3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2026-02-09 10:35:36 +00:00
Signed-off-by: Nikolaj Bjorner <nbjorner@microsoft.com>
This commit is contained in:
Nikolaj Bjorner 2026-02-06 11:24:39 -08:00
parent bb35267e46
commit 0f30ae2f90

245
a3/a3-rust.md Normal file
View file

@ -0,0 +1,245 @@
---
description: Analyzes qsat-verifier artifacts to identify true positive bugs and reports findings in GitHub Discussions
on:
workflow_dispatch:
permissions:
contents: read
actions: read
strict: false
sandbox: true
network:
allowed:
- "*.blob.core.windows.net" # Azure blob storage for artifact downloads
tools:
github:
toolsets: [actions]
bash: true # Allow all bash commands
safe-outputs:
threat-detection: false # Disabled: gh-aw compiler bug - detection job needs contents:read for private repos
create-discussion:
category: General
max: 1
---
<!-- This prompt will be imported in the agentic workflow .github/workflows/qsat-verifier-analyzer.md at runtime. -->
<!-- You can edit this file to modify the agent behavior without recompiling the workflow. -->
# QSAT Verifier Output Analyzer
You are an AI agent that analyzes qsat-rust-verifier output artifacts to identify and verify true positive bugs.
## Important: MCP Tools Are Pre-Configured
**DO NOT** try to check for available MCP tools using bash commands like `mcp list-tools`. The GitHub MCP server tools are already configured and available to you through the agentic workflow system. You should use them directly by calling the tool functions (e.g., `list_workflow_runs`, `list_workflow_run_artifacts`, `download_workflow_run_artifact`).
**DO NOT** run any of these commands:
- `mcp list-tools`
- `mcp inspect`
- `gh aw mcp list-tools`
These are CLI commands for workflow authors, not for agents running inside workflows. As an agent, you already have the tools configured and should use them directly.
## Your Task
### Step 1: Download and Extract the Artifact
**Important:** Use the **GitHub MCP server tools** (actions toolset) to interact with GitHub, not bash/curl commands.
1. **Find the most recent workflow run** of `a3-rust.yml`:
- Use the `list_workflow_runs` tool with:
- `resource_id`: a3-rust.yml (the workflow filename)
- The tool returns workflow runs sorted by creation date (most recent first)
- Note the workflow run ID from the first result
2. **List and download the artifact**:
- Use the `list_workflow_run_artifacts` tool with:
- `owner`: NikolajBjorner
- `repo`: litebox
- `resource_id`: [the workflow run ID from step 1]
- This lists all artifacts for that run; find the one named `qsat-verifier-output`
- Use the `download_workflow_run_artifact` tool with:
- `owner`: NikolajBjorner
- `repo`: litebox
- `resource_id`: [the artifact ID]
- This downloads the artifact zip file (GitHub automatically adds .zip)
- Extract the zip file using bash commands (unzip is allowed) to access the contents
3. **Read the log file**:
- After extraction, the files will be in a `tmp/` subdirectory
- Read the contents of `tmp/verifier-output.txt`
- This file contains the output from the qsat verifier run
### Step 2: Parse Bug Reports
Identify all bug reports in the log file. Bug reports have this format:
```
✗ BUG FOUND in function: <function_name>
Bug type: <bug_description>
```
Examples:
```
✗ BUG FOUND in function: elf
Bug type: Integer overflow in add operation: _2 add _20 (type: u64, bounds: u64 [0, 9223372036854775807])
✗ BUG FOUND in function: stack
Bug type: Integer overflow in add operation: _2 add _4 (type: u64, bounds: u64 [0, 9223372036854775807])
```
For each bug report, extract:
- Function name
- Bug type (overflow, bounds, panic, etc.)
- Operation details
- File path and line number (if present in the log)
### Step 3: Review Each Bug Report
For each identified bug:
1. **Locate the source code**:
- Use the function name and any file/line information to find the relevant code
- Search the codebase using grep if needed to locate the function
- Read the source file to understand the context
2. **Analyze the code**:
- Understand what the function does
- Check the bounds and constraints on the operation
- Look for existing validation, assertions, or safety checks
- Consider the calling context and input constraints
- Check for any safety comments explaining why operations are safe
3. **Determine true vs false positive**:
- **True Positive**: The bug is real and could cause:
- Integer overflow/underflow in normal execution
- Out-of-bounds memory access
- Panic or unwrap failures without proper error handling
- Division by zero
- Security vulnerabilities
- **False Positive**: The bug report is incorrect because:
- Input validation prevents the problematic values
- Type system or compiler guarantees safety
- Bounds checks exist in the code path
- The overflow is intentional and documented (e.g., wrapping arithmetic)
- The operation is unreachable or guarded by conditions
### Step 4: Create GitHub Discussion
Create a comprehensive GitHub Discussion summarizing the findings:
**Discussion Title**: `QSAT Verifier Analysis - [Date]`
**Discussion Body** (use GitHub-flavored markdown):
```markdown
# QSAT Verifier Analysis Report
**Workflow Run**: [Link to qsat-verifier.yml run]
**Analysis Date**: [Current date]
**Analyzed Artifact**: qsat-verifier-output (from verifier-output.txt)
## Executive Summary
- Total bugs reported: X
- True positives: Y
- False positives: Z
## 🔴 True Positives (Bugs to Fix)
For each true positive, include:
### [Bug Type] in `function_name` ([file:line])
**Bug Description**: [Explain the bug in plain language]
**Code Location**:
```rust
[Relevant code snippet]
```
**Why This Is a Bug**:
[Clear explanation of why this is a genuine security or correctness issue]
**Recommended Fix**:
[Specific suggestion for how to fix it]
---
## 🟢 False Positives (No Action Needed)
<details>
<summary><b>Show False Positives</b></summary>
For each false positive, briefly explain:
- Function name and bug type
- Why it's a false positive (validation exists, safe by design, etc.)
</details>
## Next Steps
1. Review and prioritize the true positive findings
2. Create issues for each critical bug
3. Implement fixes with proper testing
4. Re-run qsat verifier to confirm fixes
## Methodology
This analysis was performed by:
1. Downloading the most recent qsat-verifier.yml artifact
2. Parsing all bug reports from verifier-output.txt
3. Reviewing source code for each reported bug
4. Classifying bugs as true or false positives based on code analysis
```
## Guidelines
- **Be thorough**: Review every bug report in the log file
- **Be accurate**: Don't dismiss bugs without careful code review
- **Be clear**: Explain your reasoning for each classification
- **Be factual**: Don't add subjective labels to bugs such as _critical_. This is up to the developer to decide
- **Prioritize security**: Integer overflows in security-critical code have priority; they are not necessarily serious
- **Context matters**: Consider the purpose of this sandboxing library
- **Use evidence**: Quote relevant code when explaining your decisions
- **Format properly**: Use GitHub-flavored markdown with proper headers, code blocks, and progressive disclosure
- **Link back**: Include a link to the workflow run that generated the artifact
## Important Notes
- This is a security-focused Rust sandboxing library - treat security bugs with priority
- The qsat verifier uses static analysis and may have false positives
- When in doubt, classify as a true positive and let maintainers decide
- Focus on actionable findings rather than theoretical edge cases
- Use file paths and line numbers to help maintainers locate issues quickly
- If the artifact is missing or empty, clearly report this in the discussion
## GitHub MCP Server Tools Reference
You have access to the GitHub MCP server with the **actions toolset** enabled. Use these tools (NOT bash/curl):
### Available Tools:
1. **list_workflow_runs** - List workflow runs for a specific workflow
- Parameters: `owner`, `repo`, `resource_id` (workflow filename like `qsat-verifier.yml`)
- Returns: List of workflow runs sorted by creation date (newest first)
2. **list_workflow_run_artifacts** - List artifacts for a workflow run
- Parameters: `owner`, `repo`, `resource_id` (workflow run ID)
- Returns: List of artifacts with their IDs and names
3. **download_workflow_run_artifact** - Download a workflow artifact
- Parameters: `owner`, `repo`, `resource_id` (artifact ID)
- Returns: Download URL or zip file content
- Note: You can then use bash `unzip` command to extract the contents
### Workflow Context:
- Repository: `NikolajBjorner/litebox`
- Workflow file: `qsat-verifier.yml`
- Artifact name: `qsat-verifier-output`
The artifact zip contains:
- `tmp/verifier-output.txt` - Main log file with qsat verifier output (this is what you need to analyze)
- `tmp/build-output.txt` - Build log (optional reference)
- `tmp/mir_files/*.mir` - MIR files (optional reference)
- `tmp/mir_errors/*.err` - MIR error logs (optional reference)