AGENTS.md - AI Assistant Development Guidelines

This document provides instructions for AI coding assistants (GitHub Copilot, Cursor, Claude Code, etc.) working on the VideoStream project. These guidelines ensure consistent code quality, proper workflow adherence, and maintainable contributions.

Version: 2.0 Last Updated: 2025-11-22 Applies To: EdgeFirst VideoStream Library

Overview
Critical Rules
Git Workflow
Code Quality Standards
Testing Requirements
License Policy
Security Practices
VideoStream-Specific Guidelines
Release Process

Overview

EdgeFirst VideoStream Library provides video I/O capabilities for embedded Linux applications, including camera capture, hardware encoding, and inter-process frame sharing.

When contributing to VideoStream, AI assistants should prioritize:

Resource efficiency: Memory, CPU, and power consumption matter on embedded devices
Code quality: Maintainability, readability, and adherence to established patterns
Testing: Comprehensive coverage with unit, integration, and edge case tests
Documentation: Clear explanations for complex logic and public APIs
License compliance: Strict adherence to approved open source licenses
Stay in project root: NEVER change directories - use paths instead

Quick Reference

Branch:

feature/EDGEAI-###-description

bugfix/EDGEAI-###-description

Commit:

EDGEAI-###: Brief description

(50-72 chars, what changed not how) PR: main=2 approvals, develop=1. Link JIRA ticket, ensure CI passes.

License Policy: ✅ MIT/Apache/BSD | ⚠️ LGPL (dynamic only) | ❌ GPL/AGPL Coverage: 70% minimum, 80%+ for core modules (lib/host.c, lib/client.c, lib/frame.c) Output: ❌ NEVER filter build/test/sbom with head/tail/grep | ✅ Use tee for long output

⚠️ Critical Rules

Rule #1: NEVER Use cd Commands

BANNED: Changing directories during command execution

❌ NEVER DO THIS:

cd build && cmake ..
cd tests && pytest

✅ ALWAYS DO THIS:

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)
pytest tests/

Why: AI context doesn't reliably track current directory. Stay in project root, use paths.

Rule #2: ALWAYS Use Python venv

❌ NEVER DO THIS:

pip install package
python script.py
pytest tests/

✅ ALWAYS DO THIS:

venv/bin/pip install package
venv/bin/python script.py
venv/bin/pytest tests/

Setup:

python3 -m venv venv
venv/bin/pip install -r requirements.txt

Rule #3: Source env.sh Before Tests (If Exists)

# env.sh contains ephemeral tokens (<48h lifespan) - NEVER commit!
[ -f env.sh ] && source env.sh
venv/bin/pytest tests/

# Or use make test (handles venv + env.sh automatically)
make test

Security: ✅ Ephemeral tokens | ❌ NEVER commit env.sh | ❌ NO passwords

Rule #4: NEVER Filter Output of Non-Trivial Commands

BANNED: Using head, tail, grep to hide output that users need to monitor

❌ NEVER DO THIS:

make test | head -20
cmake --build build 2>&1 | grep -i error
make sbom | tail -10
venv/bin/pytest tests/ | grep PASSED

✅ ALWAYS DO THIS:

# Show full output for commands users need to monitor
make test
cmake --build build
make sbom
venv/bin/pytest tests/

# For very long output, capture to file for review
make test 2>&1 | tee test-output.log
cmake --build build 2>&1 | tee build-output.log

Exception: Fast parsing commands are OK:

# Quick version checks - OK to filter
grep "VSL_VERSION" include/videostream.h
cat pyproject.toml | grep "^version"

Why:

Builds/Tests/SBOM: Users need to see progress, warnings, errors in real-time
Debugging: Filtering hides critical context (warnings before errors, timing info)
CI/CD transparency: Full output shows what actually happened
tee preserves output: User sees everything AND gets log file for detailed review

What counts as "non-trivial":

❌ Build commands (cmake, make, cargo build)
❌ Test suites (pytest, cargo test, make test)
❌ SBOM generation (make sbom)
❌ Linting/formatting with multiple files (make lint, make format)
❌ Any command taking >1 second
✅ Simple parsing (grep version from single file)
✅ File counting (wc -l, ls | wc)

Git Workflow

Branch Naming

Format:

<type>/EDGEAI-###[-description]

Types:

```
feature/
```
- New features
```
bugfix/
```
- Bug fixes
```
hotfix/
```
- Critical fixes

Examples:

feature/EDGEAI-123-add-dmabuf-support
bugfix/EDGEAI-456-fix-memory-leak

Commit Messages

Format:

EDGEAI-###: Brief description

Good commits:

EDGEAI-123: Add DmaBuf support for zero-copy frame sharing
EDGEAI-456: Fix memory leak in frame pool cleanup

Pull Requests

main: 2 approvals required
develop: 1 approval required
All CI/CD checks must pass
Link JIRA ticket in description

Code Quality Standards

Language Standards

C11:

Compiler flags:
```
-Wall -Wextra -Werror
```
Formatter: clang-format (
```
make format
```
)
Error handling: Return negative errno values

Rust:

Toolchain: Stable (1.70+)
Formatter:
```
cargo fmt
```
Linter:
```
cargo clippy -- -D warnings
```
(
```
make lint
```
)

Python:

Version: 3.8+
Formatter: autopep8 (in venv)
Always use venv

Performance Considerations

Edge AI constraints:

Memory: 512MB-2GB RAM
CPU: <2% for frame distribution
Latency: <3ms for zero-copy DmaBuf
Hardware: Leverage G2D, Hantro VPU when available

Testing Requirements

Coverage

Minimum: 70% (project-wide)
Core modules: 80%+ (lib/host.c, lib/client.c, lib/frame.c)
Edge cases: Explicit tests for boundaries
Error paths: Validate error handling

Running Tests

# Using make (recommended - handles venv + env.sh + library path)
make test

# Manual (requires setup)
[ -f env.sh ] && source env.sh
source venv/bin/activate
export VIDEOSTREAM_LIBRARY=./build/libvideostream.so.1
pytest tests/

Test files:

```
tests/test_frame.py
```
- Frame lifecycle
```
tests/test_ipc.py
```
- IPC protocol
```
tests/test_host.py
```
- Host-side management
```
tests/test_client.py
```
- Client-side acquisition

License Policy

Allowed ✅

MIT, Apache-2.0, BSD-2/3, ISC, 0BSD, Unlicense, Public Domain

Conditional ⚠️

LGPL (dynamic linking only - NEVER static) MPL-2.0 (external dependencies only)

Blocked ❌

GPL (any version), AGPL, SSPL, Commons Clause

Verification

# REQUIRED before release
make sbom

CI/CD automatically blocks GPL/AGPL violations.

Security Practices

Vulnerability Reporting

Email:

[email protected]

(Subject: "Security Vulnerability - VideoStream")

Secure Coding

Input Validation:

Validate all external inputs
Use allowlists, enforce size limits

Credentials:

NEVER hardcode credentials
Use ephemeral tokens (<48h) in env.sh
NEVER commit env.sh

DmaBuf Security:

Understand FDs grant direct memory access
Close FDs immediately after use
Don't share with untrusted processes

VideoStream-Specific Guidelines

Technology Stack

Language: C11 (ISO/IEC 9899:2011)
Build system: CMake 3.10+
Dependencies:
- GStreamer 1.4+ (LGPL-2.1)
- GLib 2.0+ (LGPL-2.1)
- pthread
- stb libraries (Public Domain)
Platforms: NXP i.MX8, ARM64/ARMv7, x86_64
Acceleration: G2D, Hantro VPU, V4L2 DmaBuf
Kernel: Linux 4.14+ (5.6+ recommended)

Architecture

Pattern: Host/Client IPC with event-driven frame sharing

Layers:

Public API (vsl_host_, vsl_client_, vsl_frame_*)
IPC Protocol (UNIX sockets, FD passing)
Frame Management (reference counting)
Memory Allocation (DmaBuf / POSIX shm)

Threading:

Host: Main + GStreamer task thread
Client: Main + POSIX timer thread

Error handling: Negative errno values (e.g., -EINVAL), 0 for success

See ARCHITECTURE.md for details.

Build Commands

IMPORTANT: Use modern CMake workflow - stay in project root.

# Build (Debug)
cmake -S . -B build -DCMAKE_BUILD_TYPE=Debug
cmake --build build -j$(nproc)

# Build (Release)
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)

# Run tests
make test

# Format code
make format

# Lint code
make lint

# Generate SBOM
make sbom

# Verify version consistency
make verify-version

# Generate PDF documentation
make doc

# All pre-release checks
make pre-release

# Cross-compile for ARM64
source /opt/fsl-imx-xwayland/5.4-zeus/environment-setup-aarch64-poky-linux
cmake -S . -B build-arm64 -DCMAKE_BUILD_TYPE=Release
cmake --build build-arm64 -j$(nproc)

# Install
cmake --install build

# Clean
make clean

CMake Options:

```
-DENABLE_GSTREAMER=ON/OFF
```
- Build GStreamer plugins (default: ON)
```
-DENABLE_DMABUF=ON/OFF
```
- Enable DmaBuf support (default: ON on Linux)
```
-DENABLE_G2D=ON/OFF
```
- Enable G2D acceleration (default: ON on Linux)
```
-DENABLE_VPU=ON/OFF
```
- Enable VPU encode/decode (default: ON on Linux)
```
-DBUILD_TESTING=ON/OFF
```
- Build test suite (default: OFF)
```
-DENABLE_COVER=ON/OFF
```
- Enable code coverage (default: OFF)

Performance Targets

NXP i.MX8M Plus:

Frame latency: <3ms (DmaBuf), <10ms (shm)
Throughput: 1080p@60fps (3 clients), 4K@30fps (1 client)
Memory overhead: <100KB per client
CPU usage: <2% for distribution (1080p30)
Startup: <100ms

Critical paths:

Frame registration: <0.1ms
Event broadcast: <0.5ms per client
Lock/unlock round-trip: <2ms

Hardware Platforms

Supported:

NXP i.MX8M Plus: Full (G2D, VPU, DmaBuf)
NXP i.MX8M: DmaBuf + basic acceleration
Generic ARM64/ARMv7: POSIX shm fallback
x86_64: Development/testing (shm mode)

Acceleration:

G2D: Format conversion, scaling, rotation
Hantro VPU: H.264/H.265 encode/decode
V4L2 DMABUF: Zero-copy camera capture

Platform quirks:

i.MX8: G2D requires contiguous physical memory (CMA)
DmaBuf heap:
```
/dev/dma_heap/system
```
on Linux 5.6+
Permissions: User must be in
```
video
```
group

Release Process

CRITICAL: VideoStream uses manual release requiring version sync across 6 files.

Version Files

```
Cargo.toml
```
-
```
version = "X.Y.Z"
```

include/videostream.h

#define VSL_VERSION "X.Y.Z"

```
pyproject.toml
```
-
```
version = "X.Y.Z"
```
```
doc/conf.py
```
-
```
version = 'X.Y.Z'
```
(single quotes!)

debian/changelog

videostream (X.Y.Z-1) stable

```
CHANGELOG.md
```
-
```
## [X.Y.Z] - YYYY-MM-DD
```

Note:

CMakeLists.txt

parses from

include/videostream.h

automatically.

Release Workflow

Step 1: Pre-release checks

make pre-release
# Runs: format, lint, verify-version, test, sbom

Step 2: Determine version

PATCH (X.Y.Z): Bug fixes, non-breaking changes
MINOR (X.Y.0): New features, breaking changes
MAJOR (X.0.0): Only on explicit request

Step 3: Update CHANGELOG.md

Move items from
```
## [Unreleased]
```
to new version
Follow Keep a Changelog format

Step 4: Update ALL 6 version files

make verify-version

Step 5: Commit

git commit -a -m "Prepare Version X.Y.Z

- Updated all version files to X.Y.Z
- Finalized CHANGELOG.md
- See CHANGELOG.md for details"

git push origin main

Step 6: Wait for CI/CD (all green checkmarks)

Step 7: Tag and push

# v prefix REQUIRED to trigger release workflow
git tag -a -m "Version X.Y.Z" vX.Y.Z
git push origin vX.Y.Z

Common Mistakes

❌ Forgetting to update all 6 version files ❌ Tagging before CI/CD passes ❌ Missing

prefix on tag ❌ Running tests without venv or library path ❌ Using

cd

commands ❌ Filtering output of build/test/sbom commands

✅ Run

make pre-release

, wait for CI, tag with

vX.Y.Z

AI Assistant Best Practices

Verify:

APIs exist (no hallucinations)
License compatibility before adding deps
Test coverage meets 70% minimum
Version files synchronized before release
CI/CD passes before tagging

Avoid:

Using
```
cd
```
commands (stay in root)
System Python (always use venv)
GPL/AGPL dependencies
Hardcoded secrets
Committing env.sh
Filtering output with head/tail/grep on non-trivial commands

Remember: Review all AI-generated code thoroughly. YOU are the author, AI is a tool.

Getting Help

Documentation:

README.md - Overview and quick start
ARCHITECTURE.md - Internal design
DESIGN.md - High-level architecture
CONTRIBUTING.md - Detailed contribution guide
EdgeFirst Docs

Support: [email protected]

v2.0 | 2025-11-22 | EdgeFirst VideoStream Library

This file helps AI assistants contribute effectively to VideoStream while maintaining quality, security, and consistency.

AGENTS.md - AI Assistant Development Guidelines

Version: 2.0 Last Updated: 2025-11-22 Applies To: EdgeFirst VideoStream Library

Overview
Critical Rules
Git Workflow
Code Quality Standards
Testing Requirements
License Policy
Security Practices
VideoStream-Specific Guidelines
Release Process

Overview

EdgeFirst VideoStream Library provides video I/O capabilities for embedded Linux applications, including camera capture, hardware encoding, and inter-process frame sharing.

When contributing to VideoStream, AI assistants should prioritize:

Resource efficiency: Memory, CPU, and power consumption matter on embedded devices
Code quality: Maintainability, readability, and adherence to established patterns
Testing: Comprehensive coverage with unit, integration, and edge case tests
Documentation: Clear explanations for complex logic and public APIs
License compliance: Strict adherence to approved open source licenses
Stay in project root: NEVER change directories - use paths instead

Quick Reference

Branch:

feature/EDGEAI-###-description

bugfix/EDGEAI-###-description

Commit:

EDGEAI-###: Brief description

(50-72 chars, what changed not how) PR: main=2 approvals, develop=1. Link JIRA ticket, ensure CI passes.

⚠️ Critical Rules

Rule #1: NEVER Use cd Commands

BANNED: Changing directories during command execution

❌ NEVER DO THIS:

cd build && cmake ..
cd tests && pytest

✅ ALWAYS DO THIS:

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)
pytest tests/

Why: AI context doesn't reliably track current directory. Stay in project root, use paths.

Rule #2: ALWAYS Use Python venv

❌ NEVER DO THIS:

pip install package
python script.py
pytest tests/

✅ ALWAYS DO THIS:

venv/bin/pip install package
venv/bin/python script.py
venv/bin/pytest tests/

Setup:

python3 -m venv venv
venv/bin/pip install -r requirements.txt

Rule #3: Source env.sh Before Tests (If Exists)

# env.sh contains ephemeral tokens (<48h lifespan) - NEVER commit!
[ -f env.sh ] && source env.sh
venv/bin/pytest tests/

# Or use make test (handles venv + env.sh automatically)
make test

Security: ✅ Ephemeral tokens | ❌ NEVER commit env.sh | ❌ NO passwords

Rule #4: NEVER Filter Output of Non-Trivial Commands

BANNED: Using head, tail, grep to hide output that users need to monitor

❌ NEVER DO THIS:

make test | head -20
cmake --build build 2>&1 | grep -i error
make sbom | tail -10
venv/bin/pytest tests/ | grep PASSED

✅ ALWAYS DO THIS:

# Show full output for commands users need to monitor
make test
cmake --build build
make sbom
venv/bin/pytest tests/

# For very long output, capture to file for review
make test 2>&1 | tee test-output.log
cmake --build build 2>&1 | tee build-output.log

Exception: Fast parsing commands are OK:

# Quick version checks - OK to filter
grep "VSL_VERSION" include/videostream.h
cat pyproject.toml | grep "^version"

Why:

Builds/Tests/SBOM: Users need to see progress, warnings, errors in real-time
Debugging: Filtering hides critical context (warnings before errors, timing info)
CI/CD transparency: Full output shows what actually happened
tee preserves output: User sees everything AND gets log file for detailed review

What counts as "non-trivial":

❌ Build commands (cmake, make, cargo build)
❌ Test suites (pytest, cargo test, make test)
❌ SBOM generation (make sbom)
❌ Linting/formatting with multiple files (make lint, make format)
❌ Any command taking >1 second
✅ Simple parsing (grep version from single file)
✅ File counting (wc -l, ls | wc)

Git Workflow

Branch Naming

Format:

<type>/EDGEAI-###[-description]

Types:

```
feature/
```
- New features
```
bugfix/
```
- Bug fixes
```
hotfix/
```
- Critical fixes

Examples:

feature/EDGEAI-123-add-dmabuf-support
bugfix/EDGEAI-456-fix-memory-leak

Commit Messages

Format:

EDGEAI-###: Brief description

Good commits:

EDGEAI-123: Add DmaBuf support for zero-copy frame sharing
EDGEAI-456: Fix memory leak in frame pool cleanup

Pull Requests

main: 2 approvals required
develop: 1 approval required
All CI/CD checks must pass
Link JIRA ticket in description

Code Quality Standards

Language Standards

C11:

Compiler flags:
```
-Wall -Wextra -Werror
```
Formatter: clang-format (
```
make format
```
)
Error handling: Return negative errno values

Rust:

Toolchain: Stable (1.70+)
Formatter:
```
cargo fmt
```
Linter:
```
cargo clippy -- -D warnings
```
(
```
make lint
```
)

Python:

Version: 3.8+
Formatter: autopep8 (in venv)
Always use venv

Performance Considerations

Edge AI constraints:

Memory: 512MB-2GB RAM
CPU: <2% for frame distribution
Latency: <3ms for zero-copy DmaBuf
Hardware: Leverage G2D, Hantro VPU when available

Testing Requirements

Coverage

Minimum: 70% (project-wide)
Core modules: 80%+ (lib/host.c, lib/client.c, lib/frame.c)
Edge cases: Explicit tests for boundaries
Error paths: Validate error handling

Running Tests

# Using make (recommended - handles venv + env.sh + library path)
make test

# Manual (requires setup)
[ -f env.sh ] && source env.sh
source venv/bin/activate
export VIDEOSTREAM_LIBRARY=./build/libvideostream.so.1
pytest tests/

Test files:

```
tests/test_frame.py
```
- Frame lifecycle
```
tests/test_ipc.py
```
- IPC protocol
```
tests/test_host.py
```
- Host-side management
```
tests/test_client.py
```
- Client-side acquisition

# REQUIRED before release
make sbom

CI/CD automatically blocks GPL/AGPL violations.

Security Practices

Vulnerability Reporting

Email:

[email protected]

(Subject: "Security Vulnerability - VideoStream")

Secure Coding

Input Validation:

Validate all external inputs
Use allowlists, enforce size limits

Credentials:

NEVER hardcode credentials
Use ephemeral tokens (<48h) in env.sh
NEVER commit env.sh

DmaBuf Security:

Understand FDs grant direct memory access
Close FDs immediately after use
Don't share with untrusted processes

VideoStream-Specific Guidelines

Technology Stack

Language: C11 (ISO/IEC 9899:2011)
Build system: CMake 3.10+
Dependencies:
- GStreamer 1.4+ (LGPL-2.1)
- GLib 2.0+ (LGPL-2.1)
- pthread
- stb libraries (Public Domain)
Platforms: NXP i.MX8, ARM64/ARMv7, x86_64
Acceleration: G2D, Hantro VPU, V4L2 DmaBuf
Kernel: Linux 4.14+ (5.6+ recommended)

Architecture

Pattern: Host/Client IPC with event-driven frame sharing

Layers:

Public API (vsl_host_, vsl_client_, vsl_frame_*)
IPC Protocol (UNIX sockets, FD passing)
Frame Management (reference counting)
Memory Allocation (DmaBuf / POSIX shm)

Threading:

Host: Main + GStreamer task thread
Client: Main + POSIX timer thread

Error handling: Negative errno values (e.g., -EINVAL), 0 for success

See ARCHITECTURE.md for details.

Build Commands

IMPORTANT: Use modern CMake workflow - stay in project root.

# Build (Debug)
cmake -S . -B build -DCMAKE_BUILD_TYPE=Debug
cmake --build build -j$(nproc)

# Build (Release)
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)

# Run tests
make test

# Format code
make format

# Lint code
make lint

# Generate SBOM
make sbom

# Verify version consistency
make verify-version

# Generate PDF documentation
make doc

# All pre-release checks
make pre-release

# Cross-compile for ARM64
source /opt/fsl-imx-xwayland/5.4-zeus/environment-setup-aarch64-poky-linux
cmake -S . -B build-arm64 -DCMAKE_BUILD_TYPE=Release
cmake --build build-arm64 -j$(nproc)

# Install
cmake --install build

# Clean
make clean

CMake Options:

```
-DENABLE_GSTREAMER=ON/OFF
```
- Build GStreamer plugins (default: ON)
```
-DENABLE_DMABUF=ON/OFF
```
- Enable DmaBuf support (default: ON on Linux)
```
-DENABLE_G2D=ON/OFF
```
- Enable G2D acceleration (default: ON on Linux)
```
-DENABLE_VPU=ON/OFF
```
- Enable VPU encode/decode (default: ON on Linux)
```
-DBUILD_TESTING=ON/OFF
```
- Build test suite (default: OFF)
```
-DENABLE_COVER=ON/OFF
```
- Enable code coverage (default: OFF)

Performance Targets

NXP i.MX8M Plus:

Frame latency: <3ms (DmaBuf), <10ms (shm)
Throughput: 1080p@60fps (3 clients), 4K@30fps (1 client)
Memory overhead: <100KB per client
CPU usage: <2% for distribution (1080p30)
Startup: <100ms

Critical paths:

Frame registration: <0.1ms
Event broadcast: <0.5ms per client
Lock/unlock round-trip: <2ms

Hardware Platforms

Supported:

NXP i.MX8M Plus: Full (G2D, VPU, DmaBuf)
NXP i.MX8M: DmaBuf + basic acceleration
Generic ARM64/ARMv7: POSIX shm fallback
x86_64: Development/testing (shm mode)

Acceleration:

G2D: Format conversion, scaling, rotation
Hantro VPU: H.264/H.265 encode/decode
V4L2 DMABUF: Zero-copy camera capture

Platform quirks:

i.MX8: G2D requires contiguous physical memory (CMA)
DmaBuf heap:
```
/dev/dma_heap/system
```
on Linux 5.6+
Permissions: User must be in
```
video
```
group

Release Process

CRITICAL: VideoStream uses manual release requiring version sync across 6 files.

Version Files

```
Cargo.toml
```
-
```
version = "X.Y.Z"
```

include/videostream.h

#define VSL_VERSION "X.Y.Z"

```
pyproject.toml
```
-
```
version = "X.Y.Z"
```
```
doc/conf.py
```
-
```
version = 'X.Y.Z'
```
(single quotes!)

debian/changelog

videostream (X.Y.Z-1) stable

```
CHANGELOG.md
```
-
```
## [X.Y.Z] - YYYY-MM-DD
```

Note:

CMakeLists.txt

parses from

include/videostream.h

automatically.

Release Workflow

Step 1: Pre-release checks

make pre-release
# Runs: format, lint, verify-version, test, sbom

Step 2: Determine version

PATCH (X.Y.Z): Bug fixes, non-breaking changes
MINOR (X.Y.0): New features, breaking changes
MAJOR (X.0.0): Only on explicit request

Step 3: Update CHANGELOG.md

Move items from
```
## [Unreleased]
```
to new version
Follow Keep a Changelog format

Step 4: Update ALL 6 version files

make verify-version

Step 5: Commit

git commit -a -m "Prepare Version X.Y.Z

- Updated all version files to X.Y.Z
- Finalized CHANGELOG.md
- See CHANGELOG.md for details"

git push origin main

Step 6: Wait for CI/CD (all green checkmarks)

Step 7: Tag and push

# v prefix REQUIRED to trigger release workflow
git tag -a -m "Version X.Y.Z" vX.Y.Z
git push origin vX.Y.Z

Common Mistakes

❌ Forgetting to update all 6 version files ❌ Tagging before CI/CD passes ❌ Missing

prefix on tag ❌ Running tests without venv or library path ❌ Using

cd

commands ❌ Filtering output of build/test/sbom commands

✅ Run

make pre-release

, wait for CI, tag with

vX.Y.Z

AI Assistant Best Practices

Verify:

APIs exist (no hallucinations)
License compatibility before adding deps
Test coverage meets 70% minimum
Version files synchronized before release
CI/CD passes before tagging

Avoid:

Using
```
cd
```
commands (stay in root)
System Python (always use venv)
GPL/AGPL dependencies
Hardcoded secrets
Committing env.sh
Filtering output with head/tail/grep on non-trivial commands

Remember: Review all AI-generated code thoroughly. YOU are the author, AI is a tool.

Getting Help

Documentation:

README.md - Overview and quick start
ARCHITECTURE.md - Internal design
DESIGN.md - High-level architecture
CONTRIBUTING.md - Detailed contribution guide
EdgeFirst Docs

Support: [email protected]

v2.0 | 2025-11-22 | EdgeFirst VideoStream Library

This file helps AI assistants contribute effectively to VideoStream while maintaining quality, security, and consistency.

AGENTS.md - AI Assistant Development Guidelines

Related Skills

Markdown Converter

Nano Banana Pro

1password