Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
165 commits
Select commit Hold shift + click to select a range
a5922c8
plan(tui-render-cost): register the render-cost line and its six mile…
dbowm91 Nov 10, 2026
338ac65
perf(tui): stop the frame paying for work no interaction can spend
dbowm91 Nov 10, 2026
779f1e6
perf(tui): cache the file tree's filtered rows behind a proven key
dbowm91 Nov 10, 2026
3c991b6
perf(tui): build the chat rows once per content change, borrow the re…
dbowm91 Nov 10, 2026
0d0becb
perf(tui): remove the transcript's second copies and the collapsed pa…
dbowm91 Nov 10, 2026
229f33c
perf(tui): bound the editor's row read and seed its highlight prefix …
dbowm91 Nov 10, 2026
4f69126
perf(tui): make the styled wrap's span work linear
dbowm91 Nov 10, 2026
ba5d4f0
fix(tui): re-parse a streaming tail from the block before its last
dbowm91 Nov 10, 2026
70ff170
fix(tui): stop counting a word's tail run twice when wrapping
dbowm91 Nov 10, 2026
3913d22
fix(tui): keep the reasoning header after a searched prose part
dbowm91 Nov 10, 2026
0894cb7
fix(tui): leave room for a revised message's marker
dbowm91 Nov 10, 2026
7642b9e
test(tui): guard the tree's row inputs and record the line's fixes
dbowm91 Nov 10, 2026
fe98d86
feat(tui): land the five planning lines and close their residue
dbowm91 Nov 13, 2026
a976f7f
fix(sandbox): narrow the /proc and /var/run grants and qualify CI
dbowm91 Nov 13, 2026
dace38c
fix(sandbox): resolve /proc/self in the helper, and give the grants a…
dbowm91 Nov 13, 2026
6c7f7da
fix(ci): create the deps directory before copying the helper
dbowm91 Nov 13, 2026
e46eb67
fix(sandbox): keep the /proc aliases out of the mount view's identity…
dbowm91 Nov 13, 2026
3fb5af9
ci(sandbox): register the grant lane's filter
dbowm91 Nov 13, 2026
8a74f8c
ci(sandbox): watch the filter registry from the grant lane
dbowm91 Nov 13, 2026
0a5be70
test(sandbox): name the helper path when the fixture cannot launch it
dbowm91 Nov 13, 2026
ae34768
test(sandbox): run the grant fixtures on one test thread
dbowm91 Nov 13, 2026
a985afc
test(sandbox): probe for the managed helper instead of assuming one l…
dbowm91 Nov 13, 2026
f0c7c0d
chore(comment-budget): record the rounded-up ceiling for the fixture
dbowm91 Nov 13, 2026
c6898a4
test(sandbox): grant the device writes the product grants
dbowm91 Nov 14, 2026
6a9cae3
docs(sandbox): record the hosted grant evidence and the pre-existing …
dbowm91 Nov 14, 2026
34802dc
feat(tui): take the seven owner decisions
dbowm91 Nov 14, 2026
4848424
docs(closure): record the full run after the decisions
dbowm91 Nov 14, 2026
98740f1
Add provider stream stall diagnostics
dbowm91 Oct 31, 2026
a51032c
Record first provider stall diagnostic run
dbowm91 Oct 31, 2026
f0d8a13
Record reproduced provider stream idle timeout
dbowm91 Oct 31, 2026
b3df9cc
Allow observed provider event latency margin
dbowm91 Oct 31, 2026
cdb7846
Admit corrected provider stream confirmation run
dbowm91 Oct 31, 2026
771422f
Record corrected stream diagnostic run three
dbowm91 Oct 31, 2026
6609889
Admit fourth corrected stream diagnostic run
dbowm91 Oct 31, 2026
389cf41
Close S015 with live stream confirmation
dbowm91 Oct 31, 2026
1c3c9f7
Authorize DeepSeek H007 long-run loop
dbowm91 Oct 31, 2026
6e0e5b9
Record start of DeepSeek long-run attempt one
dbowm91 Oct 31, 2026
b2465d9
Correct initial S016 disk sample
dbowm91 Oct 31, 2026
12c7545
Record DeepSeek long-run failure and stream diagnosis
dbowm91 Oct 31, 2026
89c97ac
Add bounded SSE framing progress diagnostics
dbowm91 Oct 31, 2026
f4f2f76
Record S016 DeepSeek diagnostic run 2
dbowm91 Oct 31, 2026
f759fe9
Record S016 DeepSeek long run 3 start
dbowm91 Oct 31, 2026
a4ebc40
Record S016 DeepSeek long run 3 evidence
dbowm91 Oct 31, 2026
84ecada
Record S016 DeepSeek diagnostic run 4 start
dbowm91 Oct 31, 2026
3595ec3
Record S016 DeepSeek diagnostic run 4 evidence
dbowm91 Oct 31, 2026
b850e6e
Calibrate S016 H007 sustained task prompt
dbowm91 Oct 31, 2026
5467260
Record S016 DeepSeek long run 5 start
dbowm91 Oct 31, 2026
2cda675
Record S016 DeepSeek run 5 compaction failure
dbowm91 Oct 31, 2026
c7cbbbf
Record S016 DeepSeek job diagnostic start
dbowm91 Oct 31, 2026
a7b27a4
Record S016 diagnostic preflight refusal
dbowm91 Oct 31, 2026
b7c61dc
Record S016 DeepSeek job diagnostic run 7 start
dbowm91 Oct 31, 2026
b298a98
Preserve tool pairs during compatibility context compaction
dbowm91 Oct 31, 2026
bb6bbff
Record S016 DeepSeek long run 8 start
dbowm91 Oct 31, 2026
a180357
Record S016 DeepSeek run 8 job-sequencing failure
dbowm91 Oct 31, 2026
369a5dd
Record S016 DeepSeek job diagnostic run 9 start
dbowm91 Oct 31, 2026
1727228
Record S016 diagnostic and clarify long-run job IDs
dbowm91 Oct 31, 2026
2b7285e
Record S016 long qualification run 10 start
dbowm91 Oct 31, 2026
1f124ab
Record S016 run 10 and job sequence diagnostic
dbowm91 Oct 31, 2026
1da2ebc
Record S016 job-sequence diagnostic run 11 start
dbowm91 Oct 31, 2026
82a6ce4
Record S016 job sequence diagnostic and prompt v3
dbowm91 Oct 31, 2026
fb2f2e8
Record S016 long qualification run 12 start
dbowm91 Oct 31, 2026
34673d4
Record S016 run 12 and overlap diagnostic
dbowm91 Oct 31, 2026
6faebb7
Record S016 run 13 diagnostic start
dbowm91 Oct 31, 2026
b280bcf
Record S016 diagnostic preflight refusal
dbowm91 Oct 31, 2026
d7f3625
Record S016 run 14 diagnostic start
dbowm91 Oct 31, 2026
0f76287
Record S016 run 14 diagnostic and prompt sequence
dbowm91 Oct 31, 2026
eb83a50
Record S016 run 15 diagnostic start
dbowm91 Oct 31, 2026
4c51dc6
Record S016 run 15 and calibrate long prompt v4
dbowm91 Oct 31, 2026
aa2a453
Record S016 run 16 long qualification start
dbowm91 Oct 31, 2026
9a2bf5f
Record S016 run 16 and isolate exact job polling
dbowm91 Oct 31, 2026
701ec9b
Record S016 run 17 diagnostic start
dbowm91 Oct 31, 2026
a5fded2
Record S016 diagnostic budget issue and prompt v5
dbowm91 Oct 31, 2026
91e06e4
Record S016 run 18 long qualification start
dbowm91 Oct 31, 2026
025f751
Record S016 DeepSeek run 18 and job overlap diagnostic
dbowm91 Oct 31, 2026
7b11cc9
Authorize S016 run 19 job overlap diagnostic
dbowm91 Oct 31, 2026
9a7a4cc
Record S016 run 19 and background test prompt fix
dbowm91 Oct 31, 2026
53fa74b
Authorize final S016 DeepSeek long qualification
dbowm91 Oct 31, 2026
0c83987
Close S016 at exhausted live run authorization
dbowm91 Oct 31, 2026
927a292
fix(harness-lab): raise S016 request horizon ceiling
dbowm91 Nov 2, 2026
fc1aa27
docs(harness-lab): record raised-cap S016 run
dbowm91 Nov 2, 2026
fbb1080
fix(agent): account for full provider context in long runs
dbowm91 Nov 2, 2026
d39b44c
docs(harness-lab): record S016 diagnosis and prompt v8
dbowm91 Nov 2, 2026
1dbdc95
fix(harness-lab): compact long DeepSeek sessions before context exhau…
dbowm91 Nov 2, 2026
54db72f
docs(harness-lab): record context diagnosis and run 32
dbowm91 Nov 2, 2026
5744505
harness lab: diagnose short H007 horizon and calibrate probes
dbowm91 Nov 2, 2026
1844d07
harness lab: record attempt 34 and deepen workload probes
dbowm91 Nov 2, 2026
3359ec5
harness lab: diagnose premature end-turn with pending child
dbowm91 Nov 2, 2026
1db8e1d
harness lab: record completed workload and add bundle stage
dbowm91 Nov 2, 2026
7eafbec
harness lab: correct overlap scheduling and add catalog delta workload
dbowm91 Nov 2, 2026
d900913
harness lab: use argv-safe job and deepen merge workload
dbowm91 Nov 2, 2026
f3bbb93
harness lab: poll exact job immediately and add replication
dbowm91 Nov 2, 2026
d8620b5
harness lab: record final authorized long-run result
dbowm91 Nov 2, 2026
5560c28
harness lab: prepare Muse v17 requalification
dbowm91 Nov 3, 2026
043ddfb
harness lab: fix staged fixture and extend Muse task
dbowm91 Nov 3, 2026
ad2209e
harness lab: calibrate Muse workload beyond early task pass
dbowm91 Nov 3, 2026
9257c0c
harness lab: record Muse attempt two outcome
dbowm91 Nov 3, 2026
a978b2e
harness lab: correct Muse run three regressions
dbowm91 Nov 3, 2026
f37ab4d
harness lab: avoid blocking Muse job waits
dbowm91 Nov 3, 2026
e4d1c88
harness lab: align stall cancellation with tool progress
dbowm91 Nov 3, 2026
638cca6
harness lab: target slow debug qualification setup
dbowm91 Nov 3, 2026
96ee1de
harness lab: record Muse attempt 7 diagnostics
dbowm91 Nov 3, 2026
9617cf5
harness lab: document Muse attempt 8 verifier miss
dbowm91 Nov 3, 2026
9671ff8
harness lab: target Muse attempt 9 acceptance gaps
dbowm91 Nov 3, 2026
5507cdd
harness lab: checkpoint S019 Muse requalification
dbowm91 Nov 12, 2026
82336c5
harness lab: attribute failed job status safely
dbowm91 Nov 12, 2026
ce3f98f
harness lab: project event measurement causes
dbowm91 Nov 12, 2026
beadc35
fix H007 continuation diagnostic projection
dbowm91 Nov 12, 2026
8d230d8
harness lab: correct continuation diagnostics
dbowm91 Nov 12, 2026
26e2c33
harness lab: reconcile interrupted attempt 64
dbowm91 Nov 13, 2026
a1db46d
harness lab: add S052 attribution diagnostics
dbowm91 Nov 13, 2026
101dc3d
harness lab: close S052 from attempt 66 diagnostics
dbowm91 Nov 13, 2026
6f9ccbd
harness lab: record attempt 67 outcome
dbowm91 Nov 13, 2026
8f03736
Record H007 attempt 68 evidence
dbowm91 Nov 13, 2026
91cad7d
Correct attempt 68 layout interpretation
dbowm91 Nov 13, 2026
269cdc8
Preserve failed test submit error class in safe evidence
dbowm91 Nov 13, 2026
433ec75
harness-lab: add verifier-directed H007 recovery
dbowm91 Nov 14, 2026
79272a4
test: update compaction request fixture
dbowm91 Nov 14, 2026
d0b9e88
docs: require exact-head H007 binary pair
dbowm91 Nov 14, 2026
7c6d86b
harness-lab: reconcile interrupted attempt 71
dbowm91 Nov 14, 2026
44f6f9a
docs: clarify attempt 72 supervision signals
dbowm91 Nov 14, 2026
b9d1248
Merge corrective/s016-raised-request-horizon-20261102 into the five-l…
dbowm91 Nov 14, 2026
a8417b8
fix: clear the four reconciliation defects the hosted lanes found
dbowm91 Nov 14, 2026
443e5ca
fix(tool): assert the description wording the incoming line chose
dbowm91 Nov 14, 2026
ebc9bf0
fix(tool-advisor): move the receipt canary with the wording it guards
dbowm91 Nov 14, 2026
ad7a389
test(agent-loop): assert the continuation bound in the wording that c…
dbowm91 Nov 14, 2026
2d457a8
diag: make the two undiagnosable CI failures report their own cause
dbowm91 Nov 14, 2026
526dd79
fix(install): resolve the profile directory that owns deps, not just …
dbowm91 Nov 14, 2026
e6ecf1d
diag: classify a failed job from the observation the harness actually…
dbowm91 Nov 14, 2026
acf035f
diag: name the executable whose directory the helper refusal derived
dbowm91 Nov 14, 2026
fa61d05
fix(install): resolve the profile directory structurally, not by prob…
dbowm91 Nov 14, 2026
45db7d6
docs: polish user-facing guides
dbowm91 Nov 15, 2026
9d195ce
fix: resolve sandbox CI qualification failures
dbowm91 Nov 16, 2026
f34641f
test: preserve bounded CI failure diagnostics
dbowm91 Nov 16, 2026
e9cb7fe
fix: address Linux sandbox and H006 CI failures
dbowm91 Nov 16, 2026
95e3987
Fix Linux sandbox CI and H006 job observation
dbowm91 Nov 16, 2026
267091c
Fix constrained sandbox test setup
dbowm91 Nov 16, 2026
05299e2
Add bounded H006 managed-test diagnostics
dbowm91 Nov 16, 2026
fa3d8a5
Refine bounded H006 compiler diagnostics
dbowm91 Nov 16, 2026
369f78e
Emit bounded H006 Cargo diagnostics
dbowm91 Nov 16, 2026
d860473
Count bounded H006 Cargo diagnostic events
dbowm91 Nov 16, 2026
e9691af
Capture bounded H006 Cargo error context
dbowm91 Nov 16, 2026
38c62f8
Report bounded H006 log sizes
dbowm91 Nov 16, 2026
9cce009
Report bounded H006 working-directory scope
dbowm91 Nov 16, 2026
2f8bf92
Classify bounded H006 Cargo failure details
dbowm91 Nov 16, 2026
acf60d4
Identify bounded H006 Cargo error origin
dbowm91 Nov 16, 2026
2dc1a59
Classify bounded H006 compiler setup errors
dbowm91 Nov 16, 2026
fbebbdd
Keep H006 Cargo target inside test grant
dbowm91 Nov 16, 2026
e33636a
Forward contained Cargo target to test runner
dbowm91 Nov 16, 2026
1cc4444
Classify bounded H006 Cargo failures
dbowm91 Nov 16, 2026
e122b88
Keep Cargo temporary files inside test grant
dbowm91 Nov 16, 2026
d096ae4
Split H006 filesystem diagnostic classes
dbowm91 Nov 16, 2026
a0a3677
Add bounded H006 write-path diagnostics
dbowm91 Nov 16, 2026
73ad5d3
Classify bounded Cargo write subroles
dbowm91 Nov 16, 2026
d7c55fd
chore: update embedding.md
dbowm91 Nov 16, 2026
97c39f2
docs: reconcile ADR identities after rebase
dbowm91 Nov 17, 2026
1d2e6ee
reconcile corrective S016 branch content
dbowm91 Nov 17, 2026
c5de15c
fix(ci): re-bank comment budget after reconciliation
dbowm91 Nov 17, 2026
28fd5b0
test(harness): isolate missing-helper preflight from disk
dbowm91 Nov 17, 2026
7532ae7
fix(harness): ignore live product state in revision snapshots
dbowm91 Nov 17, 2026
cb4ba8f
test(harness): make H006 write failures decidable
dbowm91 Nov 18, 2026
1d0d8cd
test(harness): trace H006 Linux EXDEV syscalls
dbowm91 Nov 18, 2026
9d3d5e1
test(harness): project H006 Cargo target subroles
dbowm91 Nov 18, 2026
637fc10
test(harness): report H006 syscall path relationships
dbowm91 Nov 18, 2026
3c03f62
test(harness): capture H006 Cargo mount relationships
dbowm91 Nov 18, 2026
6cb5d52
Add bounded H006 filesystem labels
dbowm91 Nov 18, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
89 changes: 89 additions & 0 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -102,9 +102,23 @@ jobs:
- name: codegg-core boundary guard
run: ./scripts/check-core-boundary.sh

# The sandbox guards protect the silent containment defects (dropped
# profile thread, widened child ceiling, degraded uncontained launch,
# unregistered backend, Seatbelt deny order, hardlinked deny refusal,
# durable-payload containment rebuild). Both run with their self-tests:
# a guard that cannot fail on an injected defect is not evidence.
- name: Sandbox contract guard (self-test)
run: python3 scripts/check_sandbox_contract.py --self-test

- name: Sandbox contract guard
run: python3 scripts/check_sandbox_contract.py

- name: Sandbox policy wiring guard (self-test)
run: python3 scripts/check_sandbox_policy_wiring.py --self-test

- name: Sandbox policy wiring guard
run: python3 scripts/check_sandbox_policy_wiring.py

# M007-D: the daemon owns the command catalog (ADR-0052). Both guards
# run here with their self-tests: a guard that cannot fail on an
# injected defect is not evidence.
Expand All @@ -120,6 +134,21 @@ jobs:
- name: Frontend action set guard
run: python3 scripts/check_frontend_action_set.py

- name: Storage catalog guard (self-test)
run: python3 scripts/check_storage_catalog.py --self-test

- name: Storage catalog guard
run: python3 scripts/check_storage_catalog.py

- name: Tool catalog guard (self-test)
run: python3 scripts/check_tool_catalog.py --self-test

- name: Tool catalog guard
run: python3 scripts/check_tool_catalog.py

- name: Project catalog invariants guard
run: python3 scripts/check_project_catalog_invariants.py

- name: ADR identity and link guard
run: python3 scripts/check_adr_integrity.py

Expand Down Expand Up @@ -248,24 +277,42 @@ jobs:
- name: TUI project authority guard
run: python3 scripts/check_tui_project_authority.py

- name: TUI editor text-authority guard self-test
run: python3 scripts/check_tui_editor_text_authority.py --self-test

- name: TUI editor text-authority guard
run: python3 scripts/check_tui_editor_text_authority.py

- name: TUI editor key-authority guard self-test
run: python3 scripts/check_tui_editor_key_authority.py --self-test

- name: TUI editor key-authority guard
run: python3 scripts/check_tui_editor_key_authority.py

- name: TUI editor leaf-authority guard self-test
run: python3 scripts/check_tui_editor_leaf_authority.py --self-test

- name: TUI editor leaf-authority guard
run: python3 scripts/check_tui_editor_leaf_authority.py

- name: TUI editor completion-authority guard
run: python3 scripts/check_tui_editor_completion_authority.py

- name: TUI editor apply-path guard self-test
run: python3 scripts/check_tui_editor_apply_path.py --self-test

- name: TUI editor apply-path guard
run: python3 scripts/check_tui_editor_apply_path.py

- name: TUI editor prefix-state guard self-test
run: python3 scripts/check_tui_editor_prefix_state.py --self-test

- name: TUI editor prefix-state guard
run: python3 scripts/check_tui_editor_prefix_state.py

- name: TUI editor doc-drift guard self-test
run: python3 scripts/check_tui_editor_doc_drift.py --self-test

- name: TUI editor doc-drift guard
run: python3 scripts/check_tui_editor_doc_drift.py

Expand Down Expand Up @@ -319,6 +366,24 @@ jobs:
- name: TUI render cost guard
run: python3 scripts/check_tui_render_cost.py

- name: TUI render-invariant baseline self-test
run: python3 scripts/check_tui_render_invariant_baseline.py --self-test

- name: TUI render-invariant baseline guard
run: python3 scripts/check_tui_render_invariant_baseline.py

- name: Session replay fixture self-test
run: python3 scripts/check_session_replay_fixtures.py --self-test

- name: Session replay fixture guard
run: python3 scripts/check_session_replay_fixtures.py

- name: Request ledger writer self-test
run: python3 scripts/check_request_ledger_writers.py --self-test

- name: Request ledger writer guard
run: python3 scripts/check_request_ledger_writers.py

- name: HTTP route disposition guard
run: python3 scripts/check_http_route_disposition.py

Expand All @@ -343,6 +408,15 @@ jobs:
- name: Config merge coverage guard
run: python3 scripts/check_config_merge_coverage.py

- name: Config merge coverage guard self-test
run: python3 scripts/check_config_merge_coverage.py --self-test

- name: Config provenance guard self-test
run: python3 scripts/check_config_provenance.py --self-test

- name: Config provenance guard
run: python3 scripts/check_config_provenance.py

- name: Reasoning intent fidelity self-test
run: python3 scripts/check_reasoning_intent_fidelity.py --self-test

Expand Down Expand Up @@ -442,6 +516,14 @@ jobs:
- name: Shell policy test-safety fixtures
run: python3 scripts/test_shell_policy_test_safety.py

# An escaped defect gets a dated postmortem naming the guard that now
# prevents it, and this guard fails when that claim stops being true.
- name: Postmortem guard self-test
run: python3 scripts/check_postmortems.py --self-test

- name: Postmortem guard
run: python3 scripts/check_postmortems.py

- name: Build CodeGG for native Harness Lab job lifecycle test
run: |
cargo build --locked --bin codegg --bin codegg-sandbox-helper
Expand Down Expand Up @@ -591,6 +673,13 @@ jobs:
# home directory, and the root stays outside every workspace root the
# tests grant.
XDG_CACHE_HOME: ${{ runner.temp }}/codegg-sweep-cache
# A qualification lane, not a best-effort one. Without this marker a
# runner that loses its Landlock backend *skips* the sandbox
# fixtures, so a regression turns the suite green; with it, the skip
# arms fail and the ABI-6 assertion runs. A pre-6.12 kernel fails
# that assertion rather than skipping it, so a green run here now
# implies ABI >= 6 and a present backend.
CODEGG_SANDBOX_QUALIFICATION: "1"
run: |
if [ "${{ steps.live.outputs.live_required }}" = "true" ]; then
echo "Live Eggwork qualification included (relevant change or main push)."
Expand Down
89 changes: 89 additions & 0 deletions .github/workflows/linux-sandbox-grants-native.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,89 @@
name: Linux sandbox grant qualification

# The narrowing's own lane. `tool_read_allowances` no longer grants `/proc` or
# `/var/run` as trees, and these fixtures are the evidence for that: another
# process's `/proc/<pid>` entry is out of reach, the command's own
# `/proc/self/exe` still reads, the toolchain still runs, and removing the
# narrow `/proc` entries breaks a probe. They need a real Landlock host, so
# they cannot run anywhere else.
#
# The marker makes the lane a qualification rather than a best effort: a
# runner without a Landlock backend fails instead of skipping, and a pre-6.12
# kernel fails the ABI-6 probe rather than passing silently.

on:
workflow_dispatch:
pull_request:
paths:
- '.github/workflows/linux-sandbox-grants-native.yml'
- 'src/bin/codegg-sandbox-helper.rs'
- 'src/security/sandbox.rs'
- 'src/security/sandbox/**'
- 'tests/security_permission/sandbox_landlock.inc'
- 'tests/security_permission/mod.rs'
# The filter registry is an input to this lane: an unregistered step is
# refused, and a registry entry that drifts from the command is a
# failure rather than a silently green subset.
- 'docs/security-test-filters.toml'
push:
paths:
- '.github/workflows/linux-sandbox-grants-native.yml'
- 'src/bin/codegg-sandbox-helper.rs'
- 'src/security/sandbox.rs'
- 'src/security/sandbox/**'
- 'tests/security_permission/sandbox_landlock.inc'
- 'tests/security_permission/mod.rs'
# The filter registry is an input to this lane: an unregistered step is
# refused, and a registry entry that drifts from the command is a
# failure rather than a silently green subset.
- 'docs/security-test-filters.toml'

permissions:
contents: read

concurrency:
group: linux-sandbox-grants-native-${{ github.ref }}
cancel-in-progress: true

jobs:
grants:
runs-on: ubuntu-24.04
timeout-minutes: 20
env:
CARGO_BUILD_JOBS: 4
# A qualification lane, not a best-effort one: the skip arms fail when
# this is set, and the ABI-6 probe is asserted rather than attempted.
CODEGG_SANDBOX_QUALIFICATION: '1'
steps:
- uses: actions/checkout@v4

- uses: dtolnay/rust-toolchain@stable

- uses: Swatinem/rust-cache@v2
with:
cache-on-failure: true

- name: Record the Landlock posture this lane depends on
run: |
uname -a
printf 'LSM list: '
cat /sys/kernel/security/lsm 2>/dev/null || true
printf 'kernel release: '
uname -r

# Constrained Linux integration tests resolve this managed sidecar from
# the target profile directory shared with their Cargo `deps` binaries.
- name: Build trusted sandbox helper
run: |
cargo build --locked --bin codegg-sandbox-helper
mkdir -p target/debug/deps
cp target/debug/codegg-sandbox-helper target/debug/deps/codegg-sandbox-helper

- name: Sandbox grant fixtures
run: |
# The fixtures launch the helper as a separate process and resolve it
# beside the profile directory, so the step states what it built
# rather than leaving a missing sidecar to look like a test failure.
ls -l target/debug/codegg-sandbox-helper target/debug/deps/codegg-sandbox-helper
python3 scripts/check_test_filter_execution.py --run -- \
cargo test --locked --test security_permission sandbox_landlock -- --nocapture --test-threads=1
45 changes: 40 additions & 5 deletions .github/workflows/shared-checkout-macos-qualification.yml
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,7 @@ on:
- 'src/bin/codegg-sandbox-helper.rs'
- 'tests/shared_checkout_c006.rs'
- 'tests/macos_native_shared_checkout_containment.rs'
- 'tests/sandbox_containment.rs'
- 'tests/security_permission/**'
pull_request:
paths:
- '.github/workflows/shared-checkout-macos-qualification.yml'
Expand All @@ -52,7 +52,7 @@ on:
- 'src/bin/codegg-sandbox-helper.rs'
- 'tests/shared_checkout_c006.rs'
- 'tests/macos_native_shared_checkout_containment.rs'
- 'tests/sandbox_containment.rs'
- 'tests/security_permission/**'
workflow_dispatch:

permissions:
Expand Down Expand Up @@ -93,7 +93,9 @@ jobs:
# The constrained launch resolves this managed sidecar from the target
# profile directory shared with its Cargo `deps` binaries.
- name: Build sandbox helper for the qualification
run: cargo build --locked --bin codegg-sandbox-helper
run: |
cargo build --locked --bin codegg-sandbox-helper
test -x target/debug/codegg-sandbox-helper

# The mediation: the placement, the group authorization, the permit's
# mutual exclusion, the C005 preflight, the profile narrowing and the
Expand All @@ -118,14 +120,36 @@ jobs:
#
# `sandbox_containment` is a member of the `security_permission` family
# (M004 consolidated the root integration tests into families), so the
# target is the family and the member name is the filter. The filter
# keeps the lane's scope to the one member it names.
# target is the family and the member name is the filter.
#
# C005: `CODEGG_SANDBOX_QUALIFICATION=required` is C004's strict-mode
# marker. Unset — or set to `optional` — keeps the skip that lets a
# developer on a host without a containment backend run the suite; any
# other value makes the member fail with the platform's own reason. A
# lane whose entire purpose is the native backend must not go green
# from a skip, so it sets the marker: a macOS release that removed
# `sandbox-exec` now fails here instead of reporting a green skip.
- name: Sandbox containment suite
env:
CODEGG_SANDBOX_TEST_HELPER: ${{ github.workspace }}/target/debug/codegg-sandbox-helper
CODEGG_SANDBOX_QUALIFICATION: required
run: |
set -o pipefail
cargo test --locked -p codegg --test security_permission sandbox_containment -- --nocapture --test-threads=1 \
| tee "$RUNNER_TEMP/shared-checkout-macos/sandbox_containment.log"

# C005: the lane ran only `sandbox_containment`, so the members that
# assert the resolved sandbox profile is threaded into the registry and
# that the host admission manifest is honoured never ran on hosted
# macOS. Same family, same filter form.
- name: Sandbox policy wiring and host admission suites
run: |
set -o pipefail
cargo test --locked -p codegg --test security_permission sandbox_policy_wiring -- --nocapture --test-threads=1 \
| tee "$RUNNER_TEMP/shared-checkout-macos/sandbox_policy_wiring.log"
cargo test --locked -p codegg --test security_permission host_admission_manifest -- --nocapture --test-threads=1 \
| tee "$RUNNER_TEMP/shared-checkout-macos/host_admission_manifest.log"

- name: Record the qualification summary
if: always()
run: |
Expand All @@ -146,6 +170,17 @@ jobs:
echo '```text'
cat "$RUNNER_TEMP/shared-checkout-macos/macos_native_shared_checkout_containment.log" 2>/dev/null || echo "(not run)"
echo '```'
echo
echo "### Sandbox containment suite (strict mode)"
echo '```text'
cat "$RUNNER_TEMP/shared-checkout-macos/sandbox_containment.log" 2>/dev/null || echo "(not run)"
echo '```'
echo
echo "### Sandbox policy wiring and host admission suites"
echo '```text'
cat "$RUNNER_TEMP/shared-checkout-macos/sandbox_policy_wiring.log" 2>/dev/null || echo "(not run)"
cat "$RUNNER_TEMP/shared-checkout-macos/host_admission_manifest.log" 2>/dev/null || echo "(not run)"
echo '```'
} >> "$GITHUB_STEP_SUMMARY"

- name: Upload the qualification evidence
Expand Down
5 changes: 4 additions & 1 deletion .opencode/skills/agent/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,10 @@ instructions into the candidate. See `architecture/agent.md`.
`src/agent/team.rs` inbox is removed; the `task` tool's board/claim/release/
handoff-request actions expose bounded same-group coordination facts and
advisory intent only. They cannot message, cancel, interrupt, or reassign
another run. Derive the action set from `src/tool/task.rs`.
another run. Read-only `status`/bounded `wait` on a root child are owned by
the originating session across turns; a delegated caller must be that run's
direct parent. Mutating controls remain turn-scoped for root runs. Derive
the action set from `src/tool/task.rs`.
8. **Peer hints stay untrusted and push-only.** Deliver only allowlisted,
bounded host observations at provider boundaries; recheck after provider
completion and refuse a returned tool batch when new peer facts or a gap
Expand Down
8 changes: 4 additions & 4 deletions AGENTS.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ Rust 1.89+, edition 2021.

```bash
scripts/verify.sh quick # canonical sanity: rustfmt, the generated-asset
# checks, 84 guard scripts, and a workspace check
# checks, 94 guard scripts, and a workspace check
scripts/verify.sh full # quick + clippy (-D warnings) + nextest workspace sweep
# (profile ci) + bounded feature pass: nextest -p codegg
# --features server,plugins,lsp-test-support --profile subset
Expand Down Expand Up @@ -111,12 +111,12 @@ scripts/verify.sh scoped [base-ref] # quick's guards + only the tests the chang
## Change-triggered guards

`verify.sh quick` runs the routine subset: rustfmt, the two generated-asset
checks, 84 guard scripts, and a workspace `cargo check`. CI
checks, 94 guard scripts, and a workspace `cargo check`. CI
(`.github/workflows/ci.yml`) is one bounded `verify` job: the same generated-asset
checks, 70 guard scripts, fmt, clippy, and the workspace test sweep.
checks, 80 guard scripts, fmt, clippy, and the workspace test sweep.

CI is a strict **subset** of `quick` — it runs nothing that quick does not.
As measured on 2026-10-27, quick runs **14 guards CI omits**: the client and
As measured on 2026-11-11, quick runs **14 guards CI omits**: the client and
desktop boundaries, `check_daemon_listing_authority`, `check_first_party_skills`,
`check_network_intent_owner_wiring`, `check_no_hardcoded_selection_colours`,
`check_eggwork_target_routing`, `check_openai_endpoint_composition`, and the
Expand Down
1 change: 1 addition & 0 deletions Cargo.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

Loading
Loading