orca-review

name	orca-review
description	Review Orca security alerts from a JSON export, clone affected repos, and cross-reference with open Renovate PRs to identify fixes.
disable-model-invocation	false
argument-hint	<json-file> --clone-dir <path> [--comment] [--summary]
allowed-tools	Bash, Grep, Glob, Read, Task, TaskCreate, TaskUpdate, TaskList
scope	["security","dependencies"]

Orca Security Issue Review

Review Orca security alerts from a JSON export, analyze each vulnerability, and cross-reference with open Renovate PRs to identify if dependency updates would resolve the issues.

Arguments

• json-file (required): Path to the JSON file containing Orca alerts (e.g., ~/Downloads/alerts.json)
• --clone-dir <path> (required): Base directory where repositories should be cloned for analysis
• --comment (optional): Post comments on matching Renovate PRs with Orca security details and dashboard links. If omitted, PRs are identified but not commented on.
• --summary (optional): Generate a summary.md file in the working directory with an executive summary of findings in markdown format.

Results will be stored in a timestamped subdirectory: <clone-dir>/<YYYYMMDD-HHMMSS>/

Prerequisites

Before running this skill, ensure:

1. GitHub CLI authenticated: Run gh auth login if not already authenticated
2. jq installed: Required for JSON parsing (brew install jq on macOS)
3. Write access: You have write access to the clone directory
4. Repository access: You have read access to the repositories in the alerts

JSON Format

The skill expects a JSON array where each item has:

{
  "Title": "package-name Package Vulnerabilities",
  "Labels": ["fix_available", "shiftleft:vulnerability", ...],
  "CreatedAt": "2026-01-15T13:24:19+00:00",
  "Status": "open",
  "Category": "Vulnerabilities",
  "OrcaScore": 3.2,
  "Source": "./path/to/package-lock.json",
  "LastSeen": "2026-01-22T13:35:27+00:00",
  "AlertId": "orca-10390471",
  "CloudAccount": {"Name": "Org/repo (Project: Org/repo)"},
  "Inventory": {"Name": "Org/repo"}

The AlertId field is used to generate dashboard links: https://eu.app.orcasecurity.io/alerts/{AlertId}

Process

1. Setup Working Environment

Initialize variables from arguments and create the working directory structure.

1.1 Parse Arguments

Extract the JSON file path, clone directory, and optional flags from $ARGUMENTS:

# Parse arguments - expected format: <json-file> --clone-dir <path> [--comment] [--summary]

# Check for --comment flag
if [[ "$ARGUMENTS" == *"--comment"* ]]; then
  SHOULD_COMMENT=true
  ARGUMENTS="${ARGUMENTS//--comment/}"
else
  SHOULD_COMMENT=false
fi

# Check for --summary flag
if [[ "$ARGUMENTS" == *"--summary"* ]]; then
  SHOULD_SUMMARY=true
  ARGUMENTS="${ARGUMENTS//--summary/}"
else
  SHOULD_SUMMARY=false
fi

# Extract JSON file and clone dir
JSON_FILE="${ARGUMENTS%% --clone-dir*}"
CLONE_DIR="${ARGUMENTS##*--clone-dir }"

# Trim whitespace
JSON_FILE=$(echo "$JSON_FILE" | xargs)
CLONE_DIR=$(echo "$CLONE_DIR" | xargs)

# Expand paths (handle ~ and relative paths)
JSON_FILE=$(eval echo "$JSON_FILE")
CLONE_DIR=$(eval echo "$CLONE_DIR")

# Validate inputs
if [ ! -f "$JSON_FILE" ]; 
   
   1


 [ -z  ]; 
   
   1


 
 

1.2 Create Timestamped Working Directory

Create a timestamped subdirectory for this execution:

TIMESTAMP=$(date +%Y%m%d-%H%M%S)
WORK_DIR="${CLONE_DIR}/${TIMESTAMP}"
mkdir -p "$WORK_DIR"

echo "Working directory: $WORK_DIR"

2. Parse and Chunk the JSON File

The input file can be large (multiple MB). Check the file size and process accordingly.

2.1 Check File Size

FILE_SIZE=$(stat -f%z "$JSON_FILE" 2>/dev/null || stat -c%s "$JSON_FILE" 2>/dev/null)
FILE_SIZE_MB=$((FILE_SIZE / 1024 / 1024))

echo "File size: ${FILE_SIZE_MB}MB"

# Threshold: process in chunks if larger than 2MB
CHUNK_THRESHOLD=2

2.2 Process Based on Size

If file is small (< 2MB):

if [ "$FILE_SIZE_MB" -lt "$CHUNK_THRESHOLD" ]; then
  # Load entire file and parse to JSONL (one alert per line)
  cat "$JSON_FILE" | jq -c '.[]' > "$WORK_DIR/alerts.jsonl"
  echo "Parsed $(wc -l < "$WORK_DIR/alerts.jsonl") alerts"
fi

If file is large (>= 2MB):

if [ "$FILE_SIZE_MB" -ge "$CHUNK_THRESHOLD" ]; then
  # Initialize counters
  CHUNK_NUM=0
  LINES=0

  # Stream parse and chunk into 100-item batches
  cat "$JSON_FILE" | jq -c '.[]' | while IFS= read -r line; do
    echo "$line" >> "$WORK_DIR/alerts-chunk-${CHUNK_NUM}.jsonl"
    LINES=$((LINES + 1))
    if [ $LINES -ge 100 ]; then
      CHUNK_NUM=$((CHUNK_NUM + 1))
      LINES=0
    fi
  done

  echo "Created $((CHUNK_NUM + 1)) chunk files"
fi

This creates JSONL files (one JSON object per line, compact format) that can be processed sequentially without loading the entire file into memory.

2.3 Group Alerts by Repository

Extract unique repositories from the alerts:

# For small files or after chunking is complete
if [ -f "$WORK_DIR/alerts.jsonl" ]; then
  jq -r '.Inventory.Name' < "$WORK_DIR/alerts.jsonl" | sort -u > "$WORK_DIR/repos.txt"
else
  # Combine all chunks
  cat "$WORK_DIR"/alerts-chunk-*.jsonl | jq -r '.Inventory.Name' | sort -u > "$WORK_DIR/repos.txt"
fi

echo "Found $(wc -l < "$WORK_DIR/repos.txt") unique repositories"

3. Process Each Repository

For each unique repository in the alerts:

3.1 Clone the Repository

Clone repositories into the timestamped directory, preserving org structure:

while read -r REPO; do
  ORG=$(dirname "$REPO")
  REPO_NAME=$(basename "$REPO")

  mkdir -p "$WORK_DIR/$ORG"

  if gh repo clone "$REPO" "$WORK_DIR/$ORG/$REPO_NAME" -- --depth=1 2>/dev/null; then
    echo "Cloned: $REPO"
  else
    echo "Warning: Failed to clone $REPO (may lack access or repo deleted)"
  fi
done < "$WORK_DIR/repos.txt"

3.2 List Open Renovate PRs

For each repository, fetch open Renovate PRs and cache them:

while read -r REPO; do
  PR_FILE="$WORK_DIR/renovate-prs-$(echo "$REPO" | tr '/' '-').json"

  gh pr list --repo "$REPO" --author "renovate[bot]" --state open \
    --json number,title,url > "$PR_FILE" 2>/dev/null || echo "[]" > "$PR_FILE"

  PR_COUNT=$(jq 'length' < "$PR_FILE")
  echo "$REPO: $PR_COUNT open Renovate PRs"
done < "$WORK_DIR/repos.txt"

4. Analyze Each Alert

For each alert item, perform the following:

4.1 Extract Package Information

From each alert, extract the package name and alert ID:

# Example: "@remix-run/router Package Vulnerabilities" → "@remix-run/router"
PACKAGE_NAME=$(echo "$TITLE" | sed 's/ Package Vulnerabilities$//')

# Extract AlertId for dashboard link
ALERT_ID=$(echo "$ALERT" | jq -r '.AlertId')

4.2 Assess Severity

Use the OrcaScore to determine severity level:

Score Range	Severity	Priority
7.0 - 10.0	Critical	P1 - Immediate action
5.0 - 6.9	High	P2 - Address this sprint
3.0 - 4.9	Medium	P3 - Plan for next sprint
0.0 - 2.9	Low	P4 - Backlog

4.3 Check for Fix Availability

If Labels array contains "fix_available":

• A newer version of the package exists that addresses the vulnerability
• Check if Renovate has created a PR for this package

4.4 Cross-Reference with Renovate PRs

Search the cached Renovate PRs for the affected package:

PR_FILE="$WORK_DIR/renovate-prs-$(echo "$REPO" | tr '/' '-').json"

# Search for matching PR (case-sensitive, full package name)
MATCHING_PR=$(jq -r --arg pkg "$PACKAGE_NAME" \
  '.[] | select(.title | contains($pkg)) | @json' < "$PR_FILE" | head -1)

if [ -n "$MATCHING_PR" ]; then
  PR_NUMBER=$(echo "$MATCHING_PR" | jq -r '.number')
  PR_TITLE=$(echo "$MATCHING_PR" | jq -r '.title')
  PR_URL=$(echo "$MATCHING_PR" | jq -r '.url')
  echo "Found matching PR: #$PR_NUMBER - $PR_TITLE"
fi

5. Comment on Matching PRs (if --comment flag provided)

When a Renovate PR is found that updates a vulnerable package, optionally add a comment linking the security context and Orca dashboard.

Only post comments if the --comment flag was included in the arguments.

If --comment is provided:

• Build the Orca dashboard link from AlertId: https://eu.app.orcasecurity.io/alerts/$ALERT_ID
• Read the comment template from resources/pr-comment-template.md
• Fill in the variables ($PACKAGE_NAME, $ORCA_SCORE, $SEVERITY, $CREATED_AT, $ORCA_LINK, $SOURCE_PATHS, $LABELS)
• Post the filled template as a comment on the PR using gh pr comment

If --comment is NOT provided, skip commenting and only track the match for the summary report.

Note: When multiple alerts match the same PR (same package in different lock files), consolidate into a single comment listing all affected sources.

6. Create Tasks for PR Review

For each Renovate PR that matches a vulnerable package, create a Claude Task for follow-up review using the /renovate-review skill:

{
  "subject": "Review and validate Renovate PR for security fix",
  "description": "Review Renovate PR updating <package-name> which addresses Orca vulnerability (Score: <score>)\n\nRepository: <org>/<repo>\nPR: #<pr-number> - <pr-title>\nURL: <pr-url>\n\nVulnerability Details:\n- Package: <package-name>\n- Orca Score: <score> (<severity>)\n- Affected Files: <sources>\n- Fix Available: Yes\n- Alert Created: <created-at>\n- Orca Dashboard: https://eu.app.orcasecurity.io/alerts/<alert-id>\n\nAction: Review PR, validate dependency changes, and merge if tests pass.",
  "activeForm": "Reviewing Renovate security fix PR"
}

These tasks are collected for the summary output and can be processed by the /renovate-review skill.

7. Generate Summary Report

After processing all alerts, output a summary to the terminal. Use the structure defined in resources/summary-template.md — it includes sections for executive summary, severity distribution, PR coverage, alerts without PRs, and recommended actions. Fill in all variables with the computed values.

8. Save Results to Output File

Save the structured results to a JSON file in the working directory:

OUTPUT_FILE="$WORK_DIR/orca-review-results.json"

The output file structure is defined in resources/output-example.json. It includes execution metadata, tasks for PR review, and summary statistics by severity.

9. Generate Summary File (if --summary flag provided)

When the --summary flag is provided, write the executive summary to a markdown file.

Only generate the file if the --summary flag was included in the arguments.

If --summary is provided, read the template from resources/summary-template.md, fill in all variables with computed values, and write to $WORK_DIR/summary.md.

If --summary is NOT provided, skip file generation and only display the summary in the terminal.

What Happens Next

After the skill completes:

1. Review the summary displayed in the terminal (or read summary.md if --summary was used)
2. Check the output file at $WORK_DIR/orca-review-results.json for structured data (see resources/output-example.json)
3. Process the tasks using one of these approaches:
   • Run /renovate-review <pr-url> manually for each PR
   • Use the task list to work through PRs systematically
4. Monitor unmatched alerts for future Renovate PRs

The skill does not automatically merge PRs or proceed to the next phase. You control when and how to act on the findings.

Examples:

# Basic run (no comments, no summary file)
/orca-review ~/Downloads/alerts.json --clone-dir ~/orca-workspace

# With PR comments
/orca-review ~/Downloads/alerts.json --clone-dir ~/orca-workspace --comment

# With summary file
/orca-review ~/Downloads/alerts.json --clone-dir ~/orca-workspace --summary

# Full run with comments and summary
/orca-review ~/Downloads/alerts.json --clone-dir ~/orca-workspace --comment --summary

Directory Structure

Results are organized by execution timestamp for audit trails:

<clone-dir>/
├── 20260129-143022/                    # First execution
│   ├── NebulaTerra/
│   │   └── modular-store/              # Cloned repo
│   ├── OtherOrg/
│   │   └── another-repo/               # Cloned repo
│   ├── alerts.jsonl                    # Parsed alerts (small files)
│   ├── alerts-chunk-0.jsonl            # Chunked alerts (large files)
│   ├── repos.txt                       # Unique repository list
│   ├── renovate-prs-NebulaTerra-modular-store.json
│   ├── orca-review-results.json        # Final output (JSON)
│   └── summary.md                      # Executive summary (if --summary used)
├── 20260129-145030/                    # Second execution
│   └── ...

Each timestamped directory is independent, allowing you to:

• Compare results across multiple runs
• Track vulnerability remediation progress over time
• Reference specific execution results for audit purposes
• Run multiple analyses in parallel without conflicts

Guidelines

Deduplication

• Multiple alerts may exist for the same package across different lock files in the same repo
• Group these together and only create one task per PR
• In the task description and PR comment, list all affected source files
• Count duplicate issues only once in summary statistics

PR Matching

• Match package names case-sensitively
• For scoped packages (e.g., @remix-run/router), search for the full scoped name
• A PR title like "Update dependency @remix-run/router to v6.22.0" matches @remix-run/router
• Use fuzzy matching for PR titles if exact match fails (e.g., version numbers may vary)

Task Generation

When generating tasks for the Claude Task system:

• Each task represents ONE Renovate PR that fixes one or more Orca vulnerabilities
• Include the PR URL, repository, and all affected packages/issues
• Set severity based on the highest OrcaScore among matched issues
• Tasks are designed for the /renovate-review skill to handle PR review and merging
• Output the complete task list in structured JSON format for easy consumption

Summary Statistics

Calculate coverage metrics:

Issues with PRs = count of unique (repo, package) pairs with matching Renovate PR
Issues without PRs = count of unique (repo, package) pairs without matching PR
Coverage % = (Issues with PRs / Total Issues) * 100

Segment by severity to show coverage at each level.

Large File Handling

The JSON export can be quite large. The skill automatically handles files efficiently:

• File size check: Determines processing strategy based on file size
• Small files (< 2MB): Loaded entirely with jq
• Large files (>= 2MB): Streamed and chunked into 100-alert batches
• JSONL format: Intermediate files use compact newline-delimited JSON (jq -c)
• Sequential processing: Chunks are processed one at a time to avoid memory spikes
• Temporary files: All intermediate files are stored in $WORK_DIR

Error Handling

• If a repository cannot be cloned (permissions, deleted, etc.), log the error and continue
• If gh CLI is not authenticated, prompt user to run gh auth login
• If JSON file is malformed, report the specific parsing error and exit gracefully
• Skip duplicate alerts for the same package in the same repo (log as skipped)
• If jq fails during chunking, report the error with the chunk number and continue with next chunk

Rate Limiting

When processing many repositories, be mindful of GitHub API rate limits:

• The gh CLI handles rate limiting automatically with backoff
• For 50+ repositories: expect slower processing as rate limits are hit
• For 100+ repositories: consider splitting the JSON file and running in batches
• Recommended batch size: 50 repositories per execution
• Check rate limit status: gh api rate_limit

Labels to Watch For

• fix_available: A patched version exists (prioritize creating tasks for these)
• easy_exploitation: Vulnerability is easily exploitable - mark as high priority
• shiftleft:vulnerability: Indicates a ShiftLeft/Orca detected vulnerability
• Use labels to highlight high-risk vulnerabilities in task descriptions