fix(env): pin the infra images' base images by digest, and let Dependabot bump them (#283)

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Author: earakely-scale

.github/dependabot.yml

@@ -31,6 +31,19 @@ updates:
     groups:
       npm-minor-and-patch:
         update-types: ["minor", "patch"]
+  # The infra images the release publishes. Each base is pinned by digest, and a release rebuilds an image only when its
+  # sources change, so a bump here is what picks up a base's fixes: the next release builds and publishes the image.
+  - package-ecosystem: "docker"
+    directories:
+      - "/src/agent_env/env/gateway"
+      - "/src/agent_env/env/envs/service_db"
+      - "/src/agent_env/env/envs/website_browser"
+    schedule:
+      interval: "weekly"
+    open-pull-requests-limit: 5
+    groups:
+      infra-base-images:
+        patterns: ["*"]
   - package-ecosystem: "github-actions"
     directory: "/"
     schedule:

AGENTS.md

@@ -281,7 +281,8 @@ package to PyPI. Run by hand on `main`, the workflow publishes the images alone.
 Its jobs that run agent-env (`.github/scripts/infra_images.py`) hold no token that writes; the jobs that push
 images or publish to PyPI install nothing and check what they're handed (`.github/scripts/infra_build.py`,
 standard library and the docker CLI only).
-Actions are pinned to commit SHAs; Dependabot updates them, the pinned uv, `uv.lock` and the UI's npm packages.
+Actions are pinned to commit SHAs; Dependabot updates them, the pinned uv, `uv.lock`, the UI's npm packages and
+the infra images' base images, each pinned by digest so that a bump rebuilds and republishes the image.
 
 ## Conventions
 

CONTRIBUTING.md

@@ -91,6 +91,8 @@ the previous release, so a clear pull request title is also the release note. A
 A release also publishes the infra images agent-env ships, once the `PUBLISH_INFRA_IMAGES` repository variable is
 `true`: one per Dockerfile under `src/agent_env/env`, at `ghcr.io/scaleapi/agentenv-<name>:src-<hex>` for linux/amd64
 and linux/arm64, where the tag is the build context's source digest, so an unchanged Dockerfile is not built again.
+Each of them pins its base images by digest, and Dependabot bumps those pins, so a base's fixes reach the published
+image: a bump changes the sources, and the next release builds it again.
 Installs pull them instead of building the infra images: a run on local stores registers a kind's images when GHCR has
 every one for the sources installed, by the digest the wheel records or, with no record, the one GHCR serves, and
 builds them otherwise; a sandbox that can't pull one builds it from its build context. The wheel and the release notes

src/agent_env/env/envs/service_db/Dockerfile

@@ -1,4 +1,4 @@
-FROM public.ecr.aws/docker/library/postgres:16-alpine
+FROM public.ecr.aws/docker/library/postgres:16-alpine@sha256:721873c34ceb9f8d8fc265984940dc982404c105f19ad51be9fdc5970a6080ea
 
 # ServiceDBEnv - PostgreSQL database for MCP services
 #

src/agent_env/env/envs/service_db/Dockerfile.db-mcp

@@ -2,7 +2,7 @@
 # The Docker Hub image (crystaldba/postgres-mcp:latest) ships MCP 1.6.0,
 # which only supports stdio and SSE transports.
 
-FROM ghcr.io/astral-sh/uv:python3.12-bookworm-slim AS builder
+FROM ghcr.io/astral-sh/uv:python3.12-bookworm-slim@sha256:e5b65587bce7de595f299855d7385fe7fca39b8a74baa261ba1b7147afa78e58 AS builder
 ENV UV_COMPILE_BYTECODE=1 UV_LINK_MODE=copy
 ENV UV_PYTHON_DOWNLOADS=0
 
@@ -24,7 +24,7 @@ RUN sed -i 's/mcp = FastMCP("postgres-mcp")/from mcp.server.transport_security i
 
 RUN uv sync --frozen --no-dev
 
-FROM public.ecr.aws/docker/library/python:3.12-slim-bookworm
+FROM public.ecr.aws/docker/library/python:3.12-slim-bookworm@sha256:6a0e06f5ba0b373d051aabd3fffb9b873f27f9a2a3a14bcbe37dee6f0e3bbc70
 RUN groupadd -r app && useradd -r -g app app
 COPY --from=builder --chown=app:app /app /app
 

tst/unit/test_release_infra_images.py

@@ -9,6 +9,7 @@
 import pytest
 import yaml
 
+from agent_env.env import bootstrap
 from agent_env.env.published_images import RELEASE_FILE
 
 REPO = Path(__file__).resolve().parents[2]
@@ -135,3 +136,18 @@ def test_what_a_registry_can_rate_limit_is_tried_again():
     assert 'infra_build.py retry docker push "$BUILT"' in build
     assert "infra_build.py retry docker buildx imagetools create" in publish
     assert JOBS["infra-build"]["strategy"]["max-parallel"] <= 4
+
+
+def test_every_infra_image_pins_its_bases_by_digest_and_dependabot_bumps_them():
+    """A release rebuilds an image only when its sources change, so a base behind a bare tag never picks up its
+    fixes; pinned, the pin is a source, and Dependabot bumping it is what rebuilds and republishes the image."""
+    dependabot = yaml.safe_load((REPO / ".github" / "dependabot.yml").read_text())
+    (docker,) = [update for update in dependabot["updates"] if update["package-ecosystem"] == "docker"]
+
+    for image in bootstrap.PUBLISHED_IMAGES:
+        text = image.dockerfile.read_text()
+        stages = set(re.findall(r"^FROM\s+\S+\s+AS\s+(\S+)", text, re.IGNORECASE | re.MULTILINE))
+        bases = [ref for ref in re.findall(r"^FROM\s+(?:--\S+\s+)*(\S+)", text, re.IGNORECASE | re.MULTILINE)
+                 if ref not in stages]
+        assert bases and all(re.search(r"@sha256:[0-9a-f]{64}$", ref) for ref in bases), (image.name, bases)
+        assert f"/{image.dockerfile.parent.relative_to(REPO)}" in docker["directories"], image.name