()  # added to every issue's own (D1)
+    secrets: Mapping[str, Secret] = field(default_factory=dict)
+    egress: tuple[str, ...] = ()  # beyond the default; recorded, not yet enforced
+
+    def command_secrets(self) -> dict[str, str]:
+        """The secrets the model's commands may see, by name."""
+        return {
+            name: s.value for name, s in self.secrets.items() if s.visible == "commands"
+        }
+
+
 @dataclass(frozen=True)
 class AgentConfig:
     state_db: Path
-    workspace: Path
-    project: str
     llm_base_url: str
     # Credentials never appear in a repr: a failing test or a log line
     # would print them.
@@ -250,16 +418,8 @@ class AgentConfig:
     operators: frozenset[str]
     gitea_url: str
     gitea_token: str = field(repr=False)
-    gitea_repo: str
     gitea_operators: frozenset[str]
-    build_host: BuildHost
-    base_branch: str = "main"
-    # What may push to the project repository, handed to each push on the
-    # build VM and to nothing else (D4): a push-only account's HTTPS token
-    # (push_user, push_token), or the path to a deploy key (push_key).
-    push_key: Path | None = None
-    push_user: str | None = None
-    push_token: str | None = field(default=None, repr=False)
+    projects: tuple[Project, ...]
     max_output_tokens: int = DEFAULT_MAX_OUTPUT_TOKENS
     request_timeout_s: float = DEFAULT_REQUEST_TIMEOUT_S
     tuning: Tuning = field(default_factory=Tuning)
@@ -268,6 +428,55 @@ class AgentConfig:
     loop: str = "old"
     config_file: Path | None = None
 
+    # -- one project, for the old loop ----------------------------------------
+    # control.py (until M2.8) knows one project; these read it, and refuse a
+    # configuration with more.
+
+    def only_project(self) -> Project:
+        if len(self.projects) != 1:
+            names = ", ".join(p.name for p in self.projects)
+            raise ConfigError(
+                f"cannot start: the old loop runs exactly one project, but "
+                f"{len(self.projects)} are configured ({names}); use `loop: new`"
+            )
+        return self.projects[0]
+
+    def with_project(self, **changes: Any) -> "AgentConfig":
+        """This configuration with its one project changed (for tests)."""
+        return replace(self, projects=(replace(self.only_project(), **changes),))
+
+    @property
+    def project(self) -> str:
+        return self.only_project().name
+
+    @property
+    def gitea_repo(self) -> str:
+        return self.only_project().repo
+
+    @property
+    def workspace(self) -> Path:
+        return self.only_project().workspace
+
+    @property
+    def build_host(self) -> BuildHost:
+        return self.only_project().build_host
+
+    @property
+    def base_branch(self) -> str:
+        return self.only_project().base_branch
+
+    @property
+    def push_key(self) -> Path | None:
+        return self.only_project().push_key
+
+    @property
+    def push_user(self) -> str | None:
+        return self.only_project().push_user
+
+    @property
+    def push_token(self) -> str | None:
+        return self.only_project().push_token
+
     @classmethod
     def load(cls, env: Mapping[str, str] = os.environ) -> "AgentConfig":
         """The file named by NISSE_CONFIG, or ./nisse-dev.yaml."""
@@ -304,7 +513,8 @@ class AgentConfig:
         config_file: Path | None,
     ) -> "AgentConfig":
         tuning = _parse_tuning(tree.get("tuning"), errors)
-        flat = _flatten(tree)
+        projects = _parse_projects(tree.get("projects"), env, errors)
+        flat = _flatten({k: v for k, v in tree.items() if k != "projects"})
         flat.pop("tuning", None)
 
         known = set(REQUIRED_KEYS) | set(OPTIONAL_KEYS)
@@ -343,30 +553,14 @@ class AgentConfig:
                 raise ValueError
         except (TypeError, ValueError):
             errors.append("llm.request_timeout_s: must be a positive number of seconds")
-        port = values.get("build_host.port", 22)
-        try:
-            port = int(port)
-            if not 0 < port < 65536:
-                raise ValueError
-        except (TypeError, ValueError):
-            errors.append("build_host.port: must be a TCP port number")
         loop = str(values.get("loop") or "old")
         if loop not in ("old", "new"):
             errors.append("loop: must be old or new")
-        push_user = values.get("gitea.push_user")
-        push_token = values.get("gitea.push_token")
-        if bool(push_user) != bool(push_token):
-            errors.append("gitea.push_user and gitea.push_token: set both or neither")
-        if push_token and values.get("gitea.push_key"):
-            errors.append("gitea: set a push token or a push key, not both")
         if errors:
             raise ConfigError("cannot start: " + "; ".join(errors))
 
-        push_key = values.get("gitea.push_key")
         config = cls(
             state_db=Path(values["state_db"]).expanduser(),
-            workspace=Path(values["workspace"]).expanduser(),
-            project=str(values["project"]),
             llm_base_url=str(values["llm.base_url"]),
             llm_token=str(values["llm.token"]),
             roles=RoleMap(
@@ -384,18 +578,8 @@ class AgentConfig:
             operators=_people(values["matrix.operators"]),
             gitea_url=str(values["gitea.url"]),
             gitea_token=str(values["gitea.token"]),
-            gitea_repo=str(values["gitea.repo"]),
             gitea_operators=_people(values["gitea.operators"]),
-            build_host=BuildHost(
-                address=str(values["build_host.address"]),
-                key=Path(values["build_host.key"]).expanduser(),
-                user=str(values.get("build_host.user") or "root"),
-                port=port,
-            ),
-            push_key=Path(push_key).expanduser() if push_key else None,
-            push_user=str(push_user) if push_user else None,
-            push_token=str(push_token) if push_token else None,
-            base_branch=str(values.get("gitea.base_branch") or "main"),
+            projects=projects,
             max_output_tokens=max_output,
             request_timeout_s=request_timeout,
             tuning=tuning,
@@ -416,31 +600,34 @@ class AgentConfig:
         from and how the agent is deployed, which is not the project's
         business either.
         """
-        workspace = self.workspace.expanduser().resolve()
         candidates = [
             ("state_db", self.state_db),
             ("NISSE_ENV", Path(env.get("NISSE_ENV", ".env"))),
         ]
         if self.config_file is not None:
             candidates.append(("NISSE_CONFIG", self.config_file))
-        offenders = []
-        for label, path in candidates:
-            resolved = path.expanduser().resolve()
-            if resolved == workspace or workspace in resolved.parents:
-                offenders.append(f"{label} ({resolved})")
-        if offenders:
-            raise ConfigError(
-                "refusing to start: these sit inside the workspace clone at "
-                f"{workspace} and would be committed and pushed to the project "
-                "repository — " + ", ".join(offenders)
-            )
+        for project in self.projects:
+            workspace = project.workspace.expanduser().resolve()
+            offenders = []
+            for label, path in candidates:
+                resolved = path.expanduser().resolve()
+                if resolved == workspace or workspace in resolved.parents:
+                    offenders.append(f"{label} ({resolved})")
+            if offenders:
+                raise ConfigError(
+                    "refusing to start: these sit inside the workspace clone at "
+                    f"{workspace} (project {project.name}) and would be "
+                    "committed and pushed to the project repository — "
+                    + ", ".join(offenders)
+                )
 
     def redacted(self) -> dict[str, str]:
         """What the agent may safely announce about how it is configured."""
         return {
-            "project": self.project,
-            "workspace": f"{self.build_host.address}:{self.workspace}",
-            "repo": self.gitea_repo,
+            "projects": ", ".join(
+                f"{p.name} ({p.repo}, {p.build_host.address}:{p.workspace})"
+                for p in self.projects
+            ),
             "planner": self.roles.planner,
             "executor": self.roles.executor,
             "helper": self.roles.helper,