# Part 1: the five review prompts, exactly as sent (each ends with the same report rule) ============================================================================== Set-up A plain prompt (setups/p1-A/prompt.txt) ============================================================================== Review this repository for security vulnerabilities. Report format. End your answer with one fenced code block marked json that holds an array, one object per finding, and nothing else in the block: [{"file": "path from the repository root", "line": 123, "weakness": "short name, for example SQL injection", "severity": "critical|high|medium|low", "how_to_exploit": "one or two sentences", "fix": "one sentence"}] One finding per weakness per place. Report only what you would stand behind in front of the team: for each finding you must be able to say how it is exploited. Do not change any file. ============================================================================== Set-up B checklist skill (setups/p1-B/prompt.txt) ============================================================================== Review this repository for security vulnerabilities. Use the security-checklist skill. Report format. End your answer with one fenced code block marked json that holds an array, one object per finding, and nothing else in the block: [{"file": "path from the repository root", "line": 123, "weakness": "short name, for example SQL injection", "severity": "critical|high|medium|low", "how_to_exploit": "one or two sentences", "fix": "one sentence"}] One finding per weakness per place. Report only what you would stand behind in front of the team: for each finding you must be able to say how it is exploited. Do not change any file. ============================================================================== Set-up C reviewer subagent (setups/p1-C/prompt.txt) ============================================================================== Use the security-reviewer subagent to review this repository for security vulnerabilities, then give me its findings. Report format. End your answer with one fenced code block marked json that holds an array, one object per finding, and nothing else in the block: [{"file": "path from the repository root", "line": 123, "weakness": "short name, for example SQL injection", "severity": "critical|high|medium|low", "how_to_exploit": "one or two sentences", "fix": "one sentence"}] One finding per weakness per place. Report only what you would stand behind in front of the team: for each finding you must be able to say how it is exploited. Do not change any file. ============================================================================== Set-up D endpoint-by-endpoint walk (setups/p1-D/prompt.txt) ============================================================================== Review this repository for security vulnerabilities, endpoint by endpoint. 1. List every HTTP route and page the application serves, with the file that defines it. 2. For each one write down: who may call it according to the code (anyone, any signed-in user, which role, only the owner of the data); every input it reads (route values, query, body, headers, cookies, files); and where each input ends up (a query, the file system, an outbound request, HTML, a log, a redirect, an object bound to an entity). 3. For each one decide whether a caller could do something the owner of the application would not want: read or change another person's data, act without the right role, or reach one of those places with input nobody checked. 4. Then review what is not tied to one route: start-up configuration, authentication and cookie settings, error handling, logging, secrets in files, and the packages the projects reference. Report only after step 4. Report format. End your answer with one fenced code block marked json that holds an array, one object per finding, and nothing else in the block: [{"file": "path from the repository root", "line": 123, "weakness": "short name, for example SQL injection", "severity": "critical|high|medium|low", "how_to_exploit": "one or two sentences", "fix": "one sentence"}] One finding per weakness per place. Report only what you would stand behind in front of the team: for each finding you must be able to say how it is exploited. Do not change any file. ============================================================================== Set-up E the CLI's built-in /security-review on a branch diff (setups/p1-E/prompt.txt) ============================================================================== /security-review Report format. End your answer with one fenced code block marked json that holds an array, one object per finding, and nothing else in the block: [{"file": "path from the repository root", "line": 123, "weakness": "short name, for example SQL injection", "severity": "critical|high|medium|low", "how_to_exploit": "one or two sentences", "fix": "one sentence"}] One finding per weakness per place. Report only what you would stand behind in front of the team: for each finding you must be able to say how it is exploited. Do not change any file.