Skip to content

cherry-pick(release/v1.207): run a package with no project in the default project (#4630) #25223

cherry-pick(release/v1.207): run a package with no project in the default project (#4630)

cherry-pick(release/v1.207): run a package with no project in the default project (#4630) #25223

name: Claude PR Review
on:
pull_request:
types: [opened, ready_for_review]
issue_comment:
types: [created]
pull_request_review_comment:
types: [created]
pull_request_review:
types: [submitted]
# Cancel outdated runs when new commits are pushed
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.event.issue.number }}
cancel-in-progress: true
jobs:
claude-review:
name: Claude Code Review
runs-on: uipath-ubuntu-latest
timeout-minutes: 15
# Only run on non-fork PRs (secrets required), or on @claude mentions from
# trusted authors (OWNER/MEMBER/COLLABORATOR). Untrusted actors must not be
# able to trigger a privileged run via a comment — see Layer 2 in the
# prompt-injection hardening notes.
if: |
(github.event_name == 'pull_request' && !github.event.pull_request.draft && github.event.pull_request.head.repo.full_name == github.repository) ||
(github.event_name == 'issue_comment' && github.event.issue.pull_request && contains(github.event.comment.body, '@claude') && contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.comment.author_association)) ||
(github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@claude') && github.event.pull_request.head.repo.full_name == github.repository && contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.comment.author_association)) ||
(github.event_name == 'pull_request_review' && contains(github.event.review.body, '@claude') && github.event.pull_request.head.repo.full_name == github.repository && contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.review.author_association))
# Least privilege. contents stays read (no push). pull-requests/issues:write
# are required by track_progress to post and update the review comment.
# actions:read dropped — this prompt never reads workflow-run data.
permissions:
contents: read
pull-requests: write
issues: write
steps:
- name: Checkout code
uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
with:
fetch-depth: 0
- name: Claude Code Action
# Pinned to a full commit SHA (not @v1) so a moved tag cannot run
# untrusted action code with our token/secrets. SHA = v1.0.250.
uses: anthropics/claude-code-action@a26b07d8c778b0c43f64618e5fd29fd0f8af72e8 # v1.0.250
env:
AWS_REGION: ${{ vars.AWS_REGION }}
AWS_BEARER_TOKEN_BEDROCK: ${{ secrets.AWS_BEARER_TOKEN_BEDROCK }}
with:
use_bedrock: "true"
github_token: ${{ secrets.GITHUB_TOKEN }}
track_progress: true
# Comment-source allowlist (public repo). Only comments authored by
# these actors are passed to Claude as context; everyone else's
# comments — including external contributors' — are ignored, closing
# the "trusted user triggers @claude on a thread containing a planted
# injection comment" vector. NOTE: this filters COMMENTS only; the PR
# diff, changed-file contents, PR body and title are still ingested,
# so this stacks on the tool allowlist below, it does not replace it.
# Keep this list in sync with the global CODEOWNERS maintainers.
include_comments_by_actor: "gabrielavaduva,AlvinStanescu,marius-bughiu,RaduAna-Maria,uipreliga,dmetzgar,rockymadden,gozhang2,smflorentino,bai-uipath,akshaylive"
# Drop all bot comments (dependabot, renovate, etc.). Exclusion wins
# over the include list if an actor somehow matches both.
exclude_comments_by_actor: "*[bot]"
# Tool allowlist (prompt-injection containment). This review reads
# attacker-controlled content (diff, PR comments, file contents), so
# Claude is restricted to read-only commands plus the single write
# tool track_progress needs to post its review comment. Arbitrary
# Bash, network tools (WebFetch/WebSearch/curl/wget), git push, and
# `gh api`/`gh secret` are NOT in the list and therefore blocked —
# an injected instruction cannot reach env secrets (GITHUB_TOKEN /
# AWS_BEARER_TOKEN_BEDROCK are env vars, not files) or mutate the repo.
# Residual: readable repo content could be echoed into the PR comment.
claude_args: >-
--model us.anthropic.claude-opus-5-5
--allowedTools "Read,Grep,Glob,Bash(git diff:*),Bash(gh pr view:*),Bash(ls:*),Bash(cat:*),mcp__github_comment__update_claude_comment"
prompt: |
You are reviewing a PR for a Claude Code plugin repository that contains
markdown skill definitions, YAML frontmatter, shell hooks, and coder_eval
test tasks. There is NO source code, build system, or runtime — only
documentation that teaches AI agents how to build UiPath automations.
## Review steps
1. Read `.github/code_review.md` for the full review criteria and output format.
2. Read `CLAUDE.md` for project conventions.
3. Read `.claude/rules/skill-review.md` for the skill scoring framework.
4. Run `git diff origin/${{ github.base_ref }}...HEAD` to see the full diff.
5. Run `gh pr view ${{ github.event.pull_request.number || github.event.issue.number }} --comments` to read existing PR conversation — do not repeat resolved issues.
6. For each changed file, read the FULL file (not just the diff) for surrounding context.
## Skill-specific checks
7. If the PR adds a new skill folder under `skills/`:
a. Check that `tests/tasks/<skill-name>/` exists with at least one `.yaml` task file.
If missing, flag as **Medium** (most skills are not yet test-compliant).
b. If tests exist, verify:
- At least one task tagged `smoke` and at least one tagged `e2e` (both required).
- Each task's `tags` uses only valid test types: `smoke`, `integration`, or `e2e`.
- Each task's `tags` includes the skill directory name (e.g., `uipath-maestro-flow`) as the first tag.
- `task_id` matches `skill-<domain>-<capability>`.
- `success_criteria` has at least one check.
If smoke or e2e coverage is missing, flag as **Medium**.
8. Check that `CODEOWNERS` includes an entry for any new skill path.
9. Verify all relative links in changed markdown files resolve to files that
actually exist in the repo (use `ls` or `cat` to confirm).
10. Check for cross-skill references — skills under `skills/` must be fully
self-contained with no imports or links to other skill folders.
11. Verify no secrets, tokens, API keys, or personal filesystem paths appear
in any changed file.
## Output
Follow the output format specified in `.github/code_review.md`.
Always include the "What's Missing" section, even if empty.