Conversation
Remove upstream CI/release/docs-deploy workflows that depend on Vertex AI, Cohere, PyPI, npm secrets and multi-platform builds we don't need. Replace with a lightweight CI (lint, pure unit tests, package build, slim Docker build validation) and a release workflow that pushes slim images to ghcr.io/franchb on v* tags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
📝 WalkthroughWalkthroughThe pull request consolidates GitHub Actions workflows by introducing a new streamlined CI pipeline in Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (4)
.github/workflows/release.yml (2)
44-52: Redundantlatesttag —latest=auto+type=raw,value=latestboth producelatest-slim.The
flavor: latest=autoalready generates alatest-slimtag for semver-tagged pushes. The explicittype=raw,value=lateston line 52 will also producelatest-slim(due to the suffix). The metadata action deduplicates, so it's not broken, but the raw entry is unnecessary.♻️ Remove redundant raw latest tag
tags: | type=semver,pattern={{version}},value=${{ steps.get_version.outputs.VERSION }} type=semver,pattern={{major}}.{{minor}},value=${{ steps.get_version.outputs.VERSION }} type=semver,pattern={{major}},value=${{ steps.get_version.outputs.VERSION }} - type=raw,value=latest🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.github/workflows/release.yml around lines 44 - 52, The workflow currently uses both flavor: latest=auto (which produces latest-slim due to suffix) and an explicit tags entry type=raw,value=latest that duplicates that same latest-slim tag; remove the redundant raw tag line (the tags entry "type=raw,value=latest") so only flavor: latest=auto generates the latest-slim tag, leaving the semver tag lines and flavor/suffix intact.
54-66: Release builds lack layer caching — each tag push rebuilds from scratch.Without
cache-from/cache-to, every release rebuild pulls and builds all layers. For a slim image this is likely tolerable, but adding GHA cache is low effort and speeds up subsequent builds.♻️ Optional: add GHA build cache
- name: Build and push uses: docker/build-push-action@v6 with: context: . file: docker/standalone/Dockerfile target: ${{ matrix.target }} build-args: | INCLUDE_LOCAL_MODELS=false PRELOAD_ML_MODELS=false push: true platforms: linux/amd64 tags: ${{ steps.meta.outputs.tags }} labels: ${{ steps.meta.outputs.labels }} + cache-from: type=gha + cache-to: type=gha,mode=max🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.github/workflows/release.yml around lines 54 - 66, The Build and push step using docker/build-push-action@v6 has no layer caching configured so each tag rebuilds from scratch; add cache inputs to the step by including cache-from and cache-to entries under the same "with" for the "Build and push" step (the step that uses docker/build-push-action@v6) — e.g. add cache-from: type=gha,ref=${{ github.ref }} (or github.sha) and cache-to: type=gha,mode=max,ref=${{ github.ref }} to enable GitHub Actions layer caching and speed up subsequent matrix.target builds while keeping push, tags and platforms as-is..github/workflows/ci.yml (2)
73-87: Hardcoded test file list will silently skip new tests.Explicitly listing test files means any new pure-unit-test file added under
tests/won't run in CI until someone remembers to update this workflow. If the intent is to run only a known-safe subset (no external deps), consider using a pytest marker (e.g.,@pytest.mark.unit) instead, which is self-documenting and automatically picks up new tests.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.github/workflows/ci.yml around lines 73 - 87, The CI job "Run unit tests" currently hardcodes a file list in the pytest invocation (the "uv run pytest" command), which will silently skip any newly added tests; update the workflow so pytest uses discovery or a marker instead of an explicit file list — either remove the explicit filenames so the command runs `pytest -v --timeout 120 -n auto --dist loadgroup` to discover all tests under tests/, or switch to using a pytest marker (e.g., require `@pytest.mark.unit` on true unit tests and run pytest with `-m unit`) and update the "Run unit tests" step accordingly to ensure new unit tests are picked up automatically.
43-46: Fragile directory navigation with chainedcdcommands.The relative
cd ../navigation couples each line to the previous working directory. If a directory is renamed or reordered, it silently breaks. Consider using absolute paths from$GITHUB_WORKSPACEor separaterunblocks with explicitworking-directory.♻️ Suggested improvement
- name: Install Python dependencies run: | - cd hindsight-api && uv sync --frozen --index-strategy unsafe-best-match - cd ../hindsight-dev && uv sync --frozen --index-strategy unsafe-best-match - cd ../hindsight-embed && uv sync --frozen --index-strategy unsafe-best-match + for dir in hindsight-api hindsight-dev hindsight-embed; do + pushd "$dir" + uv sync --frozen --index-strategy unsafe-best-match + popd + done🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.github/workflows/ci.yml around lines 43 - 46, Replace fragile chained relative cd commands (the three lines that start with "cd hindsight-api && uv sync --frozen --index-strategy unsafe-best-match", "cd ../hindsight-dev && uv sync --frozen --index-strategy unsafe-best-match", and "cd ../hindsight-embed && uv sync --frozen --index-strategy unsafe-best-match") with either: 1) explicit paths using $GITHUB_WORKSPACE (e.g. "$GITHUB_WORKSPACE/hindsight-api" etc.) in the same run block, or 2) split into separate run steps that use the actions/checkout-provided working-directory key pointing at hindsight-api, hindsight-dev, and hindsight-embed respectively; update each step to call "uv sync --frozen --index-strategy unsafe-best-match" without relying on chained cd ../ navigation.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.github/workflows/ci.yml:
- Around line 89-107: The build-api job is missing the UV_INDEX environment
variable needed to resolve torch, so add UV_INDEX with the value
"pytorch=https://download.pytorch.org/whl/cpu" to the build-api job's
environment (apply it at the job level or to the steps that run uv, e.g., before
the "Install uv" or "Build hindsight-api" steps) so that the uv build command in
the Build hindsight-api step can fetch PyTorch; reference the build-api job name
and the UV_INDEX env var when making the change.
---
Nitpick comments:
In @.github/workflows/ci.yml:
- Around line 73-87: The CI job "Run unit tests" currently hardcodes a file list
in the pytest invocation (the "uv run pytest" command), which will silently skip
any newly added tests; update the workflow so pytest uses discovery or a marker
instead of an explicit file list — either remove the explicit filenames so the
command runs `pytest -v --timeout 120 -n auto --dist loadgroup` to discover all
tests under tests/, or switch to using a pytest marker (e.g., require
`@pytest.mark.unit` on true unit tests and run pytest with `-m unit`) and update
the "Run unit tests" step accordingly to ensure new unit tests are picked up
automatically.
- Around line 43-46: Replace fragile chained relative cd commands (the three
lines that start with "cd hindsight-api && uv sync --frozen --index-strategy
unsafe-best-match", "cd ../hindsight-dev && uv sync --frozen --index-strategy
unsafe-best-match", and "cd ../hindsight-embed && uv sync --frozen
--index-strategy unsafe-best-match") with either: 1) explicit paths using
$GITHUB_WORKSPACE (e.g. "$GITHUB_WORKSPACE/hindsight-api" etc.) in the same run
block, or 2) split into separate run steps that use the
actions/checkout-provided working-directory key pointing at hindsight-api,
hindsight-dev, and hindsight-embed respectively; update each step to call "uv
sync --frozen --index-strategy unsafe-best-match" without relying on chained cd
../ navigation.
In @.github/workflows/release.yml:
- Around line 44-52: The workflow currently uses both flavor: latest=auto (which
produces latest-slim due to suffix) and an explicit tags entry
type=raw,value=latest that duplicates that same latest-slim tag; remove the
redundant raw tag line (the tags entry "type=raw,value=latest") so only flavor:
latest=auto generates the latest-slim tag, leaving the semver tag lines and
flavor/suffix intact.
- Around line 54-66: The Build and push step using docker/build-push-action@v6
has no layer caching configured so each tag rebuilds from scratch; add cache
inputs to the step by including cache-from and cache-to entries under the same
"with" for the "Build and push" step (the step that uses
docker/build-push-action@v6) — e.g. add cache-from: type=gha,ref=${{ github.ref
}} (or github.sha) and cache-to: type=gha,mode=max,ref=${{ github.ref }} to
enable GitHub Actions layer caching and speed up subsequent matrix.target builds
while keeping push, tags and platforms as-is.
Comment on lines
+89
to
+107
| build-api: | ||
| runs-on: ubuntu-latest | ||
|
|
||
| steps: | ||
| - uses: actions/checkout@v4 | ||
|
|
||
| - name: Set up Python | ||
| uses: actions/setup-python@v5 | ||
| with: | ||
| python-version: '3.11' | ||
|
|
||
| - name: Install uv | ||
| uses: astral-sh/setup-uv@v5 | ||
| with: | ||
| enable-cache: true | ||
|
|
||
| - name: Build hindsight-api | ||
| working-directory: ./hindsight-api | ||
| run: uv build |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Check if hindsight-api has PyTorch-related dependency
echo "=== Checking hindsight-api/pyproject.toml for PyTorch dependencies ==="
if [ -f "hindsight-api/pyproject.toml" ]; then
cat hindsight-api/pyproject.toml | grep -iE 'torch|pytorch'
else
echo "hindsight-api/pyproject.toml not found"
fiRepository: franchb/hindsight
Length of output: 266
🏁 Script executed:
#!/bin/bash
# Check the entire CI workflow to see UV_INDEX usage in all jobs
echo "=== Checking CI workflow for UV_INDEX environment variable ==="
cat .github/workflows/ci.yml | grep -A 5 -B 5 "UV_INDEX"Repository: franchb/hindsight
Length of output: 527
🏁 Script executed:
#!/bin/bash
# Get the structure of the workflow file around build-api, lint, and unit-tests jobs
echo "=== Checking workflow structure ==="
cat .github/workflows/ci.ymlRepository: franchb/hindsight
Length of output: 3621
Add missing UV_INDEX env var to build-api job
The build-api job is missing the UV_INDEX: pytorch=https://download.pytorch.org/whl/cpu environment variable that both lint and unit-tests jobs define. Since hindsight-api has a torch dependency (torch>=2.6.0), the uv build command will fail to resolve it without the PyTorch index configured.
Proposed fix
build-api:
runs-on: ubuntu-latest
+ env:
+ UV_INDEX: pytorch=https://download.pytorch.org/whl/cpu
steps:
- uses: actions/checkout@v4📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| build-api: | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Set up Python | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: '3.11' | |
| - name: Install uv | |
| uses: astral-sh/setup-uv@v5 | |
| with: | |
| enable-cache: true | |
| - name: Build hindsight-api | |
| working-directory: ./hindsight-api | |
| run: uv build | |
| build-api: | |
| runs-on: ubuntu-latest | |
| env: | |
| UV_INDEX: pytorch=https://download.pytorch.org/whl/cpu | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Set up Python | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: '3.11' | |
| - name: Install uv | |
| uses: astral-sh/setup-uv@v5 | |
| with: | |
| enable-cache: true | |
| - name: Build hindsight-api | |
| working-directory: ./hindsight-api | |
| run: uv build |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.github/workflows/ci.yml around lines 89 - 107, The build-api job is missing
the UV_INDEX environment variable needed to resolve torch, so add UV_INDEX with
the value "pytorch=https://download.pytorch.org/whl/cpu" to the build-api job's
environment (apply it at the job level or to the steps that run uv, e.g., before
the "Install uv" or "Build hindsight-api" steps) so that the uv build command in
the Build hindsight-api step can fetch PyTorch; reference the build-api job name
and the UV_INDEX env var when making the change.
This was referenced Feb 22, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
ci.ymlwith 4 jobs: lint, pure unit tests, package build, slim Docker build validationrelease.ymlthat pushes slim-only images (hindsight-api,hindsight) toghcr.io/franchb/onv*tagsTest plan
v0.0.1-testand verify images appear atghcr.io/franchb/hindsight-api:0.0.1-test-slimandghcr.io/franchb/hindsight:0.0.1-test-slim🤖 Generated with Claude Code
Summary by CodeRabbit