CONTEXT/1.2 Workspace

Deterministic capsule format for Flagship agents: a shared language between the Reasoning Core, Memory Brain, and Tooling Mesh.

Goal. Guarantee that every .context capsule conforms to the specification and passes the golden suite before entering Modular Intellegence Systems production pipelines.

Repository Map

Path	Purpose
docs/context_spec_1_2.md	Authoritative CONTEXT/1.2 specification with CTX-CANON/3 annexes.
docs/testing.md	Guide for running and extending the golden suites.
.agents/tools/ctx_lint.py	Reference parser/linter used across all validations.
tests/context/	Positive and negative .context scenarios.
tests/outcomes/	Expected outputs for the golden suite.
.github/workflows/goldens.yml	CI workflow that executes the full golden suite on every push/PR.

Quickstart

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt  # ensure python>=3.11 if the file is absent
python tests/run_goldens.py

The command must exit with status 0. Any divergence between actual and expected output marks a regression and blocks merge.

Golden Suite and CI

• goldens.yml runs on every push and pull request.
• Always run python tests/run_goldens.py locally before publishing commits.
• Specification changes must ship with updated golden cases and documentation.

Adding New Scenarios

1. Create a .context capsule in tests/context/ with a descriptive name.
2. Execute python tests/run_goldens.py to produce the digest or error metadata.
3. Store the expected output under tests/outcomes/<name>.json.
4. Update docs/context_spec_1_2.md and docs/testing.md if behavior changed.
5. Open a pull request with the execution trace, references to ADRs when relevant, and proof of a green CI run.

Status and Next Steps

• Covered: resolver metadata, chunk payloads, TTL, confidence models, signature rotation and quorum, safe hints, TAB errors, attachment hash mismatch, external relation validation, JSON round-trip placeholder.
• In progress: pack/unpack flow, tag validation, external descriptors, public registry.
• Quarterly target: broaden negative scenarios and formalize the converter audit protocol.

Contributing

• Follow AGENTS.md and the Flagship bar: zero mocks, coverage >=85 percent, cyclomatic complexity <=10.
• Each pull request includes a design brief plus evidence (test logs, ADR links).
• Use organization GitHub Discussions for questions and design clarifications.

Support

• Issues are the preferred channel for requests or improvements.
• Contact: magraytlinov@gmail.com - core team replies within one business day.
• Context is pinned on the Modular Intellegence Systems overview as the entry point into the modular ecosystem.