Merge branch 'main' into fix/orphan-agent-lifecycle

README.md

@@ -8,7 +8,7 @@
 > | 能力 | 状态 |
 > |---|---|
 > | 任务分解（规划回退）、能力路由派发、专家执行（OpenAI 兼容）、评审/重做循环、结果汇总、peer 协作消息路由、WebSocket Agent 协议、Redis 持久化、Prometheus 指标 | ✅ 已实现（仓内可运行） |
-> | Manager↔Runtime 生命周期契约（对齐 `heicode-am-contract`）、HMAC 签名回调事件 envelope、稳定的 `deployment_id`/`workflow_id`/`trace_id` | 🟡 待接入（见 `docs/integration/runtime-contract.md`；事件 envelope 与签名待 Manager 对齐） |
+> | Manager↔Runtime 生命周期契约（对齐 `heicode-am-contract`）、HMAC 签名回调事件 envelope、稳定的 `deployment_id`/`workflow_id`/`trace_id` | 🟡 机制已实现，待主链路联调验收（生命周期接口见 `docs/integration/runtime-contract.md §3`；envelope/HMAC 签名机制已实现且规范与 HM 对齐，见 `docs/integration/event-schema.md §1`，待 Manager 端验签端到端联调验收） |
 > | 统一 usage/计费聚合（归属 NewAPI）、审计/lineage trace、前端事件 API、统一 Agent registry/scheduling、统一 secret/workspace/tool/MCP/tenant 安全边界 | 🟡 待接入（需与 Billing / Audit / Frontend / Infra / Security Team 对齐） |
 > | Benchmark 自证（`Benchmark_Agent`、`S_swarm`、`G_E`、`G_E,c`、治理/协作/通信/鲁棒性指标、baseline 对比、telemetry 架构） | 🟡 采集器已落地（`benchmark/`：指标公式 `metrics.py`、活体 run 采集 `collectors/`、4 基线 runner `runners/`、`baselines.compare` 算 `G_E`/`G_E,c`、自证合流 `selfcert_collector`、telemetry 导出 Cosmos/Blob `export/`）；**评分标准待定**——`O`（可观测性）公式、`BASE_COEFFICIENTS`、`S_gain` 阈值需累积真实用户使用数据后经验标定（长期工单）。诚实原则：缺真实输入的量返回 `NaN`、不伪造分值（标准见 `docs/benchmark/`） |
 >

docs/integration/event-schema.md

@@ -14,7 +14,7 @@
 
 - Swarm → HM：`POST {callback.url}`（create 时下发）。
 - 鉴权二选一：服务令牌 `X-Agent-Service-Token`（或 `Authorization: Bearer`），或 HMAC 签名。
-- **HMAC 签名（已实现，与 HM 一致）**：
+- **HMAC 签名（机制已实现，规范与 HM 对齐；待主链路端到端联调验收）**：
   - 头：`X-Agent-Timestamp`（Unix 毫秒）、`X-Agent-Signature`、`X-Agent-Event-Id`、`X-Correlation-ID`（均含 `X-Agnet-` 兼容别名）。
   - 规范串：`canonical = f"{timestamp}.{event_id}.{raw_body}"`。
   - 签名：`X-Agent-Signature = "sha256=" + hex(HMAC_SHA256(secret, canonical))`。

docs/integration/runtime-contract.md

@@ -85,10 +85,14 @@ Swarm 拉起每个专家 agent 时注入以下进程环境变量（`agent/main.p
 | `AGENT_ID` | agent 实例 id（唯一） | **Swarm launcher** |
 | `AGENT_CAPABILITIES` | 能力集合（逗号分隔） | **Swarm launcher**（`AGENT_LAUNCH_CAPABILITIES` 池策略） |
 | `HEICODE_USER_ID` | 发起用户；用于每用户并发上限 `MAX_AGENTS_PER_USER`（注册时强制） | **Swarm 从 create `x-user-id` 透传** |
-| `WORKSPACE_DIR` / `GIT_REPO_URL` | 可选：工作区 / 代码仓 | 资源授权 |
+| `WORKSPACE_DIR` | 可选：工作区根 | 部署（`WORKSPACE_DIR`） |
+| `GIT_REPO_URL` | 代码仓克隆地址；agent 有此值才 clone | **Swarm 从 git `resource_grant.metadata.repo_url` 解析** |
+| `GIT_USERNAME` / `GIT_PASSWORD` | git HTTPS 凭据（注入 clone URL） | **Swarm 从 git `resource_grant.secret_ref`（`azkv://`）解析** |
+| `GIT_BASE_BRANCH` | 可选：基线分支（默认 `main`） | git grant `metadata.base_branch` |
 
 **约束**：
 - 模型 key 由 **Swarm 服务端从 `billing_context.secret_ref`（`azkv://`）解析**（`resolve_model_key`：override → azkv（部署 SecretResolver / dev `HEICODE_SECRET_<name>`）→ 编排器 `OPENAI_API_KEY` 兜底）后注入被拉起 agent 的 env。key **不入** create 请求体 / 回调 / 日志 / argv（满足 §3.1）；解析不到则 agent keyless 启动并明确报错（不伪造）。`billing_context` 的 `newapi_user_ref`/`quota_ref` 仍仅为归因元数据。见 [usage-billing-schema.md §2](./usage-billing-schema.md)。
+- **git 仓库绑定**（agent_swarm#63 / HM #92）由 **Swarm 服务端从 `resource_grants` 里 `resource_type=git` 的 grant 解析**（`resolve_git_grant`）：仓库地址 = `grant.metadata.repo_url`（非密文，inline 注入 `GIT_REPO_URL`）；git 凭据 = `grant.secret_ref`（`azkv://`，同模型 key 路径解析 → 注入 `GIT_USERNAME`/`GIT_PASSWORD`）。git 凭据**不入** create 请求体 / 回调 / 日志 / argv；k8s 后端 `GIT_PASSWORD` 经 per-swarm Secret 的 `secretKeyRef` 注入（绝不内联 PodSpec）。grant 有 repo 无凭据时仍注入 `GIT_REPO_URL`（公有仓可 clone；私有仓 clone 报错，不伪造）。git KV secret 值约定为 JSON `{"git_username","git_password"}`（接受 `git_token`/`token` 形式 + 裸 token），**待 HM #92 对齐**。同一用户 agent 池注入同一 git grant。
 - 同一用户的专家 agent 注入**同一把** `sk-` + `OPENAI_API_BASE=HM/v1`，计费归一到发起用户；`task_executor` 随模型请求带 `X-Agent-*` 归因头供 HM/NewAPI 关联。token 吊销走 §3.3.1 A.5 的 `swarm.pool_terminated` 握手（per-user，全部 run stop 后吊销）。
 - **运行时事件前置**：只有 (i) HM 真把 create 派发到编排器（`SWARM_RUNTIME_ENABLED=true`，非 manager-local 适配器）且 (ii) Swarm 已按上表拉起 agent 并连入，编排器才回推 `task.*`/`swarm.*` 运行时事件（否则 events feed 仅有 HM 控制面 `deployment.status_changed`，见 #39）。
 

docs/integration/security-boundary.md

@@ -60,7 +60,8 @@
 - 服务间鉴权：HM→Swarm 用 `AGENT_RUNTIME_SERVICE_TOKEN`（Bearer）；回调 HMAC 签名。
 - **每用户并发 Agent 配额**：一个 `user_id` 同时连接的 Agent 数上限为 `MAX_AGENTS_PER_USER`（env，默认 10）。注册（WS `register` 消息携带 `user_id`）超额即被拒绝（回 `registration_rejected` 并关闭，code 1008），断开后释放名额。归因主轴仍为 `user.id`/`channelId`。未带 `user_id` 的 Agent 为 unbound，不计入该配额。实现：`ConnectionManager.can_bind_user/bind_user/unbind` + 注册处强制；测试 `scripts/test-max-agents-per-user.py`。
 - **Swarm 拉起 agent + 服务端解析 key（team 决议，runtime-contract §3.3）**：由 **Swarm 运行时**（`orchestrator/agent_launcher.py`）拉起专家 agent 池（拉起数 `min(池大小, MAX_AGENTS_PER_USER − 已连)`，与上面的注册兜底一致）。模型 key 由 **Swarm 从 `billing_context.secret_ref`（`azkv://`）服务端解析**后注入被拉起 agent——**不入** create 请求体 / 回调 / 日志 / argv。azkv 真实解析为部署侧 SecretResolver；dev/CI 用 `HEICODE_SECRET_<name>`。解析不到即 keyless 启动并明确报错（不伪造）。
-- **K8s pod 边界（`AGENT_LAUNCH_BACKEND=kubernetes`，生产）**：**每 agent 一个 Pod**，带 CPU/内存 requests+limits、标签（`heicode-swarm-id`/`heicode-user-id`）。模型 key 经**每-swarm k8s Secret**（manifest via stdin 应用）由 Pod `secretKeyRef` 引用，**绝不内联进 PodSpec env**（否则暴露于 etcd / `kubectl get pod -o yaml`）。编排器需 `kubectl` + 一个仅对 `AGENT_POD_NAMESPACE` 有 pod/secret 权限的 **ServiceAccount（最小 RBAC）**；Pod 出网由 NetworkPolicy 限定到 HM `/v1` + git；停止/删除按标签 `kubectl delete pod,secret`。**更硬化（推荐 Infra 评估）**：azkv **CSI SecretProviderClass** 让 Pod 直接从 Key Vault 挂载密钥，编排器**全程不接触明文**。代码执行沙箱仍按 §8.1 在 Pod 层强制。
+- **Git 仓库绑定凭据服务端解析（agent_swarm#63 / HM #92）**：git 绑定经 `resource_grants` 里 `resource_type=git` 的 grant 下发。Swarm 同模型 key 方式**服务端解析**（`resolve_git_grant`）：`metadata.repo_url`（非密文）→ `GIT_REPO_URL`；`secret_ref`（`azkv://` git 凭据引用）→ `GIT_USERNAME`/`GIT_PASSWORD`。git 凭据**不入** create 请求体 / 回调 / 日志 / argv；明文 git token/密码不得进 create 体（`_reject_plaintext_secrets` 已覆盖 `resource_grants`）。git KV secret 值约定 JSON `{"git_username","git_password"}`（接受 token 形式），**待 HM #92 对齐**。
+- **K8s pod 边界（`AGENT_LAUNCH_BACKEND=kubernetes`，生产）**：**每 agent 一个 Pod**，带 CPU/内存 requests+limits、标签（`heicode-swarm-id`/`heicode-user-id`）。模型 key 与 **git 凭据（`GIT_PASSWORD`）** 均经**每-swarm k8s Secret**（manifest via stdin 应用）由 Pod `secretKeyRef` 引用，**绝不内联进 PodSpec env**（否则暴露于 etcd / `kubectl get pod -o yaml`）；`GIT_REPO_URL`/`GIT_USERNAME` 为非密文，inline 注入。编排器需 `kubectl` + 一个仅对 `AGENT_POD_NAMESPACE` 有 pod/secret 权限的 **ServiceAccount（最小 RBAC）**；Pod 出网由 NetworkPolicy 限定到 HM `/v1` + git；停止/删除按标签 `kubectl delete pod,secret`。**更硬化（推荐 Infra 评估）**：azkv **CSI SecretProviderClass** 让 Pod 直接从 Key Vault 挂载密钥，编排器**全程不接触明文**。代码执行沙箱仍按 §8.1 在 Pod 层强制。
 - 🟡 待接入：多租户运行时隔离（命名空间/网络/配额）由 Agent 平台（AKS Workload Identity）承载，非本仓编排器；归因主轴为 `user.id`/`channelId`（见 `usage-billing-schema.md`），不引入 tenant 概念。
 
 ## 7. 外部 API 与传输

orchestrator/agent_launcher.py

@@ -25,6 +25,14 @@ Secret handling: the model key (per-user ``sk-``) is **resolved here from
 — it never travels in the create request body (runtime-contract §3.1) or in events/logs. Real
 Azure Key Vault resolution is a deployment adapter (`SECRET_RESOLVER`); dev/CI use an env map.
 
+Git binding (agent_swarm#63 / HM #92): a swarm-run needs a repo to operate on. HM ships the git
+binding as a ``resource_type=git`` entry in ``resource_grants`` — ``metadata.repo_url`` (clone URL,
+non-secret) + ``secret_ref`` (``azkv://`` ref to the git credential, never plaintext). The launcher
+resolves it here the same server-side way as the model key (``resolve_git_grant``) and injects
+``GIT_REPO_URL`` (+ ``GIT_USERNAME``/``GIT_PASSWORD`` when a credential resolves) into the agent env;
+the agent (`agent/main.py` + `agent/git_operations.py`) clones from those vars. Without this the
+agent has no repo URL → cannot clone → the "需绑定 git 仓库才能使用" symptom in HM #92.
+
 Everything is **fail-soft**: a launch error is logged and swallowed so run creation never fails on
 it. Pure planning (`plan_launch_specs`) is separated from side-effecting launch for testability.
 """
@@ -192,13 +200,14 @@ def _parse_azkv_ref(secret_ref: str):
     return vault_url, name, version
 
 
-def _resolve_from_keyvault(secret_ref: str) -> Optional[str]:
-    """Read the model key from Azure Key Vault using the pod's workload identity.
+def _read_kv_secret_value(secret_ref: str) -> Optional[str]:
+    """Read the **raw** secret value string from Azure Key Vault using the pod's workload identity.
 
-    `DefaultAzureCredential` picks up the federated token the AKS workload-identity webhook injects
-    (see `_azkv_enabled`). Lazy-imports the azure SDK so dev/CI without it are unaffected. Returns
-    None on ANY failure (not enabled / unparseable ref / SDK missing / no credential / network /
-    secret absent) — never fabricates, never raises.
+    Shared by the model-key and git-credential resolvers (each applies its own extractor to the raw
+    value). `DefaultAzureCredential` picks up the federated token the AKS workload-identity webhook
+    injects (see `_azkv_enabled`). Lazy-imports the azure SDK so dev/CI without it are unaffected.
+    Returns None on ANY failure (not enabled / unparseable ref / SDK missing / no credential /
+    network / secret absent) — never fabricates, never raises.
     """
     if not _azkv_enabled():
         return None
@@ -216,12 +225,114 @@ def _resolve_from_keyvault(secret_ref: str) -> Optional[str]:
     try:
         client = SecretClient(vault_url=vault_url, credential=DefaultAzureCredential())
         secret = client.get_secret(name, version) if version else client.get_secret(name)
-        return _extract_model_key(secret.value)
+        return secret.value
     except Exception as exc:  # no credential / RBAC / network / missing secret
         logger.warning("azkv resolver: failed to read secret '%s' from %s: %s", name, vault_url, exc)
         return None
 
 
+def _resolve_from_keyvault(secret_ref: str) -> Optional[str]:
+    """Read the model key from Azure Key Vault (raw value → `_extract_model_key`). None on failure."""
+    raw = _read_kv_secret_value(secret_ref)
+    return _extract_model_key(raw) if raw is not None else None
+
+
+# ── Git binding resolution (agent_swarm#63 / HM #92) ───────────────────────────
+#: Default git username when the KV secret carries only a token/PAT (GitHub HTTPS convention: any
+#: non-empty username + PAT-as-password works; ``x-access-token`` is the documented placeholder).
+DEFAULT_GIT_USERNAME = "x-access-token"
+
+
+def _extract_git_credentials(secret_value: str) -> Optional[Dict[str, str]]:
+    """Pull git credentials out of a resolved KV secret value.
+
+    Contract (mirrors the model-key JSON convention; pending HM #92 alignment): the git KV secret is
+    JSON ``{"git_username": "...", "git_password": "..."}``. Accepted aliases:
+    ``username``/``password``, and a token form ``git_token``/``token`` (username defaults to
+    ``x-access-token``). A bare string value is treated as a token. Returns
+    ``{"username","password"}`` or None (never fabricates — missing password → None).
+    """
+    value = (secret_value or "").strip()
+    if not value:
+        return None
+    if value.startswith("{"):
+        try:
+            obj = json.loads(value)
+        except ValueError:
+            return None
+        if not isinstance(obj, dict):
+            return None
+        username = (obj.get("git_username") or obj.get("username") or "").strip() or DEFAULT_GIT_USERNAME
+        password = (obj.get("git_password") or obj.get("git_token")
+                    or obj.get("password") or obj.get("token") or "").strip()
+        return {"username": username, "password": password} if password else None
+    return {"username": DEFAULT_GIT_USERNAME, "password": value}  # bare token string
+
+
+def _resolve_git_secret_ref(secret_ref: str) -> Optional[Dict[str, str]]:
+    """Resolve an azkv:// git ``secret_ref`` to ``{"username","password"}``.
+
+    Same source order as the model key: dev/CI env map ``HEICODE_SECRET_<name>`` (hermetic) →
+    production Azure Key Vault via workload identity. None when unavailable — never fabricated.
+    """
+    name = secret_ref.rstrip("/").rsplit("/", 1)[-1]
+    raw = os.getenv(f"HEICODE_SECRET_{name}")
+    if raw is None:
+        raw = _read_kv_secret_value(secret_ref)
+    return _extract_git_credentials(raw) if raw is not None else None
+
+
+def _iter_resource_grants(body: Dict[str, Any]):
+    """Yield every resource grant in the create body (top-level + per-agent), in declaration order."""
+    yield from (body.get("resource_grants") or [])
+    plan = body.get("orchestration_plan") or {}
+    for agent in (plan.get("agents") or []):
+        yield from (agent.get("resource_grants") or [])
+    for agent in (body.get("agents") or []):
+        yield from (agent.get("resource_grants") or [])
+
+
+def _first_git_grant(body: Dict[str, Any]) -> Optional[Dict[str, Any]]:
+    """First grant with ``resource_type`` (or ``type``) == ``git``, or None."""
+    for grant in _iter_resource_grants(body):
+        if not isinstance(grant, dict):
+            continue
+        if (grant.get("resource_type") or grant.get("type")) == "git":
+            return grant
+    return None
+
+
+def resolve_git_grant(body: Dict[str, Any]) -> Optional[Dict[str, str]]:
+    """Resolve the git binding for launched agents — server-side, mirroring `resolve_model_key`.
+
+    Returns the git env to inject (``GIT_REPO_URL`` + optional ``GIT_USERNAME``/``GIT_PASSWORD`` +
+    optional ``GIT_BASE_BRANCH``) or None when there is no git grant / no repo URL. The credential
+    (``secret_ref`` → azkv) is resolved here and never travels in the create body / events / logs.
+    A grant with a repo URL but unresolved credential still yields ``GIT_REPO_URL`` (public-repo
+    clone works; a private repo then fails to clone with a clear git error — never fabricated).
+    """
+    grant = _first_git_grant(body)
+    if not grant:
+        return None
+    meta = grant.get("metadata") or {}
+    repo_url = (meta.get("repo_url") or grant.get("repo_url") or "").strip()
+    if not repo_url:
+        return None
+    env: Dict[str, str] = {"GIT_REPO_URL": repo_url}
+    branch = (meta.get("base_branch") or meta.get("branch") or "").strip()
+    if branch:
+        env["GIT_BASE_BRANCH"] = branch
+    secret_ref = (grant.get("secret_ref") or grant.get("ref") or "").strip()
+    if secret_ref.startswith("azkv://"):
+        creds = _resolve_git_secret_ref(secret_ref)
+        if creds:
+            env["GIT_USERNAME"] = creds["username"]
+            env["GIT_PASSWORD"] = creds["password"]
+        else:
+            logger.warning("git grant: secret_ref present but unresolved; agent will clone keyless (private repo will fail)")
+    return env
+
+
 def model_api_base() -> str:
     return os.getenv("AGENT_OPENAI_API_BASE") or os.getenv("OPENAI_API_BASE") or "https://api.openai.com/v1"
 
@@ -421,6 +532,11 @@ async def _launch_command(spec: AgentLaunchSpec) -> None:
 # (orchestrator never touches plaintext) — deployment/Infra option, see security-boundary §6.
 _k8s_swarms: set = set()
 
+#: Env keys that carry secrets — delivered via the per-swarm k8s Secret (`secretKeyRef`), NEVER
+#: inlined into the PodSpec (would leak into etcd / `kubectl get pod -o yaml`). `GIT_USERNAME` and
+#: `GIT_REPO_URL` are non-secret and stay inline.
+SENSITIVE_ENV_KEYS = ("OPENAI_API_KEY", "GIT_PASSWORD")
+
 
 def pod_image() -> str:
     return os.getenv("AGENT_POD_IMAGE", "heicode/swarm-agent:latest")
@@ -461,25 +577,30 @@ def _k8s_labels(spec: AgentLaunchSpec, swarm_id: str) -> Dict[str, str]:
     return labels
 
 
-def build_secret_manifest(*, key_value: str, namespace: str, name: str, labels: Dict[str, str]) -> Dict[str, Any]:
-    """k8s Secret carrying the per-user model key (applied via stdin → value never on argv)."""
+def build_secret_manifest(*, namespace: str, name: str, labels: Dict[str, str],
+                          secret_data: Dict[str, str]) -> Dict[str, Any]:
+    """k8s Secret carrying the per-user sensitive env (model key + git password) — applied via stdin
+    so values never hit argv. `secret_data` maps env key → value (only SENSITIVE_ENV_KEYS)."""
     return {
         "apiVersion": "v1", "kind": "Secret", "type": "Opaque",
         "metadata": {"name": name, "namespace": namespace, "labels": labels},
-        "stringData": {"OPENAI_API_KEY": key_value},
+        "stringData": dict(secret_data),
     }
 
 
 def build_pod_manifest(spec: AgentLaunchSpec, *, namespace: str, swarm_id: str,
                        image: Optional[str] = None, secret_name: Optional[str] = None,
-                       service_account: Optional[str] = None,
+                       secret_keys=None, service_account: Optional[str] = None,
                        resources: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
-    """One agent Pod. Non-sensitive env inline; `OPENAI_API_KEY` via `secretKeyRef` (never inline)."""
+    """One agent Pod. Non-sensitive env inline; secret env (`OPENAI_API_KEY`/`GIT_PASSWORD`) via
+    `secretKeyRef` (never inline). `secret_keys` = the keys actually present in the Secret."""
+    keys_in_secret = set(secret_keys or ())
     env_list: List[Dict[str, Any]] = []
     for k, v in spec.env.items():
-        if k == "OPENAI_API_KEY":
-            if secret_name:  # reference the Secret; never inline the key into the PodSpec
-                env_list.append({"name": k, "valueFrom": {"secretKeyRef": {"name": secret_name, "key": "OPENAI_API_KEY"}}})
+        if k in SENSITIVE_ENV_KEYS:
+            if secret_name and k in keys_in_secret:  # reference the Secret; never inline into PodSpec
+                env_list.append({"name": k, "valueFrom": {"secretKeyRef": {"name": secret_name, "key": k}}})
+            # else: omit entirely — a secret value is never inlined
         else:
             env_list.append({"name": k, "value": v})
     pod_spec: Dict[str, Any] = {
@@ -521,19 +642,28 @@ async def _launch_kubernetes_pool(specs: List[AgentLaunchSpec], swarm_id: str) -
     ns, sa, res, img = pod_namespace(), pod_service_account(), pod_resources(), pod_image()
     labels = _k8s_labels(specs[0], swarm_id)
     secret_name = None
-    key_value = next((s.env.get("OPENAI_API_KEY") for s in specs if s.env.get("OPENAI_API_KEY")), None)
-    if key_value:
+    # Sensitive env (model key + git password) is per-user/per-run — same across the pool; collect
+    # the first non-empty value of each into one per-swarm Secret.
+    secret_data: Dict[str, str] = {}
+    for skey in SENSITIVE_ENV_KEYS:
+        val = next((s.env.get(skey) for s in specs if s.env.get(skey)), None)
+        if val:
+            secret_data[skey] = val
+    if secret_data:
         secret_name = f"swarm-agent-key-{swarm_id}"
         try:
-            await _kubectl_apply(build_secret_manifest(key_value=key_value, namespace=ns, name=secret_name, labels=labels))
-        except Exception as exc:  # without the secret the pods can't get the key; still try keyless (clear error)
+            await _kubectl_apply(build_secret_manifest(namespace=ns, name=secret_name, labels=labels, secret_data=secret_data))
+        except Exception as exc:  # without the secret the pods can't get the values; still try without (clear error)
             logger.warning("k8s secret apply failed for %s: %s", swarm_id, exc)
             secret_name = None
+            secret_data = {}
+    secret_keys = set(secret_data)
     launched: List[str] = []
     for spec in specs:
         try:
             await _kubectl_apply(build_pod_manifest(spec, namespace=ns, swarm_id=swarm_id, image=img,
-                                                    secret_name=secret_name, service_account=sa, resources=res))
+                                                    secret_name=secret_name, secret_keys=secret_keys,
+                                                    service_account=sa, resources=res))
             launched.append(spec.agent_id)
         except Exception as exc:
             logger.warning("k8s agent pod apply failed for %s: %s", spec.agent_id, exc)

orchestrator/main.py

@@ -1767,10 +1767,12 @@ async def launch_swarm_agents(run, body: Dict[str, Any]) -> None:
     connected = manager.user_agent_count(user_id) if user_id else 0
     backend = agent_launcher.launch_backend()
     model_key = agent_launcher.resolve_model_key(body)
-    git_env = agent_launcher.git_launch_env(body)
+    git_env = agent_launcher.resolve_git_grant(body)
     info: Dict[str, Any] = {
         "backend": backend, "planned": 0, "launched": 0, "launched_ids": [],
-        "model_key_resolved": bool(model_key), "repo_bound": bool(git_env.get("GIT_REPO_URL")),
+        "model_key_resolved": bool(model_key),
+        "git_repo_resolved": bool(git_env and git_env.get("GIT_REPO_URL")),
+        "git_creds_resolved": bool(git_env and git_env.get("GIT_PASSWORD")),
         "note": "",
     }
     try:

scripts/test-agent-launcher.py

@@ -91,55 +91,60 @@ def test_pool_window():
 def test_plan_specs():
     run = FakeRun()
     body = {"orchestration_plan": {"objective": "x"}, "billing_context": {"default_model_id": "gpt-x"}}
-    # 8 connected, cap 10 -> headroom 2, but MIN floor 3 wins (intentional [3,16] hard-floor-first).
+    git_env = {"GIT_REPO_URL": "https://git.example/r.git", "GIT_USERNAME": "x-access-token",
+               "GIT_PASSWORD": "ghp_xyz"}
     specs = al.plan_launch_specs(run, body, connected_user_agents=8, limit=10, pool_size=3,
-                                 model_key="sk-test", orchestrator_url="ws://orch:8000", user_id="u-1")
-    check("plan applies MIN floor over cap headroom (8 connected, cap 10 -> 3 not 2)", len(specs) == 3)
+                                 model_key="sk-test", orchestrator_url="ws://orch:8000", user_id="u-1",
+                                 git_env=git_env)
+    check("plan caps at limit (8 connected, cap 10 -> launch 2)", len(specs) == 2)
     s = specs[0]
     check("spec env has ORCHESTRATOR_URL", s.env.get("ORCHESTRATOR_URL") == "ws://orch:8000")
     check("spec env has model key (server-side injected)", s.env.get("OPENAI_API_KEY") == "sk-test")
     check("spec env has model id from billing_context", s.env.get("OPENAI_MODEL") == "gpt-x")
     check("spec env has HEICODE_USER_ID for per-user cap", s.env.get("HEICODE_USER_ID") == "u-1")
+    check("spec env has GIT_REPO_URL (server-side injected)", s.env.get("GIT_REPO_URL") == "https://git.example/r.git")
+    check("spec env has GIT_USERNAME + GIT_PASSWORD", s.env.get("GIT_USERNAME") == "x-access-token" and s.env.get("GIT_PASSWORD") == "ghp_xyz")
     check("spec has AGENT_ID + capabilities", bool(s.agent_id) and bool(s.env.get("AGENT_CAPABILITIES")))
     check("agent ids unique", len({sp.agent_id for sp in specs}) == len(specs))
-    # No key -> OPENAI_API_KEY omitted (not fabricated), no user -> HEICODE_USER_ID omitted.
+    # No key -> OPENAI_API_KEY omitted (not fabricated), no user -> HEICODE_USER_ID omitted, no git -> GIT_* omitted.
     specs2 = al.plan_launch_specs(run, body, connected_user_agents=0, limit=10, pool_size=1,
-                                  model_key=None, orchestrator_url="ws://orch", user_id=None)
+                                  model_key=None, orchestrator_url="ws://orch", user_id=None, git_env=None)
     check("no model key -> OPENAI_API_KEY omitted", "OPENAI_API_KEY" not in specs2[0].env)
     check("no user -> HEICODE_USER_ID omitted", "HEICODE_USER_ID" not in specs2[0].env)
-    # git_env (GIT_REPO_URL) merged into every agent's env so the agent clones the bound repo.
-    specs3 = al.plan_launch_specs(run, body, connected_user_agents=0, limit=10, pool_size=3,
-                                  model_key="sk-test", orchestrator_url="ws://orch", user_id="u-1",
-                                  git_env={"GIT_REPO_URL": "https://github.com/Fasthei/swe", "GIT_PROVIDER": "github"})
-    check("git_env merged into every spec (GIT_REPO_URL)",
-          all(sp.env.get("GIT_REPO_URL") == "https://github.com/Fasthei/swe" for sp in specs3))
-    check("git_env GIT_PROVIDER propagated", specs3[0].env.get("GIT_PROVIDER") == "github")
-    check("no git_env -> GIT_REPO_URL omitted", "GIT_REPO_URL" not in specs2[0].env)
+    check("no git grant -> GIT_REPO_URL omitted", "GIT_REPO_URL" not in specs2[0].env)
 
 
-def test_git_launch_env():
-    # Public repo bound via resource grant (metadata.repo_url) -> GIT_REPO_URL injected, no creds.
-    body = {"resource_grants": [{
-        "resource_type": "git", "binding_scope": "https://github.com/Fasthei/swe",
-        "metadata": {"repo_url": "https://github.com/Fasthei/swe", "provider": "github", "default_branch": "main"},
-        "ref": "azkv://heicode-vault.vault.azure.net/secrets/users-3-bindings-x",  # present but NOT resolved (public)
-    }]}
-    env = al.git_launch_env(body)
-    check("git_launch_env: GIT_REPO_URL from grant metadata.repo_url", env.get("GIT_REPO_URL") == "https://github.com/Fasthei/swe")
-    check("git_launch_env: GIT_PROVIDER from metadata", env.get("GIT_PROVIDER") == "github")
-    check("git_launch_env: GIT_DEFAULT_BRANCH + GIT_BASE_BRANCH from metadata",
-          env.get("GIT_DEFAULT_BRANCH") == "main" and env.get("GIT_BASE_BRANCH") == "main")
-    check("git_launch_env: public repo injects NO credentials (GIT_TOKEN absent)", "GIT_TOKEN" not in env)
-    # grant found via orchestration_plan.resource_grants too
-    body_plan = {"orchestration_plan": {"resource_grants": [{"resource_type": "git",
-                 "metadata": {"repo_url": "https://github.com/a/b"}}]}}
-    check("git_launch_env: reads plan.resource_grants", al.git_launch_env(body_plan).get("GIT_REPO_URL") == "https://github.com/a/b")
-    # git resource via binding_scope (no metadata.repo_url) when resource_type == git
-    body_scope = {"resource_grants": [{"resource_type": "git", "binding_scope": "https://gitea.x/o/r"}]}
-    check("git_launch_env: falls back to binding_scope for git type", al.git_launch_env(body_scope).get("GIT_REPO_URL") == "https://gitea.x/o/r")
-    # no git grant -> empty (a non-git grant is ignored)
-    check("git_launch_env: no git grant -> {}", al.git_launch_env({"resource_grants": [{"resource_type": "database", "metadata": {"host": "h"}}]}) == {})
-    check("git_launch_env: empty body -> {}", al.git_launch_env({}) == {})
+def test_resolve_git_grant():
+    # ── credential extraction (pure) ──
+    check("git creds: JSON git_username/git_password",
+          al._extract_git_credentials('{"git_username":"u","git_password":"p"}') == {"username": "u", "password": "p"})
+    check("git creds: token form -> default username",
+          al._extract_git_credentials('{"git_token":"ghp_x"}') == {"username": al.DEFAULT_GIT_USERNAME, "password": "ghp_x"})
+    check("git creds: bare string -> token",
+          al._extract_git_credentials("ghp_bare") == {"username": al.DEFAULT_GIT_USERNAME, "password": "ghp_bare"})
+    check("git creds: missing password -> None (never fabricated)", al._extract_git_credentials('{"git_username":"u"}') is None)
+    check("git creds: empty -> None", al._extract_git_credentials("") is None)
+
+    # ── grant lookup + resolution (dev env map, hermetic) ──
+    os.environ.pop("HEICODE_SECRET_res_git_1", None)
+    grant_inline = {"resource_type": "git", "secret_ref": "azkv://heicode-vault/secrets/res_git_1",
+                    "metadata": {"repo_url": "https://git.example/acme/app.git", "base_branch": "develop"}}
+    body = {"resource_grants": [grant_inline]}
+    check("no secret in env -> repo url still resolved (public-repo path)",
+          al.resolve_git_grant(body) == {"GIT_REPO_URL": "https://git.example/acme/app.git", "GIT_BASE_BRANCH": "develop"})
+    os.environ["HEICODE_SECRET_res_git_1"] = '{"git_username":"bot","git_password":"ghp_kv"}'
+    out = al.resolve_git_grant(body)
+    check("secret_ref resolved via dev map -> creds injected",
+          out.get("GIT_USERNAME") == "bot" and out.get("GIT_PASSWORD") == "ghp_kv"
+          and out.get("GIT_REPO_URL") == "https://git.example/acme/app.git")
+    os.environ.pop("HEICODE_SECRET_res_git_1")
+    check("no git grant -> None", al.resolve_git_grant({"resource_grants": [{"resource_type": "newapi"}]}) is None)
+    check("git grant without repo_url -> None",
+          al.resolve_git_grant({"resource_grants": [{"resource_type": "git", "secret_ref": "azkv://v/secrets/x"}]}) is None)
+    # per-agent grant (orchestration_plan.agents[].resource_grants) also discovered
+    body_nested = {"orchestration_plan": {"agents": [{"resource_grants": [grant_inline]}]}}
+    check("per-agent git grant discovered",
+          (al.resolve_git_grant(body_nested) or {}).get("GIT_REPO_URL") == "https://git.example/acme/app.git")
 
 
 def test_resolve_model_key():
@@ -216,17 +221,22 @@ def test_k8s_manifests():
         agent_id="swarm-abc-agent-1", capabilities="python,general",
         env={"ORCHESTRATOR_URL": "ws://orch.svc:8000", "AGENT_ID": "swarm-abc-agent-1",
              "AGENT_CAPABILITIES": "python,general", "OPENAI_API_KEY": "sk-secret-xyz",
-             "OPENAI_API_BASE": "https://hm/v1", "HEICODE_USER_ID": "u-1"},
+             "OPENAI_API_BASE": "https://hm/v1", "HEICODE_USER_ID": "u-1",
+             "GIT_REPO_URL": "https://git.example/acme/app.git", "GIT_USERNAME": "x-access-token",
+             "GIT_PASSWORD": "ghp_secret_pw"},
     )
-    # Secret manifest carries the key (applied via stdin, not argv).
-    sec = al.build_secret_manifest(key_value="sk-secret-xyz", namespace="heicode-swarm",
-                                   name="swarm-agent-key-swarm-abc", labels={"app": "heicode-swarm-agent"})
+    # Secret manifest carries BOTH sensitive values (model key + git password), applied via stdin.
+    sec = al.build_secret_manifest(namespace="heicode-swarm", name="swarm-agent-key-swarm-abc",
+                                   labels={"app": "heicode-swarm-agent"},
+                                   secret_data={"OPENAI_API_KEY": "sk-secret-xyz", "GIT_PASSWORD": "ghp_secret_pw"})
     check("secret kind/type", sec["kind"] == "Secret" and sec["type"] == "Opaque")
-    check("secret stringData has the key", sec["stringData"]["OPENAI_API_KEY"] == "sk-secret-xyz")
+    check("secret stringData has model key + git password",
+          sec["stringData"]["OPENAI_API_KEY"] == "sk-secret-xyz" and sec["stringData"]["GIT_PASSWORD"] == "ghp_secret_pw")
 
-    # Pod manifest references the key via secretKeyRef — NEVER inline.
+    # Pod manifest references secret env via secretKeyRef — NEVER inline.
     pod = al.build_pod_manifest(spec, namespace="heicode-swarm", swarm_id="swarm-abc",
                                 image="img:1", secret_name="swarm-agent-key-swarm-abc",
+                                secret_keys={"OPENAI_API_KEY", "GIT_PASSWORD"},
                                 service_account="swarm-agent-sa")
     blob = json.dumps(pod)
     check("pod kind/name", pod["kind"] == "Pod" and pod["metadata"]["name"] == "swarm-abc-agent-1")
@@ -235,17 +245,25 @@ def test_k8s_manifests():
           and pod["metadata"]["labels"].get("heicode-user-id") == "u-1")
     env = {e["name"]: e for e in pod["spec"]["containers"][0]["env"]}
     check("OPENAI_API_KEY via secretKeyRef (not inline)",
-          "value" not in env["OPENAI_API_KEY"] and env["OPENAI_API_KEY"]["valueFrom"]["secretKeyRef"]["name"] == "swarm-agent-key-swarm-abc")
-    check("raw key value NOT in pod manifest (only in Secret)", "sk-secret-xyz" not in blob)
-    check("non-secret env inline", env["ORCHESTRATOR_URL"]["value"] == "ws://orch.svc:8000")
+          "value" not in env["OPENAI_API_KEY"] and env["OPENAI_API_KEY"]["valueFrom"]["secretKeyRef"]["key"] == "OPENAI_API_KEY")
+    check("GIT_PASSWORD via secretKeyRef (not inline)",
+          "value" not in env["GIT_PASSWORD"] and env["GIT_PASSWORD"]["valueFrom"]["secretKeyRef"]["key"] == "GIT_PASSWORD")
+    check("raw secret values NOT in pod manifest (only in Secret)",
+          "sk-secret-xyz" not in blob and "ghp_secret_pw" not in blob)
+    check("non-secret env inline (orch url + git repo url + git username)",
+          env["ORCHESTRATOR_URL"]["value"] == "ws://orch.svc:8000"
+          and env["GIT_REPO_URL"]["value"] == "https://git.example/acme/app.git"
+          and env["GIT_USERNAME"]["value"] == "x-access-token")
     check("pod has resource limits + restartPolicy + serviceAccount",
           pod["spec"]["containers"][0]["resources"]["limits"]
           and pod["spec"]["restartPolicy"] == "OnFailure"
           and pod["spec"]["serviceAccountName"] == "swarm-agent-sa")
-    # No secret_name -> key omitted entirely (never inlined).
+    # No secret_name -> secret env omitted entirely (never inlined); non-secret git env still inline.
     pod2 = al.build_pod_manifest(spec, namespace="heicode-swarm", swarm_id="swarm-abc", secret_name=None)
     env2 = {e["name"] for e in pod2["spec"]["containers"][0]["env"]}
-    check("no secret -> OPENAI_API_KEY omitted from pod env", "OPENAI_API_KEY" not in env2)
+    check("no secret -> OPENAI_API_KEY + GIT_PASSWORD omitted from pod env",
+          "OPENAI_API_KEY" not in env2 and "GIT_PASSWORD" not in env2)
+    check("no secret -> non-secret GIT_REPO_URL/GIT_USERNAME still present", "GIT_REPO_URL" in env2 and "GIT_USERNAME" in env2)
 
 
 def main():
@@ -255,6 +273,7 @@ def main():
     test_plan_specs()
     test_git_launch_env()
     test_resolve_model_key()
+    test_resolve_git_grant()
     test_azkv_resolver()
     test_command_backend_build()
     test_k8s_manifests()