Files
marqdo-cli/docs/evidence/2026-11-20-local-gate.md
cflmy b6fc18233d feat(sandbox): implement probe-gated Linux strict isolation (ADR-0021)
Strict mode now re-executes the CLI inside a Bubblewrap supervisor with
user/mount/PID/net/UTS/IPC namespaces, no-new-privs, all capabilities dropped,
a read-only /project, private /proc and /dev, rlimits, output capture and
wall-clock process-group cleanup. The child verifies PID 1, NoNewPrivs, empty
CapEff, an unshared uid map and loopback-only networking before project code.

- PlatformCapabilities::detect runs a real bwrap probe; no probe, no strict
- strict L1 supports pure Marqdo + Math; Foreign/ABI/worker/kernel/grants/trace
  are rejected with exit 2 before sandboxing
- forged MARQDO_STRICT_ACTIVE is rejected; missing bwrap fails closed
- add adversarial helper, 7-test escape/resource matrix, seeded security lane
  script and self-hosted security workflow
- executed G-C lane as uid 1000 in a disposable Ubuntu 26.04 VM: 7/7 PASS
  plus positive CLI and refusal matrix; evidence in docs/evidence
- ADR-0021, spec/security.md, STATUS/COMPATIBILITY/README/CHANGELOG updated;
  ADR-0019 remains for non-Linux/fail-closed rules
2026-11-20 15:01:49 +08:00

2.8 KiB

Local gate evidence — 2026-11-20

Code commit tested: 2cf9ecedd590f57cd8a816d26dbe19417ddfed81. Workspace version: 0.1.0 (no tag). Host: Ubuntu 26.04 container, Rust 1.100.0, x86_64.

This is local evidence. It is not a GitHub Actions URL and does not replace the native Windows/macOS lanes. See COMPATIBILITY.md for the honest matrix.

Commands and results

cargo fmt --all -- --check                             PASS
cargo clippy --workspace --all-targets -- -D warnings  PASS
cargo test --workspace                                 PASS (46 test binaries)
cargo build --release --locked -p marqdo-cli           PASS

Notable suites in the run:

conformance_ast                 4 tests PASS (LF canonical + CRLF byte fixture)
integration_cli                23 tests PASS
plugin_worker                   4 tests PASS
marqdo-foreign provider         8 tests PASS
analysis_trace / tool_protocol  PASS
package_env                     PASS

Windows checkout reproduction

A fresh clone with core.autocrlf=true was made from this tree:

golden_has_CR False
crlf_fixture_has_CRLF True
cargo test -p marqdo-cli --test conformance_ast -> 4 passed

Converting one golden to CRLF before the fix reproduced the original "0 line(s) differ" symptom; the new diagnostic reported the first differing byte and classified it as line endings only.

Release candidate and clean-install smoke

The release binary was packaged with specs, examples, licenses, changelog and compatibility matrix, then extracted into a fresh temporary directory. The SHA-256 file verified before extraction. Executed from the extracted tree:

marqdo --version                     marqdo 0.1.0
marqdo --help                        exit 0
check examples/factorial.mq.md       exit 0
run examples/hello.mq.md             你好,Marqdo
call factorial 阶乘 --arg n=5        120
check multi-module --emit analysis   exit 0
analysis_consumer.py                 exit 0 (functions=2)
run multi-module --trace-file        exit 0
trace_consumer.py                    exit 0 (events=7)
run packages-multi-version           exit 0
run python-foreign --allow-foreign   28.274333882308138
run hello --mode strict              exit 1, refusal message
tool install/run/remove              exit 0

Environment limits observed

  • The process is root inside a container with no effective capabilities (capsh: Current: =) and no-new-privs=1.
  • setpriv --reuid=65534 fails (setresuid: Operation not permitted), so a true non-root run could not be performed on this host.
  • unshare (user/mount/PID/net) and bwrap fail with EPERM; cgroup2 is mounted read-only.
  • Therefore G-C could not be evaluated on that container host. It was later executed in a disposable Ubuntu 26.04 VM; see 2026-11-20-strict-gc.md and docs/security-gate.md.