architecture-decision-records
Skill2 starsChanged 5 months agoApache-2.0View on GitHub
Write and maintain Architecture Decision Records (ADRs) for the atlas project. Covers ADR format, where to store them, when to create one, and examples based on past decisions. Use when making a significant technical decision that should be recorded for future maintainers.
70 other projects use a version of this.
Adopted by
Most recently taken up by
- AtilaMedeiros/skillstheir own version, first seen Nov 16, 2026
- hmohammed7/claude-skillstheir own version, first seen Nov 16, 2026
- AndresArreguy/ecc-pi-porttheir own version, first seen Nov 16, 2026
- roronoazoroshao369/vibe-coding-ostheir own version, first seen Nov 15, 2026
- Lukk17/agent-standardstheir own version, first seen Nov 13, 2026
- ricoberger/dotfilestheir own version, first seen Nov 11, 2026
Used with
Projects with it also use nuxt4-patterns, dart-flutter-patterns and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
loulanyueSkillnuxt4-patterns
Nuxt 4 app patterns for hydration safety, performance, route rules, lazy loading, and SSR-safe data fetching with useFetch and useAsyncData.
Also in 8 projects that use this
RojiosSkilldart-flutter-patterns
Production-ready Dart and Flutter patterns covering null safety, immutable state, async composition, widget architecture, popular state management frameworks (BLoC, Riverpod, Provider), GoRouter navigation, Dio networking, Freezed code generation, and clean architecture.
Also in 5 projects that use this
royaluniondesign-sysSkillcontent-hash-cache-pattern
Cache expensive file processing results using SHA-256 content hashes — path-independent, auto-invalidating, with service layer separation.
Also in 11 projects that use this
Yog-SothoSkillpytorch-patterns
PyTorch deep learning patterns and best practices for building robust, efficient, and reproducible training pipelines, model architectures, and data loading. Always activate when the user is writing PyTorch models or training scripts, reviewing deep learning code, debugging training loops or data pipelines, optimizing GPU memory or training speed, setting up reproducible experiments, implementing transfer learning or fine-tuning, or asks anything about nn.Module, DataLoader, autograd, AMP, or torch.compile.
Also in 11 projects that use this
junimnjwSkillralphinho-rfc-pipeline
RFC-driven multi-agent DAG execution pattern with quality gates, merge queues, and work unit orchestration.
Also in 7 projects that use this
junimnjwSkillcarrier-relationship-management
Codified expertise for managing carrier portfolios, negotiating freight rates, tracking carrier performance, allocating freight, and maintaining strategic carrier relationships. Informed by transportation managers with 15+ years experience. Includes scorecarding frameworks, RFP processes, market intelligence, and compliance vetting. Use when managing carriers, negotiating rates, evaluating carrier performance, or building freight strategies.
Also in 6 projects that use this
- +2 more from junimnjw
Other versions70 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 70
- loulanyue/awesome-claude-notes★ 272same name
- tranhieutt/software_development_department★ 71same name
- Fmarzochi/EGC★ 61same name
- GSA-TTS/devCrew_s1★ 52same name
- kumaran-is/claude-code-onboarding★ 35same name
- ricoberger/dotfiles★ 19same name
- listener-He/java-harness-agent★ 16same name
- woxiangyangzhimao/skills★ 14same name
- robertoatila/jarvis-skill-registry★ 9same name
- xfstudio/skills★ 9same name
60 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository16 other files it gives its agents.
- atlas-add-toolSkill
- atlas-build-bundleSkill
- atlas-ci-releaseSkill
- atlas-optimization-patternsSkill
- atlas-rag-evaluationSkill
- embedding-strategiesSkill
- hybrid-search-implementationSkill
- mcp-builderSkill
- python-code-styleSkill
- python-observabilitySkill
- python-resilienceSkill
- python-testing-patternsSkill
- python-type-safetySkill
- readme-maintenanceSkill
- skill-creatorSkill
- uv-package-managerSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 66 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Create an ADR when you make a decision that is:
- Distinctive: line 25
Do NOT create ADRs for:
- Distinctive: line 108
Decision: Use MLX framework for Apple Silicon GPU acceleration instead of CoreML.
- Very specific: line 116
Decision: Use huggingface_hub.snapshot_download instead of transformers.cached_file to resolve the model directory.
“Common” is measured: kept by at least 6 in 10 of the 66 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
When to use
Create an ADR when you make a decision that is:
- Significant — affects multiple modules, backends, or the build process
- Non-obvious — the rationale won't be immediately clear to future readers
- Costly to reverse — changing it later would require substantial rework
- Architecturally binding — constrains future design choices
Examples from this project's history that should have been ADRs:
- Choosing MLX over CoreML for Apple Silicon
- Using
snapshot_downloadinstead ofcached_filefor ONNX tokenizer loading - Reducing CI model from BGE-base (768-dim) to BGE-small (384-dim) for timeout reasons
- Choosing int8 quantisation over FP32 for CPU builds
Do NOT create ADRs for:
- Bug fixes
- Routine dependency upgrades
- Cosmetic refactors
- Decisions documented entirely by code comments
ADR format
Each ADR is a markdown file at docs/adr/NNNN-title-with-dashes.md.
File name
NNNN-title-with-dashes.mdWhere NNNN is the next available number, zero-padded to 4 digits.
Frontmatter
---
date: 2026-06-22
status: accepted # proposed | accepted | deprecated | superseded
supersedes: # optional: ADR number this replaces
superseded_by: # optional: ADR number that replaces this
---Template
---
date: <YYYY-MM-DD>
status: <proposed|accepted|deprecated|superseded>
---
# <Number>. <Title>
## Context
What is the problem we are solving? What constraints or forces are at play?
Include relevant background, requirements, and trade-offs.
## Decision
What did we decide? State it clearly and concretely.
## Rationale
Why this option over the alternatives? List 2–3 options considered and why
each was or wasn't chosen.
## Consequences
What becomes easier or harder as a result of this decision?
What does this decision enable or foreclose?
## Compliance
How will future code or CI enforce this decision? (optional)Where to store
docs/adr/
├── index.md # Table of contents, generated
├── 0001-mlx-over-coreml.md
└── 0002-snapshot-download-for-tokenizer.mdADR lifecycle
- Proposed — Decision is being discussed. File with
status: proposed. - Accepted — Decision is final. Change status to
accepted. - Deprecated — Decision is no longer relevant but not yet superseded.
- Superseded — A newer ADR replaces this one. Set
superseded_by.
Examples
0001: Choose MLX over CoreML for Apple Silicon
Decision: Use MLX framework for Apple Silicon GPU acceleration instead of CoreML.
Rationale: CoreML's ONNX conversion flow (coremltools) was unstable — caused SIGKILL crashes during conversion and produced models 30–40× slower than CPU. MLX provides direct model loading, stable Apple Silicon acceleration, and native float16 support.
Consequences: MLX becomes the default backend on Apple Silicon. ONNX+CPU remains the CI fallback. Linux/Windows users get ONNX+CPU only.
0002: Use snapshot_download for ONNX tokenizer loading
Decision: Use huggingface_hub.snapshot_download instead of transformers.cached_file to resolve the model directory.
Rationale: cached_file downloads only the ONNX file, missing tokenizer.json, config.json, and vocab files. snapshot_download guarantees the full model snapshot lands in one directory, so tokenizer loading never fails with "file not found."
Consequences: Slightly larger initial download (all model files instead of just ONNX). snapshot_download respects HF_TOKEN environment variable.
References
README.md— contains informal design rationale that could be migrated to ADRs over time- Module-level docstrings in
atlas/embed/onnx.py,atlas/chunk.py,atlas/rag_server.py— embedded design decisions
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

