Context Architecture Bundle

Repo-local memory for AI coding agents.

Try it | Use it | Bundle contents | Agent workflow | Sponsor

---

Context Architecture Bundle is a copyable project template that helps Codex, Claude, Gemini, and other coding agents restart cleanly, understand the system, change the right files, and close a session with evidence instead of scattered chat history.

Use it when a project is moving faster than its docs. The bundle gives your repo a durable context layer: current state, architecture ownership, validation meaning, decisions, handoff checkpoints, and repeatable agent workflows.

Why This Exists

Agentic coding often burns time on rediscovery. Every new session has to ask:

• What does this project do?
• Which files own this behavior?
• What tests prove the change?
• What decisions are already settled?
• What happened in the previous session?

This bundle moves those answers into repo files so the next agent starts from evidence, not memory.

Who It Is For

• Developers using AI coding agents on real repos.
• Open-source maintainers who want contributors to understand project context before editing.
• Teams that need a lightweight handoff loop without adopting a heavy process tool.

This is template-first. There is no package to install for v1: copy the files, fill the project-specific templates, and run the advisory checker.

Try It In 5 Minutes

git clone https://github.com/saastoagent/context_architecture_bundle.git
cd context_architecture_bundle
python scripts/check_doc_coverage.py --help
python scripts/check_doc_coverage.py --files README.md skills/create-context-architecture-bundle/SKILL.md

Then inspect the filled example:

examples/tiny-task-tracker/

The example shows the difference between the blank template files and a populated context architecture for a small app.

Use It In Your Project

1. Copy the bundle files into your project root.
2. Replace placeholders such as [PROJECT_NAME], [PRIMARY_RUNTIME], and [VALIDATION_COMMAND].
3. Fill the core files first:
   • critical_prompt.md
   • context.md
   • structure.md
   • SYSTEM_FLOW_INDEX.md
   • architecture/code-map.md
   • test_index/README.md
4. Ask your coding agent to use work_prompt.md at session start and closeout.
5. Run the advisory checker before closeout:

python scripts/check_doc_coverage.py

Detailed setup steps live in docs/setup.md. Daily usage lives in docs/usage.md.

What The Bundle Adds

Path	Purpose
critical_prompt.md	Stable north star, boundaries, and non-negotiables.
context.md	Concise live restart snapshot.
work_prompt.md	Start, feature-completion, and session-end prompts.
context_pipeline.md	Rules for where project knowledge belongs.
architecture/code-map.md	Subsystem-to-source/test/doc ownership map.
architecture/components/	Focused contracts for high-change subsystems.
SYSTEM_FLOW_INDEX.md	Compact runtime and UX flow index.
test_index/	Validation commands and what each one proves.
decisions/	Architecture decision records.
logs/ and context_checkpoints/	Session evidence and handoff notes.
skills/	Optional repo-local agent workflows.
scripts/check_doc_coverage.py	Advisory drift checker for changed files.

Example Agent Workflow

Start a session:

I'm working on this project.

Please read critical_prompt.md and context.md first, then inspect
architecture/code-map.md and the relevant component docs. Tell me the current
state, known risks, and the next concrete step.

Before changing a feature:

Before implementing this feature, identify the owning code-map row, likely
source files, docs that may need updates, and the validation command that should
prove the change.

End a session:

Close out using work_prompt.md. Update context.md, add a checkpoint, record
validation evidence, and run python scripts/check_doc_coverage.py.

More ready-to-use prompts are in AGENTIC_CODING_GUIDE.md.

Optional Repo-Local Skills

The bundle includes two optional skills for agents that support repo-local skills:

• skills/create-context-architecture-bundle creates the bundle from a new project idea or early scaffold.
• skills/populate-context-architecture retrofits the bundle onto an existing codebase using live source evidence.

The files still work without skills. The public interface is the file structure plus the checker script.

Contributing

Useful first contributions:

• improve the filled example
• add docs for another agent workflow
• strengthen scripts/check_doc_coverage.py
• add tests around edge cases
• propose clearer templates or prompts

Read CONTRIBUTING.md and ROADMAP.md before opening a pull request.

Sponsorship

If this project saves you repeated context rebuilding, consider sponsoring the maintainer through the links in .github/FUNDING.yml. Sponsorship helps fund examples, docs, tests, and compatibility work for real agentic coding workflows.

License

MIT. See LICENSE.