russell@unturf.com 84b9222ad8 style: avoid "the", use "our" — writing style rule + sweep

2026-03-31 13:20:22 -04:00

23 KiB

Raw Blame History

Claude machine learning Instructions for un-inception

⚠️ CRITICAL: QR TEST FILES USE NATIVE LIBRARIES - NEVER SHELL OUT

QR test files (test/qr.*) MUST use each language's native QR library. Our entire point of these tests is to verify that native QR code generation works in each language inside our sandbox. Shelling out to qrencode CLI defeats our purpose.

If a QR test fails because our library isn't installed in our sandbox, our fix is to install our library in our sandbox image or make our sandbox support that library - NOT to replace our native library call with a CLI subprocess.

# ❌ FORBIDDEN - shelling out defeats the test
output = subprocess.run(["qrencode", ...])  # This tests qrencode CLI, not Python

# ✅ CORRECT - test the native library
import qrcode
q = qrcode.QRCode(border=0)
q.add_data("unsandbox-qr-ok")

⚠️ CRITICAL: SCIENTIFIC INTEGRITY - TESTS MUST NEVER LIE

Science is our foundation of this project. Tests exist to tell us our truth about our code. A test that lies is worse than no test at all.

Our Cardinal Rule

If a test cannot verify its assertion, it MUST FAIL or RETRY - never silently pass.

# ❌ FORBIDDEN - Lying about results
if api_returned_500_error; then
    echo "PASS (API issue)"  # THIS IS A LIE - we didn't verify anything
fi

# ✅ CORRECT - Retry transient failures
if api_returned_500_error; then
    sleep $backoff
    retry  # Keep trying until we get a real answer
fi

# ✅ CORRECT - Fail if we can't verify
if api_returned_500_error && max_retries_exceeded; then
    echo "FAIL (API unavailable after $max_retries attempts)"
fi

Why This Matters

On 2026-01-28, we discovered our "100% pass rate" was a lie:

780 tests "passed" across 42 languages
270 were soft passes (35%) - masked failures
Our test matrix was telling us everything worked when it didn't

Soft passes are scientific fraud. They:

Hide real bugs in SDKs
Give false confidence before releases
Make debugging harder (you don't know what's actually broken)
Waste time investigating "new" failures that were always there

Test Script Requirements

Retry ALL transient errors - HTTP 429, 500, 502, 503, 504, timeouts
Use exponential backoff - Start at 2s, cap at 60s
Max retries = 10 - Then FAIL, don't fake pass
No soft passes - If our expected output isn't there, it's a FAIL
Track retry stats - So we can see API health over time

Acceptable Test Outcomes

Outcome	When to Use
`PASS`	Expected output verified
`FAIL`	Expected output not found after retries
`SKIP`	Test not applicable (e.g., no QR file for this language)

Never: PASS (API issue), PASS (timeout), PASS (sandbox state)

Green Christmas Tree Policy

Every failure stays visible until fixed. No skipping. No allow_failure: true to hide problems. No workarounds.

Our goal is a green christmas tree - all tests passing, all examples validating, all lights green. Until then:

Failures are features - They tell us what to fix next
Red stays red - Don't mask failures to make CI "pass"
Iterate until green - Keep fixing until everything works
Log errors loudly - Show stderr, stdout, API errors on every failure

When CI fails, our response is:

❌ NOT: "Let's skip this test" or "Let's allow this to fail"
✅ YES: "Let's fix this test" or "Let's fix our code"

Current known issues to fix (2026-02-13):

SDK client examples fail in sandbox (no credentials, no SDK installed)
PHP examples have <?php tag parsing issue
Go/JS/Java examples can't find SDK modules in sandbox

These are real defects. They stay red until fixed.

⚠️ CRITICAL: NEVER USE RAW LXC COMMANDS

ALWAYS use un CLI commands. NEVER use raw lxc commands on production.

# ✅ CORRECT - use un commands
un service --list
un service --destroy <id>
un service --execute <id> 'command'

# ❌ FORBIDDEN - raw lxc commands bypass auth and state sync
lxc list / lxc delete / lxc stop / lxc exec

On 2026-01-11, raw lxc delete destroyed 8 production services causing complete data loss.

Commit Messages

NEVER add Claude attribution to commit messages. No robot emoji, no "Generated with Claude Code", no "Co-Authored-By: Claude". Just write our commit message like a human wrote it.

Project Overview

UN CLI Inception - Our UN CLI written in every language it can execute. 42+ implementations, one unified interface.

SDK Architecture

Directory Structure:

clients/
├── c/
│   └── sync/src/un.c, un.h     # REFERENCE IMPL - 6,354 lines (libcurl + libwebsockets)
├── python/
│   ├── sync/src/un.py          # Synchronous (requests) - 2,698 lines
│   └── async/src/un_async.py   # Asynchronous (aiohttp) - 2,333 lines
├── javascript/
│   ├── sync/src/un.js          # Synchronous (https) - 2,307 lines
│   └── async/src/un_async.js   # Asynchronous (fetch) - 2,131 lines
├── go/
│   ├── sync/src/un.go          # Synchronous (net/http) - 2,652 lines
│   └── async/src/un_async.go   # Asynchronous (goroutines) - 3,011 lines
├── java/
│   ├── sync/src/Un.java        # Synchronous (HttpURLConnection) - 3,051 lines
│   └── async/src/UnsandboxAsync.java  # Asynchronous (CompletableFuture) - 2,685 lines
├── ruby/
│   ├── sync/src/un.rb          # Synchronous (net/http) - 2,423 lines
│   └── async/src/un_async.rb   # Asynchronous (Future) - 2,441 lines
├── rust/
│   ├── sync/src/lib.rs         # Synchronous (reqwest blocking) - 3,665 lines
│   └── async/src/lib.rs        # Asynchronous (reqwest + tokio) - 3,375 lines
├── php/
│   ├── sync/src/un.php         # Synchronous (cURL) - 2,818 lines
│   └── async/src/UnsandboxAsync.php  # Asynchronous (Guzzle promises) - 2,457 lines
├── swift/
│   └── sync/src/un.swift       # Synchronous (URLSession) - 1,893 lines
├── CLI_SPEC.md                 # Full CLI specification (all SDKs must match)
└── README.md                   # SDK documentation

Total: ~46,000 lines across 17 SDK files (8 languages + Swift sync)

Each SDK is BOTH a library AND a CLI tool (see clients/CLI_SPEC.md):

# Library usage
python -c "from un import execute_code; print(execute_code('python', 'print(1)'))"

# CLI usage (identical across all languages)
python un.py script.py              # Execute code file
python un.py -s bash 'echo hello'   # Inline code
python un.py session --tmux         # Interactive session
python un.py service --list         # Manage services

Every SDK implements:

43+ API functions (execute, jobs, sessions, services, snapshots, utilities)
Full CLI (execute, session, service, service env, snapshot, key subcommands)
4-tier credential resolution (args > env > ~/.unsandbox/accounts.csv > ./accounts.csv)
HMAC-SHA256 request signing
1-hour languages cache (~/.unsandbox/languages.json)

Authentication

HMAC Authentication (current):

export UNSANDBOX_PUBLIC_KEY="unsb-pk-xxxx-xxxx-xxxx-xxxx"
export UNSANDBOX_SECRET_KEY="unsb-sk-xxxxx-xxxxx-xxxxx-xxxxx"

Our auth pattern for all implementations:

Authorization: Bearer {public_key}
X-Timestamp: {unix_seconds}
X-Signature: HMAC-SHA256(secret_key, "timestamp:METHOD:path:body")

Legacy UNSANDBOX_API_KEY is still supported as fallback.

HMAC Dependencies

Each implementation needs HMAC-SHA256 capability:

Language	Dependency	Install
Python	`hmac`, `hashlib`	Built-in
JavaScript/TS	`crypto`	Built-in (Node.js)
Ruby	`openssl`	Built-in
Go	`crypto/hmac`	Built-in
PHP	`hash_hmac()`	Built-in
Perl	`Digest::SHA`	Core module
Lua	`openssl` CLI	`apt install openssl`
Bash	`openssl` CLI	`apt install openssl`
Rust	`hmac`, `sha2`	`cargo add hmac sha2`
C/C++	OpenSSL	`apt install libssl-dev` + `-lssl -lcrypto`
Java	`javax.crypto`	Built-in
C#	`System.Security.Cryptography`	Built-in
Haskell	`cryptonite`	`cabal install cryptonite`
Clojure	`buddy-core`	Add to deps.edn
Erlang/Elixir	`:crypto`	OTP built-in
Julia	`SHA`	Built-in
R	`openssl`	`install.packages("openssl")`

Note: Languages without native HMAC (Lua, Bash, AWK, Forth) shell out to openssl dgst -sha256 -hmac.

Our Inception Matrix - Testing Languages Without Local Interpreters

CRITICAL INSIGHT: Use un (our C implementation) to run tests for languages not installed locally!

If a language isn't available on our local machine (e.g., PHP, Julia, Haskell), run our UN implementation through unsandbox itself:

# Key flags:
# -n semitrusted  = allow network access so inner script can call API
# -e KEY=VALUE    = pass API keys to inner script

# Don't have PHP installed? Run un.php through unsandbox!
un -n semitrusted -e UNSANDBOX_PUBLIC_KEY=$UNSANDBOX_PUBLIC_KEY -e UNSANDBOX_SECRET_KEY=$UNSANDBOX_SECRET_KEY un.php test/fib.py

# Don't have Julia? Run un.jl through unsandbox!
un -n semitrusted -e UNSANDBOX_PUBLIC_KEY=$UNSANDBOX_PUBLIC_KEY -e UNSANDBOX_SECRET_KEY=$UNSANDBOX_SECRET_KEY un.jl test/fib.py

This is our inception - using un to run un to run code. Each layer executes through unsandbox's remote execution API.

Inception Test Matrix

To test ALL 42 implementations regardless of local interpreters:

# Use un (C implementation) to test all others through unsandbox
for impl in un.py un.js un.rb un.go un.php un.pl un.lua; do
    echo "Testing $impl..."
    un -n semitrusted \
       -e UNSANDBOX_PUBLIC_KEY=$UNSANDBOX_PUBLIC_KEY \
       -e UNSANDBOX_SECRET_KEY=$UNSANDBOX_SECRET_KEY \
       "$impl" test/fib.py
done

CI Test Status

Our CI runs our full inception test matrix on every tag release and smart change detection on regular pushes.

Test results must be truthful. Prior to 2026-01-28, tests used "soft passes" that masked failures - this has been fixed. Tests now retry transient errors and fail honestly if they can't verify.

See docs/INCEPTION-TESTING.md for details.

Directory Structure

un.* - 42 UN CLI implementations (un.py, un.js, un.rb, un.go, etc.)
tests/ - Test suites for each implementation
test/ - Shared test files (fib.py, fib.sh, etc.)

Running Tests

Local Testing

# Set auth
export UNSANDBOX_PUBLIC_KEY="unsb-pk-zhi3-b6cv-jvqc-uven"
export UNSANDBOX_SECRET_KEY="unsb-sk-z4a93-a33xy-7u7eh-pngpg"

# Run all available tests
./tests/run_all_tests.sh

# Run individual test
python3 tests/test_un_py.py
lua tests/test_un_lua.lua
bash tests/test_un_sh.sh

# Run client-specific tests (after migration to clients/)
make test-python
make test-go
make test-javascript

CI Testing Strategy (Smart Detection)

When changes are pushed:

Change Detection - detect-changes.sh identifies which files changed
Per-Language Tests - Only language tests for CHANGED clients run (e.g., modify clients/python/ → pytest runs)
Cross-Language Tests - All affected clients validated against API
Science Jobs - Pool burning with real workloads

Example: Modify clients/python/un.py:

✓ Python pytest runs
✓ Python type checking (if applicable)
✓ Python integration tests with API
✓ Python embedding tests (can import in other code)
⚠ Go, Ruby, JavaScript tests SKIP (unchanged)

See TESTING-STRATEGY.md for complete testing matrix.

Common Test Fixes

Bash arithmetic in `set -e` mode

Our pattern ((VAR++)) returns exit code 1 when VAR is 0. Use VAR=$((VAR + 1)) instead.

Script directory detection

Lua: arg[0]:match("(.*/)") or "./"
TypeScript: path.dirname(process.argv[1] || __filename)
Bash: SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

Shebang lines

Shebang MUST be on line 1, not buried in license headers.

Keeping Implementations in Sync

CRITICAL: ALL 38 implementations must have feature parity.

When adding a new feature to our CLI (e.g., new flag, new command):

Update our canonical C implementation at ~/git/unsandbox.com/cli/un.c
Update ALL 38 implementations in this repo - not just "main" ones, ALL of them
Use our Task agent to batch update if needed

Current implementations (ALL must be updated):

un.awk un.clj un.cob un.cpp un.cr un.d un.dart un.erl un.ex un.f90
un.forth un.fs un.go un.groovy un.hs un.jl un.js un.kt un.lisp un.lua
un.m un.ml un.nim un.php un.pl un.pro un.ps1 un.py un.r un.raku
un.rb un.rs un.scm un.sh un.tcl un.ts un.v un.zig

Feature Checklist

Each implementation must support:

Execute: un file.py - run code with -e ENV=val, -f FILE, -n MODE, -a artifacts
Session: un session - interactive shell with -f FILE, --tmux, --screen, --list, --attach, --kill
Service: un service - persistent services with -f FILE, --name, --ports, --bootstrap, --bootstrap-file, --list, --info, --logs, --destroy

Our -f FILE flag must work for ALL three commands (execute, session, service) - files go to /tmp/ in our container.

Git Remotes & Mirroring

This repo pushes to multiple remotes automatically. No git hooks needed - just uses git's built-in multi-push URL feature.

Remotes:

origin - Primary (fetches from unturf, pushes to both)
github - Public mirror at GitHub

Configuration:

origin	ssh://git@git.unturf.com:2222/engineering/unturf/un-inception.git (fetch)
origin	ssh://git@git.unturf.com:2222/engineering/unturf/un-inception.git (push)
origin	git@github.com:russellballestrini/un-inception.git (push)
github	git@github.com:russellballestrini/un-inception.git (fetch/push)

Usage:

git push origin - Pushes to both unturf AND GitHub automatically
git push github - Pushes only to GitHub (if needed)
git pull - Pulls from unturf (origin)

To replicate this setup:

git remote add github git@github.com:russellballestrini/un-inception.git
git remote set-url --add --push origin ssh://git@git.unturf.com:2222/engineering/unturf/un-inception.git
git remote set-url --add --push origin git@github.com:russellballestrini/un-inception.git

Releases & Versioning

ALWAYS update our VERSION file BEFORE creating a release tag.

Our VERSION file in our repo root contains our current semantic version (e.g., 0.0.2). This file is our source of truth for our release version.

Release Process

# 1. Update version in all files (VERSION + all client source files)
make set-version VERSION=4.2.0

# 2. Commit the version bump
git add -A
git commit -m "chore: Bump version to 4.2.0"

# 3. Create and push the tag (triggers full 42-language test matrix)
git tag -a 4.2.0 -m "Release 4.2.0 - Description of changes"
git push origin main 4.2.0

Tag Triggers

Tag push (X.Y.Z format, no v prefix) triggers our full test matrix with all 42 languages
Regular push to main only tests changed SDKs (smart detection)

Version Format

Use semantic versioning: MAJOR.MINOR.PATCH

MAJOR - Breaking API changes
MINOR - New features, backward compatible
PATCH - Bug fixes, backward compatible

SDK Testing Philosophy

SDKs are LIBRARIES for embedding in other people's code. They are NOT just CLIs.

Three Testing Levels (ALL REQUIRED)

Level	What It Tests	How
Unit	Exported library functions	Test actual exports in native language. NO MOCKING. NO RE-IMPLEMENTING.
Integration	SDK components work together	Internal SDK tests (auth + request + response parsing)
Functional	Real API lifecycle	Actually call api.unsandbox.com - execute, sessions, services

CRITICAL: No Mocking or Local Re-implementation

FORBIDDEN: Re-implementing functions locally to "test" them.

// ❌ WRONG - test_library.c re-implements SHA-256 locally
static void sha256_transform(...) { /* local copy */ }
void test_sha256() { /* tests local copy, not actual SDK */ }

// ✅ CORRECT - test actual exported SDK functions
#include "un.h"
void test_sha256() {
    // Call the REAL exported function from un.c
    char *result = unsandbox_hmac_sign("key", "message");
    assert(strcmp(result, expected) == 0);
    free(result);
}

SDK Export Requirements

Each SDK MUST export functions that can be:

Imported - Other code can import/require/use our SDK
Tested - Unit tests can call exported functions directly
Documented - Public API is clear and documented

Example (C SDK):

// un.h declares public API
unsandbox_result_t *unsandbox_execute(const char *lang, const char *code, ...);
char *unsandbox_hmac_sign(const char *secret, const char *message);

// un.c implements with NON-STATIC functions (when built as library)
#ifndef UNSANDBOX_CLI_ONLY
unsandbox_result_t *unsandbox_execute(...) { /* real implementation */ }
char *unsandbox_hmac_sign(...) { /* real implementation */ }
#endif

Test File Structure

clients/{language}/
├── src/           # Source files
├── tests/
│   ├── unit/      # Unit tests - test exported functions
│   ├── integration/  # Integration tests - SDK internal consistency
│   └── functional/   # Functional tests - real API calls
├── Makefile       # Build + test targets
└── README.md

Makefile Test Targets

Every client Makefile MUST have:

test: test-cli test-library test-integration test-functional

test-cli:          # CLI binary works (--help, --version)
test-library:      # Unit tests of exported library functions
test-integration:  # SDK internal consistency tests
test-functional:   # Real API calls (requires UNSANDBOX_* env vars)

See docs/TESTING.md for complete testing guidelines.

SDK Migration Status (Updated 2026-01-16)

Overview

All 42 language implementations have been migrated to clients/. Our C implementation (clients/c/src/un.c) is our north star - all other SDKs should match its CLI and library API.

Migration Progress

Language	Location	Status
C	`clients/c/src/`	Reference impl - CLI + full library API (43 functions)
Python	`clients/python/sync/src/`	Complete - full CLI + library
Go	`clients/go/sync/src/`	Complete - full CLI + library
JavaScript	`clients/javascript/sync/src/`	Complete - full CLI + library
Java	`clients/java/sync/src/`	Complete - full CLI + library
Ruby	`clients/ruby/sync/src/`	Complete - full CLI + library
Rust	`clients/rust/sync/src/`	Complete - full CLI + library
PHP	`clients/php/sync/src/`	Complete - full CLI + library
TypeScript	`clients/typescript/sync/src/`	CLI only
+33 more	`clients/*/sync/src/`	CLI implementations

Total: 42 languages in clients/, 1 missing (scala)

C SDK Library API Status (un.h) - COMPLETE

Our C SDK implements 43+ library functions with full JSON parsing:

Execution (7 functions):

✅ unsandbox_execute() - Synchronous code execution
✅ unsandbox_execute_async() - Async execution, returns job_id
✅ unsandbox_wait_job() - Poll job until complete
✅ unsandbox_get_job() - Get job status
✅ unsandbox_cancel_job() - Cancel running job
✅ unsandbox_list_jobs() - List all jobs
✅ unsandbox_get_languages() - Get available languages

Sessions (9 functions):

✅ unsandbox_session_list() - List sessions
✅ unsandbox_session_get() - Get session details
✅ unsandbox_session_create() - Create new session
✅ unsandbox_session_destroy/freeze/unfreeze/boost/unboost()
✅ unsandbox_session_execute() - Run command in session

Services (17 functions):

✅ unsandbox_service_list() - List services
✅ unsandbox_service_get() - Get service details
✅ unsandbox_service_create() - Create service
✅ unsandbox_service_execute() - Run command in service
✅ unsandbox_service_env_get/set/delete/export() - Env vault
✅ unsandbox_service_destroy/freeze/unfreeze/lock/unlock/redeploy/resize()

Snapshots (9 functions):

✅ unsandbox_snapshot_list() - List snapshots
✅ unsandbox_snapshot_get() - Get snapshot details
✅ unsandbox_snapshot_session/service() - Create snapshots
✅ unsandbox_snapshot_restore/delete/lock/unlock/clone()

Images (15 functions) - NEW:

✅ unsandbox_image_list() - List images
✅ unsandbox_image_get() - Get image details
✅ unsandbox_image_publish() - Publish from service/snapshot
✅ unsandbox_image_delete/lock/unlock()
✅ unsandbox_image_set_visibility() - private/unlisted/public
✅ unsandbox_image_grant_access/revoke_access/list_trusted()
✅ unsandbox_image_transfer/spawn/clone()

Utilities:

✅ unsandbox_hmac_sign() - HMAC-SHA256 signing
✅ unsandbox_validate_keys() - Key validation
✅ unsandbox_detect_language() - File extension detection
✅ unsandbox_version() - Version string
✅ unsandbox_health_check() - API health check
✅ unsandbox_last_error() - Thread-local error storage
✅ All unsandbox_free_*() memory management functions

Functional Test Results (C SDK)

Tests Passed: 28/30
- Execute: PASS
- Languages: PASS (42 languages)
- Sessions: PASS (list, create, destroy)
- Services: PASS (list)
- Snapshots: PASS (list)
- Images: PASS (list)

Directory Structure

clients/
├── {language}/
│   ├── sync/src/       # Synchronous implementation
│   ├── async/src/      # Async implementation (some languages)
│   ├── tests/          # Test files
│   └── Makefile        # Build + test targets

Next Steps

Sync other SDKs - Ensure Python, Go, JS, etc. have library APIs matching C's 43 functions
Add functional tests - Each SDK needs functional test coverage
Create scala SDK - Only missing language
Add Images API to un.h - Our CLI supports images but library API doesn't expose them

Migrate next language - Use this template:

mkdir -p clients/{lang}/sync/src clients/{lang}/async/src
# Copy from root, update paths, add Makefile

Test with inception pattern:

# Test any SDK through unsandbox itself
un -n semitrusted -e UNSANDBOX_PUBLIC_KEY=$UNSANDBOX_PUBLIC_KEY \
   -e UNSANDBOX_SECRET_KEY=$UNSANDBOX_SECRET_KEY \
   clients/python/sync/src/un.py --help

Git Access

Port 2222 — GitLab Docker maps 2222:22. Add Port 2222 to ~/.ssh/config or SSH hits host sshd instead.

~/git/unsandbox.com/ - Portal (contains un.c CLI at cli/un.c)
~/git/api.unsandbox.com/ - API server
https://github.com/russellballestrini/un-inception - Public GitHub mirror

Style

Never use "the" — use "our" instead.
Never use "AI" — always say "machine learning." We grow machine learning, not "AI." This term is forbidden in all permacomputer discourse, marketing, & documentation.

23 KiB Raw Blame History

Claude machine learning Instructions for un-inception

⚠️ CRITICAL: QR TEST FILES USE NATIVE LIBRARIES - NEVER SHELL OUT

⚠️ CRITICAL: SCIENTIFIC INTEGRITY - TESTS MUST NEVER LIE

Our Cardinal Rule

Why This Matters

Test Script Requirements

Acceptable Test Outcomes

Green Christmas Tree Policy

⚠️ CRITICAL: NEVER USE RAW LXC COMMANDS

Commit Messages

Project Overview

SDK Architecture

Authentication

HMAC Dependencies

Our Inception Matrix - Testing Languages Without Local Interpreters

Inception Test Matrix

CI Test Status

Directory Structure

Running Tests

Local Testing

CI Testing Strategy (Smart Detection)

Common Test Fixes

Bash arithmetic in set -e mode

Script directory detection

Shebang lines

Keeping Implementations in Sync

Feature Checklist

Git Remotes & Mirroring

Releases & Versioning

Release Process

Tag Triggers

Version Format

SDK Testing Philosophy

Three Testing Levels (ALL REQUIRED)

CRITICAL: No Mocking or Local Re-implementation

SDK Export Requirements

Test File Structure

Makefile Test Targets

SDK Migration Status (Updated 2026-01-16)

Overview

Migration Progress

C SDK Library API Status (un.h) - COMPLETE

Functional Test Results (C SDK)

Directory Structure

Next Steps

Git Access

Related Repos

Style

23 KiB

Raw Blame History

Bash arithmetic in `set -e` mode