a7-py is pre-1.0. Security fixes are made on master unless a release branch is created later.

a7-py is not a sandbox.

The compiler reads A7 source, emits Zig, and the verification scripts build and execute native binaries. Compiled A7 programs can do anything the generated program and host runtime allow. Only compile and run A7 source you trust.

Before release or broad testing, run:

./run_all_tests.sh
uv build
uvx --from pip-audit==2.10.0 pip-audit --strict
uvx --from bandit==1.9.4 bandit -r a7 scripts main.py -q --skip B404,B603
uv run python scripts/check_no_secrets.py
(cd site && bun run build)
(cd site && bun audit --audit-level=moderate)

The full gate includes parser/tokenizer tests, semantic tests, Zig backend tests, example runtime verification, debug/release artifact builds, error-stage checks, docs style checks, and full pytest.

• No sandboxing for compiled programs.
• No ownership or borrow-style lifetime model yet.
• fall is supported only as the final direct statement of a non-final match case; other placements are rejected during semantic validation.
• Built-in stdlib imports are virtual and not yet unified with file-based module semantics.
• File-backed imports are restricted to configured module search paths; absolute paths and parent-directory traversal are rejected.
• Zig example checks are runtime smoke tests, not a proof for all possible programs.
• The current release workflow builds package distributions and attaches them to draft GitHub releases; it does not publish to a package registry.
• Tag-created release artifacts receive GitHub artifact attestations; verify them with gh attestation verify <file> --repo code5717/a7-py before publishing or consuming a release.
• Workflow actions are pinned to immutable commits resolved from current upstream release tags on 2026-05-08.
• Dependency audits check known advisories, not unknown vulnerabilities.
• The committed-secrets check is pattern- and filename-based and should be treated as a guard, not a complete data-loss prevention system.

These limitations are tracked in docs/STATUS.md and docs/RELEASE.md.

Open a private security advisory or contact the repository owner before opening public issues for exploitable crashes, unsafe generated code, or release pipeline compromise.