القائمة
Use when reviewing a pull request for code quality, test coverage, and security. Triggers - "review PR", "review pull request", PR number or link provided, code review request.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when producing product artifacts - epics, user stories, acceptance criteria, or analytics specs.
Use when defining product direction, vision document sections, or roadmap. Triggers on strategic product planning, vision creation, or vision document quality review.
Use when designing UI, choosing colors, typography, spacing, creating visual hierarchy, or making design decisions. Based on Refactoring UI by Wathan & Schoger.
Use when someone needs a plain-language explanation of how an existing feature or edge case behaves in the codebase. Triggers - non-technical stakeholder asks how something works, need code-informed functional explanation, need to understand rules or edge cases without implementation planning.
Use when a developer wants to go from rough idea to working code autonomously. Triggers - imperfect prompt, solo dev wants full pipeline, "just build it", idea to implementation without manual steps.
Use when creating a new agentic workflow skill. Triggers - need new multi-step orchestrated process, want to add workflow to agentic framework, building spec-driven pipeline.
| name | pr-review |
| description | Use when reviewing a pull request for code quality, test coverage, and security. Triggers - "review PR", "review pull request", PR number or link provided, code review request. |
| argument-hint | <PR number | PR URL> |
Reviews a pull request using QA, Test QA, and Security QA agents. Outputs either a local markdown report or GitHub PR comments.
/agentic:workflow:pr-review <input> [<input2> ...]
Arguments (one or more):
123 or #123 — PR number (current repo)https://github.com/owner/repo/pull/123 — full PR URLowner/repo#123 — cross-repo PR referenceBatch mode: Multiple PR refs launch parallel reviews:
/agentic:workflow:pr-review 123 456 789
/agentic:workflow:pr-review #42 owner/repo#99 https://github.com/acme/app/pull/7
You are the main agent orchestrating this workflow. You:
You do NOT:
You MUST delegate agent work using the Task tool. You MUST NOT perform agent work yourself.
When a step says "invoke subagent", you:
If you catch yourself doing agent work, STOP and use Task tool.
Always use {subagentTypeGeneralPurpose} subagent type:
Task(subagent_type="{subagentTypeGeneralPurpose}", prompt="You are the Architect agent. {ide-invoke-prefix}{ide-folder}/agents/agentic-agent-architect.md for your full instructions. ...")
Task(subagent_type="{subagentTypeGeneralPurpose}", prompt="You are the QA agent. {ide-invoke-prefix}{ide-folder}/agents/agentic-agent-qa.md for your full instructions. ...")
Task(subagent_type="{subagentTypeGeneralPurpose}", prompt="You are the Test QA agent. {ide-invoke-prefix}{ide-folder}/agents/agentic-agent-test-qa.md for your full instructions. ...")
Task(subagent_type="{subagentTypeGeneralPurpose}", prompt="You are the Security QA agent. {ide-invoke-prefix}{ide-folder}/agents/agentic-agent-security-qa.md for your full instructions. ...")
Available agents: agentic:agent:architect, agentic:agent:qa, agentic:agent:test-qa, agentic:agent:security-qa
Classify by argument count:
1 PR ref → Single: Classify → Choose Mode → Context → Reviews → Output
N PR refs → Batch: Classify All → Choose Mode (once) → Dispatch N parallel subagents
Execute steps in order. Read step file before executing each step.
| Step | File | Description |
|---|---|---|
| 1 | steps/step-01-classify-input.md | Parse PR reference, verify it exists, extract metadata |
| 2 | steps/step-02-choose-output-mode.md | Ask user: output mode + verbosity level |
| 3 | steps/step-03-gather-context.md | Architect gathers technical context from PR diff |
| 4 | steps/step-04-dispatch-reviews.md | QA + Test QA + Security QA review the PR |
| 5 | steps/step-05-output.md | Aggregate findings, produce output in chosen mode |
Execution pattern:
workflow-state.yaml after each stepgh pr view returned errorAll outputs: {ide-folder}/{outputFolder}/task/pr-review/{pr-identifier}/{instance_id}/
| Artifact | Description |
|---|---|
workflow-state.yaml | Current workflow state |
pr-metadata.md | PR title, description, author, files changed |
pr-diff.patch | Full PR diff |
technical-context.md | Codebase context for affected areas |
qa-review.md | Code quality review |
test-qa-review.md | Test quality review |
security-review.md | Security review |
review-summary.md | Aggregated review (always saved locally) |
/agentic:workflow:pr-review <PR ref>
|
v
+-------------------------------+
| STEP 1: Classify Input |
| Parse PR ref, verify exists |
+-------------------------------+
| |
FOUND NOT FOUND
| |
v v
Continue HALT: PR not found
|
v
+-------------------------------+
| STEP 2: Output Mode & |
| Verbosity Level |
+-------------------------------+
|
v
+-------------------------------+
| STEP 3: Gather Context |
| Architect reads PR diff |
+-------------------------------+
|
v
+-------------------------------+
| STEP 4: Dispatch Reviews |
| QA + Test QA + Security QA |
+-------------------------------+
|
v
+-------------------------------+
| STEP 5: Output |
| Local: md + chat summary |
| PR: gh pr review comments |
+-------------------------------+
|
v
WORKFLOW COMPLETE
/agentic:workflow:pr-review <PR1> <PR2> ... <PRn>
|
v
+-------------------------------+
| STEP 1: Classify All |
| Parse + verify each PR |
| (parallel) |
+-------------------------------+
|
v
+-------------------------------+
| STEP 2: Output Mode & |
| Verbosity Level (ask ONCE) |
+-------------------------------+
|
v
+----------+----------+----------+
| | | |
v v v v
+------+ +------+ +------+ +------+
| PR 1 | | PR 2 | | PR 3 | | PR N |
| S3-5 | | S3-5 | | S3-5 | | S3-5 |
+------+ +------+ +------+ +------+
| | | |
v v v v
+----------+----------+----------+
|
v
+-------------------------------+
| Collect + Summarize |
| Batch summary table in chat |
+-------------------------------+
|
v
WORKFLOW COMPLETE
When multiple PR refs are provided, the orchestrator parallelizes entire reviews:
Classify all inputs (Step 1 for each PR): Run step-01-classify-input.md for each PR ref. These are independent — run in parallel. If any PR fails validation, report the error and continue with the valid ones.
Choose output mode + verbosity once (Step 2): Run step-02-choose-output-mode.md once. The chosen mode and verbosity apply to all PRs.
Dispatch parallel review subagents: For each validated PR, launch one subagent that runs steps 3–5 autonomously:
Task(subagent_type="{subagentTypeGeneralPurpose}", prompt="
You are reviewing PR #{pr_number}: {pr_title}
Output mode: {output_mode}
Verbosity level: {verbosity_level}
Output path: {output_path}
PR diff: {output_path}/pr-diff.patch
PR metadata: {output_path}/pr-metadata.md
Workflow state: {output_path}/workflow-state.yaml
Execute these steps in order:
1. Read and execute {ide-invoke-prefix}{path_to_this_workflow}/steps/step-03-gather-context.md
2. Read and execute {ide-invoke-prefix}{path_to_this_workflow}/steps/step-04-dispatch-reviews.md
3. Read and execute {ide-invoke-prefix}{path_to_this_workflow}/steps/step-05-output.md
You are autonomous. Do not ask the user for input. The output mode is already decided.
")
Launch ALL subagents in parallel (single message with multiple Task calls).
## Batch PR Review Complete
| PR | Verdict | Blockers | Majors | Minors | Nits |
|----|---------|----------|--------|--------|------|
| #123 — {title} | APPROVE | 0 | 0 | 2 | 1 |
| #456 — {title} | REQUEST_CHANGES | 1 | 3 | 0 | 0 |
| #789 — {title} | COMMENT | 0 | 1 | 4 | 2 |
Reports: {output_path_base}/
Each PR gets its own output directory:
{ide-folder}/{outputFolder}/task/pr-review/pr-{N}/{instance_id}/
No shared state file — each PR has its own workflow-state.yaml.
Read steps/step-01-classify-input.md and begin execution.