agent-desktop/.github/workflows/ci.yml
Lahfir 18daaa8215
feat: windows toolchain, ci & com bootstrap (sub-phase 2.1) (#112)
* docs: correct the windows sub-phase facts research disproved

Sixteen replacements, no annotations. The document is the product's source of
truth and now reads true.

The COM bootstrap is split by consumer. CoInitializeEx is right for the CLI and
wrong for the cdylib, which fails RPC_E_CHANGED_MODE against any host thread
already in an STA and whose per-thread balance can never be released from a Drop
on another thread; the library path uses CoIncrementMTAUsage, whose cookie is
thread-agnostic. MTA is recorded as a requirement rather than a preference,
because the single-threaded apartment can prevent handler removal outright.

Two things the sub-phase asserted were not reachable. Its private-file seam named
a boundary that does not exist: every write site is in core with no adapter
handle, and core may not depend on the platform crate. The document now states
that constraint and leaves the mechanism to the plan. Its CI paragraph claimed
all runners enforce clippy, isolation, and the size cap; only the macOS job does,
and the other two lanes each run a single lib invocation.

The event invariant said handlers are drained on the dedicated MTA thread. They
are not: delivery arrives on several UIA-owned threads at once, the registering
worker among them and the main thread never, so handler state has to be safe for
concurrent delivery.

Runner registration gains the hardening a public repository requires, and gains
the measurement it already owed the ledger: it closes the deferred RDP row rather
than merely documenting session isolation. The DPI bootstrap records that
awareness cannot be verified by read-back, since the query API has no enumerant
for the value being set.

The ledger's hunk index moves with them, 37 to 40. Hunks backed by research
rather than by a probe are labelled as such, so an external citation cannot pass
itself off as a measurement.

* docs: move the windows runner registration to the sub-phase that needs it

Sub-phase 2.1 was to register a self-hosted interactive Windows runner that
nothing in 2.1 through 2.11 uses. On a public repository that is a persistently
registered runner sitting idle behind ten sub-phases, and GitHub's own guidance
is that self-hosted runners should almost never be used for public repositories
at all. Registration moves to 2.12, the first sub-phase whose gate needs a real
desktop, and it takes the hardening requirements with it: dispatch-only
triggering, a written fork-PR approval policy, and an explicit
ephemeral-versus-persistent decision.

The deferred RDP row moves with it. Its closure was never documentation -- it is
a measurement that only exists once a second, non-console session does, so 2.12
now owns closing it rather than 2.1 owning a promise it could not keep.

Two probe scripts hardcoded the old closure sub-phase in text they emit. The
scripts were corrected and their probes re-run, so the captures were regenerated
rather than edited by hand.

* docs: plan the windows toolchain, ci, and com bootstrap sub-phase

Eight units covering the CI lane extension, Win32 bindings in the platform
crate, the COM and DPI bootstrap, the adapter session, the permission probe,
and a private-file layer rebuilt from scratch against sub-phase 2.0's measured
evidence.

Three reviewers found what the first draft got wrong. It installed both the
apartment and the private-file seam from ad_init, which the FFI header labels
optional and whose own doc promises it initialises nothing -- so any host that
skipped it would have run every private write through the unhardened default
with nothing to distinguish the two states. The install moves to the site the
macOS arm already uses, where a consumer cannot skip it.

It also planned a layer that validates security state it never authors. The
deleted implementation at least built an owner-only protected descriptor at
creation; the rebuild now does too, and rejects reparse points per component
the way the unix path rejects symlinks.

The ACE validation order was still wrong in the same shape as the defect it
exists to prevent: checking the type before forming the SID pointer still
dereferences the header at an offset nothing has bounds-checked. The order is
now extent-first, with the ACE count treated as advisory rather than trusted.

The seam routes at core's five private-file primitives rather than at a list of
call sites, which covers the lock and liveness paths a list had missed and
needs no call-site edits at all.

* docs: cut acl validation from the windows private-file design

Reverses the ancestor-walk decision on measurement. A plain leaf under the user
profile already inherits SYSTEM, Administrators and the user with no Users
entry, all inherited and none explicit, so the walk would verify a property
Windows already guarantees. Unix has to author its permission because the
default there is world-readable; Windows does not, because the default is
already private.

The only principal the walk would additionally exclude is another
administrator, who holds SeTakeOwnershipPrivilege and can seize any object
whatever its DACL. Building a chain-walking validator against someone who can
bypass it costs more than it buys, and the cost is specific: parsing ACEs is
what the previous layer did, and its AceSize handling is the defect that took
the layer down. Removing the parser removes the bug class rather than fixing
it, so no successor module exists and a test asserts the ACE symbols never come
back.

What survives is the control with real unix parity and a real attack behind it:
per-component reparse-point rejection. That is an integrity control, not a
confidentiality one -- a junction planted on the path redirects where the
product writes, which no permission on the intended destination prevents.

Cross-admin isolation is now answered rather than open: out of scope, for the
same reason macOS never raised the question.

* test: resolve the test binary per platform via CARGO_BIN_EXE

* feat: add target-gated windows-sys bindings to the windows crate

* feat: route private-file primitives through a core-defined platform seam

* feat: establish com apartment and dpi awareness at unskippable call sites

* feat: report windows permissions truthfully from a host-independent seam

* ci: extend the windows lane to the full package surface

* feat: give the windows adapter a session that owns apartment lifetime

* feat: rebuild windows private-file hardening behind the core seam

* test: pin the inherited-acl assumption module-free on a real profile leaf

* feat: install the windows private-file seam at both unskippable consumers

* fix: scope the temp lease per write and fold duplicated windows helpers

* fix: route windows session and trace directory creation through the private-file seam

* fix: scope the com bootstrap to adapter commands and harden review-flagged tests

* test: cover the owner, locality, replace-acl, and sweep-race branches of the windows hardening

* fix: pin the private-file directory chain to close the junction-swap window

* docs: settle the user-output seam exclusion and pin it with a test

* refactor: split the trace file plumbing and ffi adapter tests into sibling modules
2026-07-27 18:35:51 -07:00

561 lines
24 KiB
YAML
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

name: CI
on:
push:
branches: [main, master]
pull_request:
workflow_dispatch:
# Cancel in-progress runs for the same ref so stale PR builds don't burn minutes.
concurrency:
group: ci-${{ github.ref }}
cancel-in-progress: true
# Deny all permissions at workflow level; each job declares only what it needs.
permissions: {}
env:
CARGO_TERM_COLOR: always
RUST_BACKTRACE: 1
jobs:
fmt:
name: Format
runs-on: ubuntu-latest
timeout-minutes: 10
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- run: rustup component add rustfmt
- run: cargo fmt --all -- --check
- name: Check E2E scripts without running native automation
run: |
scripts/check-bash3-compat.sh
shellcheck -x -e SC2034,SC2154 tests/e2e/run.sh tests/e2e/electron-live.sh tests/e2e/permission-contract.sh tests/e2e/guard-command.sh scripts/*.sh .githooks/pre-commit
PYTHONPYCACHEPREFIX=/tmp/agent-desktop-pycache python3 -m py_compile tests/e2e/*.py
PYTHONPYCACHEPREFIX=/tmp/agent-desktop-pycache python3 -m unittest discover -s tests/e2e -p 'test_*.py'
- name: Validate GitHub Actions workflows
run: go run github.com/rhysd/actionlint/cmd/actionlint@v1.7.11
msrv:
name: Rust 1.89 MSRV
runs-on: ubuntu-latest
timeout-minutes: 20
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- run: rustup toolchain install 1.89.0 --profile minimal
- run: cargo +1.89.0 check --locked -p agent-desktop-core -p agent-desktop-linux -p agent-desktop --all-targets
platform-check:
name: Native check (${{ matrix.platform }})
runs-on: ${{ matrix.os }}
timeout-minutes: 25
permissions:
contents: read
strategy:
fail-fast: false
matrix:
include:
- platform: Linux
os: ubuntu-latest
package: agent-desktop-linux
- platform: Windows
os: windows-latest
package: agent-desktop-windows
- platform: macOS
os: macos-latest
package: agent-desktop-macos
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install pinned Rust toolchain
run: rustup show
- name: Check native adapter and binary
run: cargo check --locked -p agent-desktop-core -p ${{ matrix.package }} -p agent-desktop --all-targets
test:
name: Test
runs-on: macos-latest
timeout-minutes: 30
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: |
rustup show
rustup component add clippy
- name: Verify macOS Bash 3.2 compatibility
run: /bin/bash scripts/check-bash3-compat.sh
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Cache build artifacts
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: target/
key: ${{ runner.os }}-build-ci-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}-${{ hashFiles('crates/**/*.rs', 'src/**/*.rs', 'tests/**/*.rs') }}
restore-keys: |
${{ runner.os }}-build-ci-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}-
${{ runner.os }}-build-ci-
- name: Check dependency isolation
run: |
TREE=$(cargo tree --locked -p agent-desktop-core --edges normal,build)
if printf '%s\n' "$TREE" | grep -E 'agent-desktop-(macos|windows|linux)'; then
echo "FAIL: core crate depends on platform crates"
exit 1
fi
echo "OK: core crate has no platform dependencies"
- name: Check release metadata consistency
run: scripts/check-release-consistency.sh
- name: Enforce Rust source rules
run: scripts/check-rust-file-size.sh
- name: Clippy
run: cargo clippy --locked -p agent-desktop-core -p agent-desktop-macos -p agent-desktop -p agent-desktop-ffi --all-targets -- -D warnings
- name: Core and macOS unit tests
run: scripts/cargo-test-isolated-home.sh test --locked -p agent-desktop-core -p agent-desktop-macos --lib
- name: Locator benchmark adapter contract
run: scripts/cargo-test-isolated-home.sh test --locked -p agent-desktop-core --example locator_benchmark
- name: Deterministic permission acceptance contract
run: bash tests/e2e/permission-contract.sh
- name: Binary command tests
run: scripts/cargo-test-isolated-home.sh test --locked -p agent-desktop
# The FFI crate ships integration harnesses under crates/ffi/tests/ that
# exercise the C-ABI from outside the crate (raw extern "C" decls, enum
# fuzzing, out-param zeroing, last-error lifetimes). The workspace --lib
# run skips them, so wire them in explicitly so regressions fail PR CI.
- name: FFI integration tests
run: scripts/cargo-test-isolated-home.sh test --locked -p agent-desktop-ffi --tests
- name: Build stripped release binary
run: |
cargo build --locked --release -p agent-desktop
cargo build --locked --release -p agent-desktop-macos --bin agent-desktop-macos-helper
- name: Verify shipped binary version flag
run: |
PACKAGE_ID=$(cargo pkgid -p agent-desktop)
EXPECTED_VERSION=${PACKAGE_ID##*@}
EXPECTED_OUTPUT="agent-desktop $EXPECTED_VERSION"
ACTUAL_OUTPUT=$(target/release/agent-desktop --version)
if [ "$ACTUAL_OUTPUT" != "$EXPECTED_OUTPUT" ]; then
echo "FAIL: expected '$EXPECTED_OUTPUT', got '$ACTUAL_OUTPUT'"
exit 1
fi
- name: Check shipped binary size
run: |
SIZE=$(stat -f%z target/release/agent-desktop)
HELPER_SIZE=$(stat -f%z target/release/agent-desktop-macos-helper)
LIMIT=$((15 * 1024 * 1024))
echo "Binary size: $(du -sh target/release/agent-desktop | cut -f1)"
echo "Helper size: $(du -sh target/release/agent-desktop-macos-helper | cut -f1)"
if [ "$SIZE" -gt "$LIMIT" ] || [ "$HELPER_SIZE" -gt "$LIMIT" ]; then
echo "FAIL: shipped executable exceeds 15MB limit (binary=${SIZE}, helper=${HELPER_SIZE} bytes)"
exit 1
fi
echo "OK: shipped executables are within the 15MB per-file limit"
- name: FFI cdylib build (release-ffi profile)
run: cargo build --locked --profile release-ffi -p agent-desktop-ffi
- name: FFI dylib-adjacent helper discovery smoke
run: scripts/ci-ffi-helper-discovery-smoke.sh
- uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
with:
node-version: '24'
- name: NPM package contents
run: |
node --test tests/npm/*.test.js
node scripts/check-npm-package.js
- name: NPM wrapper smoke
run: scripts/ci-npm-wrapper-smoke.sh
test-linux:
name: Test (Linux)
runs-on: ubuntu-latest
timeout-minutes: 20
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: rustup show
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Core and Linux unit tests
run: scripts/cargo-test-isolated-home.sh test --locked -p agent-desktop-core -p agent-desktop-linux --lib
test-windows:
name: Test (Windows)
runs-on: windows-latest
timeout-minutes: 45
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: rustup show
# Core resolves its data dir via HOME then USERPROFILE (crates/core/src/refs.rs
# fn home_dir), and env assigned inside one run: block does not survive into
# the next step, so the isolation is hoisted here and exported through
# GITHUB_ENV for every later step. CARGO_HOME and RUSTUP_HOME must be resolved
# from the real USERPROFILE first (rustup/cargo default them to
# %USERPROFILE%\.rustup and %USERPROFILE%\.cargo) so repointing USERPROFILE at
# the empty test home doesn't strand the installed toolchain.
# ORIGINAL_USERPROFILE feeds the trailing profile-isolation guard.
- name: Isolate HOME for the whole lane
shell: pwsh
run: |
$cargoHome = if ($env:CARGO_HOME) { $env:CARGO_HOME } else { Join-Path $env:USERPROFILE '.cargo' }
$rustupHome = if ($env:RUSTUP_HOME) { $env:RUSTUP_HOME } else { Join-Path $env:USERPROFILE '.rustup' }
$testHome = Join-Path $env:RUNNER_TEMP ("agent-desktop-test-home-" + [guid]::NewGuid())
New-Item -ItemType Directory -Path $testHome | Out-Null
Add-Content -Path $env:GITHUB_ENV -Value "ORIGINAL_USERPROFILE=$env:USERPROFILE"
Add-Content -Path $env:GITHUB_ENV -Value "CARGO_HOME=$cargoHome"
Add-Content -Path $env:GITHUB_ENV -Value "RUSTUP_HOME=$rustupHome"
Add-Content -Path $env:GITHUB_ENV -Value "HOME=$testHome"
Add-Content -Path $env:GITHUB_ENV -Value "USERPROFILE=$testHome"
# HOME and USERPROFILE now point into RUNNER_TEMP for every remaining step
# (including the post-job cache save), so ~ would miss the real registry; the
# cache paths must go through the hoisted CARGO_HOME instead.
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
${{ env.CARGO_HOME }}/registry/index/
${{ env.CARGO_HOME }}/registry/cache/
${{ env.CARGO_HOME }}/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Cache build artifacts
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: target/
key: ${{ runner.os }}-build-ci-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}-${{ hashFiles('crates/**/*.rs', 'src/**/*.rs', 'tests/**/*.rs') }}
restore-keys: |
${{ runner.os }}-build-ci-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}-
${{ runner.os }}-build-ci-
# Widened relative to the macOS-lane check: dev edges are included (a Win32
# binding crate hiding in dev-dependencies otherwise passes clean) and the
# resolution is additionally pinned to the MSVC target.
- name: Check dependency isolation
shell: pwsh
run: |
$pattern = 'agent-desktop-(macos|windows|linux)|\bwindows-sys\b|\bwindows\b|\bwinapi\b'
foreach ($target in @('', 'x86_64-pc-windows-msvc')) {
$arguments = @('tree', '--locked', '-p', 'agent-desktop-core', '--edges', 'normal,build,dev')
if ($target) { $arguments += @('--target', $target) }
$tree = cargo @arguments
if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
$violations = @($tree | Select-String -Pattern $pattern)
if ($violations.Count -gt 0) {
$violations | ForEach-Object { Write-Host $_.Line }
Write-Host "FAIL: core crate pulls platform or Win32 binding crates (target: '$target')"
exit 1
}
}
Write-Host 'OK: core crate has no platform dependencies on normal, build, or dev edges'
# The dependency check cannot see hand-written externs, so reject the raw
# Win32 surface at source level. The allowlist is the pair of portable no-op
# directory-sync shims in private_file.rs; the count is pinned so a third
# windows conditional cannot ride in on the allowlisted file either.
- name: Check core source for Win32 leakage
shell: pwsh
run: |
$sources = Get-ChildItem -Path crates/core/src -Recurse -Filter *.rs
$forbidden = @($sources | Select-String -Pattern 'extern "system|std::os::windows|\bwinapi\b|\bwindows_sys\b')
if ($forbidden.Count -gt 0) {
$forbidden | ForEach-Object { Write-Host "$($_.Path):$($_.LineNumber): $($_.Line.Trim())" }
Write-Host 'FAIL: core source declares a raw Win32 surface'
exit 1
}
$allowed = (Resolve-Path 'crates/core/src/private_file.rs').Path
$cfgMatches = @($sources | Select-String -Pattern 'cfg\(windows\)|cfg\(target_os = "windows"\)' -AllMatches)
$outside = @($cfgMatches | Where-Object { $_.Path -ne $allowed })
if ($outside.Count -gt 0) {
$outside | ForEach-Object { Write-Host "$($_.Path):$($_.LineNumber): $($_.Line.Trim())" }
Write-Host 'FAIL: windows cfg outside the private_file.rs allowlist'
exit 1
}
$shimCount = [int]($cfgMatches | ForEach-Object { $_.Matches.Count } | Measure-Object -Sum).Sum
if ($shimCount -ne 2) {
Write-Host "FAIL: expected exactly 2 windows cfg shims in private_file.rs, found $shimCount"
exit 1
}
Write-Host 'OK: core source keeps windows conditionals to the 2 allowlisted shims'
- name: Clippy
run: cargo clippy --locked -p agent-desktop-core -p agent-desktop-windows -p agent-desktop -p agent-desktop-ffi --all-targets -- -D warnings
- name: Core and Windows unit tests
run: cargo test --locked -p agent-desktop-core -p agent-desktop-windows --lib
- name: Binary command tests
run: cargo test --locked -p agent-desktop
# The FFI crate ships integration harnesses under crates/ffi/tests/ that a
# --lib run skips, and the Windows-gated reachability coverage landing there
# must execute on this runner, so wire them in explicitly.
- name: FFI integration tests
run: cargo test --locked -p agent-desktop-ffi --tests
- name: Build stripped release binary
run: cargo build --locked --release -p agent-desktop
# The macOS lane's stat -f%z is BSD-only and fails under Git Bash, so the
# size gate is native pwsh. There is no helper binary on Windows — the cap
# applies to the one shipped executable.
- name: Check shipped binary size
shell: pwsh
run: |
$binary = Get-Item target/release/agent-desktop.exe
$limit = 15MB
Write-Host "Binary size: $($binary.Length) bytes"
if ($binary.Length -gt $limit) {
Write-Host "FAIL: shipped executable exceeds 15MB limit (binary=$($binary.Length) bytes)"
exit 1
}
Write-Host 'OK: shipped executable is within the 15MB per-file limit'
# The isolation scenario this lane must prove: a test that writes a private
# artifact observes it under RUNNER_TEMP, never the runner profile.
- name: Guard profile isolation
shell: pwsh
run: |
if (-not $env:HOME -or -not $env:HOME.StartsWith($env:RUNNER_TEMP, [System.StringComparison]::OrdinalIgnoreCase)) {
Write-Host "FAIL: HOME is not under RUNNER_TEMP: $env:HOME"
exit 1
}
if (-not $env:USERPROFILE -or -not $env:USERPROFILE.StartsWith($env:RUNNER_TEMP, [System.StringComparison]::OrdinalIgnoreCase)) {
Write-Host "FAIL: USERPROFILE is not under RUNNER_TEMP: $env:USERPROFILE"
exit 1
}
$leak = Join-Path $env:ORIGINAL_USERPROFILE '.agent-desktop'
if (Test-Path $leak) {
Write-Host "FAIL: a test escaped the isolated HOME and wrote $leak"
exit 1
}
Write-Host 'OK: private artifacts stayed under RUNNER_TEMP'
ffi-python-smoke:
name: FFI Python Smoke
runs-on: macos-latest
timeout-minutes: 20
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: rustup show
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Build FFI dylib (stub adapter)
run: >
cargo build --locked --profile release-ffi
-p agent-desktop-ffi --features stub-adapter
- name: Locate dylib
id: dylib
run: |
DYLIB=$(find target/release-ffi -name 'libagent_desktop_ffi.*' \
\( -name '*.dylib' -o -name '*.so' -o -name '*.dll' \) | head -1)
if [ -z "$DYLIB" ]; then
echo "FAIL: dylib not found under target/release-ffi"
exit 1
fi
echo "path=$DYLIB" >> "$GITHUB_OUTPUT"
- uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
with:
python-version: '3.12'
- name: Run FFI smoke harness
env:
AD_EXPECT_STUB: "1"
run: >
python3 tests/ffi-python/smoke.py
"${{ steps.dylib.outputs.path }}"
crates/ffi/include/agent_desktop.h
# Gate 1 — Header-drift: the committed crates/ffi/include/agent_desktop.h must
# match what cbindgen 0.29.4 would generate from the current source. A stale
# header means a consumer compiled against it sees different symbols/layouts
# than the dylib ships. cbindgen --verify exits non-zero when there is a diff.
#
# The cbindgen version is pinned so a generator upgrade cannot produce a
# false-positive failure — only a real header change triggers this gate.
ffi-header-drift:
name: FFI Header Drift
runs-on: macos-latest
timeout-minutes: 20
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: rustup show
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Install pinned cbindgen 0.29.4
run: cargo install cbindgen --version 0.29.4 --locked
- name: Verify header matches source (cbindgen --verify)
run: >
cbindgen crates/ffi
--config crates/ffi/cbindgen.toml
--output crates/ffi/include/agent_desktop.h
--verify
# Gate 2 — Panic-trap guard: the release-ffi profile must keep panic="unwind"
# so the trap_panic boundary in the shipped dylib actually catches panics. A
# flip to panic="abort" silently defeats every catch_unwind in the cdylib and
# turns any Rust panic into an abort that kills the host process.
#
# This job proves the contract in two steps:
# 1. Assert the Cargo.toml profile setting directly (simpler and more
# reliable than a build.rs env check for a dual cdylib+rlib crate).
# 2. Build the actual feature-gated cdylib panic entrypoint, load it with
# dlopen from a C process, and require the exported ABI boundary to
# return ErrInternal instead of aborting the host.
ffi-panic-guard:
name: FFI Panic Guard
runs-on: macos-latest
timeout-minutes: 20
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: rustup show
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Assert release-ffi profile keeps panic=unwind
run: |
if ! grep -A5 '\[profile\.release-ffi\]' Cargo.toml | grep -q 'panic.*=.*"unwind"'; then
echo "FAIL: [profile.release-ffi] must set panic=\"unwind\" — a flip to"
echo " panic=\"abort\" defeats every trap_panic boundary in the dylib."
exit 1
fi
echo "OK: release-ffi profile retains panic=\"unwind\""
- name: Prove the shipped cdylib catches an exported-boundary panic
run: bash crates/ffi/tests/run_cdylib_panic_probe.sh
# Gate 3 — Stub-adapter passthrough tests: the Family-B command-backed
# entrypoints (ad_snapshot, ad_status, ad_wait, ad_execute_by_ref,
# ad_version) plus adapter lifecycle (ad_init / ad_destroy) and
# ad_check_permissions are exercised against the stub adapter. Each must
# return a structured PLATFORM_NOT_SUPPORTED envelope — or, for
# ad_check_permissions, the documented ErrPermDenied result. These tests are
# gated behind #[cfg(feature = "stub-adapter")] so the normal build is
# unaffected.
#
# Scope note: the ~35 Family-A entrypoints (ad_find, ad_execute_action,
# ad_list_windows, ad_screenshot, clipboard, notifications, etc.) are NOT
# covered by this passthrough job. Broader Family-A stub coverage is a
# documented follow-up.
#
# Per-target build coverage:
# - PR CI: this job runs on Linux and reaches the stub adapter directly to
# assert PLATFORM_NOT_SUPPORTED for every covered entrypoint, proving the
# non-macOS parity U10 targets.
# - Release: release.yml build-ffi matrix covers macOS×2 + Linux×2 +
# Windows×1 for the real cdylib. PR builds do not duplicate that
# cross-compile matrix to avoid bloating CI on every push.
ffi-passthrough:
name: FFI Stub-Adapter Passthrough
runs-on: ubuntu-latest
timeout-minutes: 20
permissions:
contents: read
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
- name: Install Rust toolchain
run: rustup show
- name: Cache cargo registry
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: ${{ runner.os }}-cargo-${{ hashFiles('Cargo.lock', 'Cargo.toml', 'crates/**/Cargo.toml', 'src/Cargo.toml') }}
restore-keys: ${{ runner.os }}-cargo-
- name: Run stub-adapter passthrough tests
run: >
scripts/cargo-test-isolated-home.sh test --locked -p agent-desktop-ffi
--features stub-adapter --test c_abi_passthrough