full workflow

Author: swghosh

plugins/oape/commands/review.md

@@ -23,7 +23,8 @@ The review covers four key modules:
 
 ## Arguments
 
-- `$1` (ticket_id): The Jira Ticket ID (e.g., OCPBUGS-12345). **Required.**
+- `$1` (ticket_id_or_NO_TICKET): The Jira Ticket ID (e.g., OCPBUGS-12345) **OR** the literal string `NO_TICKET`. **Required.**
+  - When `NO_TICKET` is provided, the review skips Jira validation and focuses on code quality, safety, and build consistency only.
 - `$2` (base_ref): The base git ref to diff against. Defaults to `origin/master`. **Optional.**
 
 
@@ -38,11 +39,16 @@ BASE_REF="${2:-origin/master}"
 ```
 
 ### Step 2: Fetch Context
-1. **Jira Issue**: Fetch the Jira issue details using curl:
-   ```bash
-   curl -s "https://issues.redhat.com/browse/$1"
-   ```
-   Focus on Acceptance Criteria as the primary validation source.
+1. **Jira Issue** (skip if `$1` is `NO_TICKET`):
+   - If `$1` is NOT `NO_TICKET`, fetch the Jira issue details using curl:
+     ```bash
+     curl -s "https://issues.redhat.com/browse/$1"
+     ```
+     Focus on Acceptance Criteria as the primary validation source.
+   - If `$1` IS `NO_TICKET`, skip Jira fetching entirely. The review will focus
+     on code quality (Modules A-D) without validating against Jira acceptance criteria.
+     In the Logic Verification module, skip the "Intent Match" check that compares
+     code against Jira requirements.
 
 2. **Git Diff**: Get the code changes:
    ```bash
@@ -61,7 +67,7 @@ Apply the following review criteria:
 #### Module A: Golang (Logic & Safety)
 
 **Logic Verification (The "Mental Sandbox")**:
-- **Intent Match:** Does the code implementation match the Jira Acceptance Criteria? Quote the Jira line that justifies the change.
+- **Intent Match:** (Skip if `NO_TICKET` mode) Does the code implementation match the Jira Acceptance Criteria? Quote the Jira line that justifies the change. In `NO_TICKET` mode, verify that the code changes are internally consistent and logically correct.
 - **Execution Trace:** Mentally simulate the function.
     - *Happy Path:* Does it succeed as expected?
     - *Error Path:* If the API fails, does it retry or return an error?
@@ -116,7 +122,7 @@ Returns a JSON report with the following structure, followed by an automatic fix
     "simplicity_score": "1-10"
   },
   "logic_verification": {
-    "jira_intent_met": true,
+    "jira_intent_met": true,  // Set to null in NO_TICKET mode
     "missing_edge_cases": ["List handled edge cases or gaps (e.g., 'Does not handle pod deletion')"]
   },
   "issues": [

server/agent.py

@@ -1,10 +1,14 @@
 """
 Core agent execution logic shared between sync and async endpoints.
+
+Supports both single-command execution and full workflow orchestration.
 """
 
 import json
 import logging
+import os
 import traceback
+import uuid
 from collections.abc import Callable
 from dataclasses import dataclass, field
 from pathlib import Path
@@ -22,6 +26,7 @@
 
 # Resolve the plugin directory (repo root) relative to this file.
 PLUGIN_DIR = str(Path(__file__).resolve().parent.parent / "plugins" / "oape")
+TEAM_REPOS_CSV = str(Path(__file__).resolve().parent.parent / "team-repos.csv")
 
 CONVERSATION_LOG = Path("/tmp/conversation.log")
 
@@ -34,10 +39,125 @@
 with open(Path(__file__).resolve().parent / "config.json") as cf:
     CONFIGS = json.loads(cf.read())
 
-# Supported commands and their corresponding plugin skill names.
-SUPPORTED_COMMANDS = {
-    "api-implement": "oape:api-implement",
-}
+# ---------------------------------------------------------------------------
+# Workflow system prompt — instructs a single long-running agent session
+# to execute the full feature development pipeline.
+# ---------------------------------------------------------------------------
+
+WORKFLOW_SYSTEM_PROMPT = """\
+You are an OpenShift operator AI feature development agent. Your job is to
+execute a complete feature development workflow given an Enhancement Proposal
+(EP) PR URL. You MUST follow every phase below **in order**, and you MUST NOT
+skip any phase. Stream your progress clearly so the user can follow along.
+
+## Inputs
+- EP_URL: the GitHub Enhancement Proposal PR URL (provided in the user prompt).
+
+## Allowed repositories
+Read the file {team_repos_csv} to get the list of allowed repositories and
+their base branches (CSV columns: product, role, repo_url, base_branch).
+
+## Phase 0 — Detect target repository
+1. Fetch the EP PR description using `gh pr view <EP_URL> --json body -q .body`
+   (or WebFetch the PR page).
+2. From the EP content, determine which operator repository it targets.
+   Match against the repos in {team_repos_csv}.
+3. Extract: REPO_SHORT_NAME, REPO_URL, BASE_BRANCH.
+   - REPO_SHORT_NAME is the last path component of the repo URL without .git
+     (e.g. "cert-manager-operator").
+4. Print a summary: "Detected repo: <REPO_SHORT_NAME>, base branch: <BASE_BRANCH>".
+
+If you cannot determine the repo, STOP and report the failure.
+
+## Phase 1 — Clone repository
+1. Run `/oape:init <REPO_SHORT_NAME>` in the current working directory.
+2. After init completes, `cd` into the cloned repo directory.
+3. Run `git fetch origin` and `git checkout <BASE_BRANCH>` to ensure you are
+   on the correct base branch.
+4. Extract the EP number from EP_URL (the numeric part after /pull/).
+
+## Phase 2 — PR #1: API Types + Tests
+1. Create and checkout a new branch: `git checkout -b oape/api-types-<EP_NUMBER>`
+2. Run `/oape:api-generate <EP_URL>`.
+3. Identify the generated API directory (typically `api/` or a subdirectory).
+4. Run `/oape:api-generate-tests <path-to-api-directory>`.
+5. Run `make generate && make manifests` (if Makefile targets exist).
+6. Run `make build && make test` to verify the code compiles and tests pass.
+   If build or tests fail, attempt to fix the issues.
+7. Run `/oape:review NO_TICKET origin/<BASE_BRANCH>` to review code quality.
+   The review will auto-fix issues it finds.
+8. Run `make build && make test` again after review fixes.
+9. Stage all changes: `git add -A`
+10. Commit: `git commit -m "oape: generate API types and tests from EP #<EP_NUMBER>"`
+11. Push: `git push -u origin oape/api-types-<EP_NUMBER>`
+12. Create PR #1:
+    ```
+    gh pr create \\
+      --base <BASE_BRANCH> \\
+      --title "oape: API types and tests from EP #<EP_NUMBER>" \\
+      --body "Auto-generated API type definitions and integration tests from <EP_URL>"
+    ```
+13. Save the PR #1 URL.
+
+## Phase 3 — PR #2: Controller Implementation
+1. From the current branch (oape/api-types-<EP_NUMBER>), create a new branch:
+   `git checkout -b oape/controller-<EP_NUMBER>`
+2. Run `/oape:api-implement <EP_URL>`.
+3. Run `make generate && make manifests` (if targets exist).
+4. Run `make build && make test`. Fix failures if any.
+5. Run `/oape:review NO_TICKET origin/<BASE_BRANCH>` to review.
+6. Run `make build && make test` again after review fixes.
+7. Stage, commit: `git commit -m "oape: implement controller from EP #<EP_NUMBER>"`
+8. Push: `git push -u origin oape/controller-<EP_NUMBER>`
+9. Create PR #2 (base = the api-types branch so it stacks):
+    ```
+    gh pr create \\
+      --base oape/api-types-<EP_NUMBER> \\
+      --title "oape: controller implementation from EP #<EP_NUMBER>" \\
+      --body "Auto-generated controller/reconciler code from <EP_URL>"
+    ```
+10. Save the PR #2 URL.
+
+## Phase 4 — PR #3: E2E Tests
+1. Go back to the api-types branch: `git checkout oape/api-types-<EP_NUMBER>`
+2. Create a new branch: `git checkout -b oape/e2e-<EP_NUMBER>`
+3. Run `/oape:e2e-generate <BASE_BRANCH>`.
+4. Run `/oape:review NO_TICKET origin/<BASE_BRANCH>` to review.
+5. Fix any issues, verify build passes.
+6. Stage, commit: `git commit -m "oape: generate e2e tests from EP #<EP_NUMBER>"`
+7. Push: `git push -u origin oape/e2e-<EP_NUMBER>`
+8. Create PR #3 (base = the api-types branch):
+    ```
+    gh pr create \\
+      --base oape/api-types-<EP_NUMBER> \\
+      --title "oape: e2e tests from EP #<EP_NUMBER>" \\
+      --body "Auto-generated e2e test artifacts from <EP_URL>"
+    ```
+9. Save the PR #3 URL.
+
+## Phase 5 — Summary
+Output a final summary in this exact format:
+
+```
+=== OAPE Workflow Complete ===
+
+Enhancement Proposal: <EP_URL>
+Target Repository:    <REPO_SHORT_NAME>
+Base Branch:          <BASE_BRANCH>
+
+PR #1 (API Types + Tests):      <PR_1_URL>
+PR #2 (Controller):             <PR_2_URL>
+PR #3 (E2E Tests):              <PR_3_URL>
+```
+
+## Critical Rules
+- NEVER skip a phase. Execute them in order.
+- If a phase fails and you cannot recover, STOP and report which phase failed and why.
+- Always use `git add -A` before committing to include all generated files.
+- The review command with NO_TICKET skips Jira validation and reviews code quality only.
+- Use `gh pr create` (not manual URL construction) for creating PRs.
+- Do NOT modify the EP or any upstream repository.
+"""
 
 
 @dataclass
@@ -55,37 +175,25 @@ def success(self) -> bool:
 
 
 async def run_agent(
-    command: str,
-    ep_url: str,
+    prompt: str,
     working_dir: str,
+    system_prompt: str,
     on_message: Callable[[dict], None] | None = None,
 ) -> AgentResult:
-    """Run the Claude agent and return the result.
+    """Run the Claude agent with an arbitrary prompt and system prompt.
 
     Args:
-        command: The command key (e.g. "api-implement").
-        ep_url: The enhancement proposal PR URL.
-        working_dir: Absolute path to the operator repo.
+        prompt: The user prompt to send to the agent.
+        working_dir: Absolute path to the working directory.
+        system_prompt: The system prompt for the agent.
         on_message: Optional callback invoked with each conversation message
             dict as it arrives, enabling real-time streaming.
 
     Returns:
         An AgentResult with the output or error.
     """
-    skill_name = SUPPORTED_COMMANDS.get(command)
-    if skill_name is None:
-        return AgentResult(
-            output="",
-            cost_usd=0.0,
-            error=f"Unsupported command: {command}. "
-            f"Supported: {', '.join(SUPPORTED_COMMANDS)}",
-        )
-
     options = ClaudeAgentOptions(
-        system_prompt=(
-            "You are an OpenShift operator code generation assistant. "
-            f"Execute the {skill_name} plugin with the provided EP URL. "
-        ),
+        system_prompt=system_prompt,
         cwd=working_dir,
         permission_mode="bypassPermissions",
         allowed_tools=CONFIGS["claude_allowed_tools"],
@@ -97,7 +205,7 @@ async def run_agent(
     cost_usd = 0.0
 
     conv_logger.info(
-        f"\n{'=' * 60}\n[request] command={command}  ep_url={ep_url}  "
+        f"\n{'=' * 60}\n[request] prompt={prompt[:120]}  "
         f"cwd={working_dir}\n{'=' * 60}"
     )
 
@@ -109,7 +217,7 @@ def _emit(entry: dict) -> None:
 
     try:
         async for message in query(
-            prompt=f"/{skill_name} {ep_url}",
+            prompt=prompt,
             options=options,
         ):
             if isinstance(message, AssistantMessage):
@@ -198,3 +306,39 @@ def _emit(entry: dict) -> None:
             error=str(exc),
             conversation=conversation,
         )
+
+
+async def run_workflow(
+    ep_url: str,
+    on_message: Callable[[dict], None] | None = None,
+) -> AgentResult:
+    """Run the full OAPE feature development workflow.
+
+    Creates a temp directory, then launches a single long-running agent session
+    that executes all phases: init, api-generate, api-generate-tests, review,
+    PR creation, api-implement, e2e-generate, etc.
+
+    Args:
+        ep_url: The enhancement proposal PR URL.
+        on_message: Optional streaming callback.
+
+    Returns:
+        An AgentResult with the final summary or error.
+    """
+    job_id = uuid.uuid4().hex[:12]
+    working_dir = f"/tmp/oape-{job_id}"
+    os.makedirs(working_dir, exist_ok=True)
+
+    system_prompt = WORKFLOW_SYSTEM_PROMPT.format(
+        team_repos_csv=TEAM_REPOS_CSV,
+    )
+
+    return await run_agent(
+        prompt=(
+            f"Execute the full OAPE feature development workflow for this "
+            f"Enhancement Proposal: {ep_url}"
+        ),
+        working_dir=working_dir,
+        system_prompt=system_prompt,
+        on_message=on_message,
+    )