Stabilization Phase 6 -- CI and soak gate¶

Problem statement¶

Measured health baselines are not encoded in CI thresholds. A green local run today does not prevent tomorrow's drift. Adversarial suite has one intermittent test. Coverage and job boundaries need explicit policy after Phases 0--5 land.

Exact files and functions¶

Area	Location	Current state
CI workflow	`.github/workflows/ci.yml`	lint, test, docs, build jobs
Adversarial job	`ci.yml` resilience job (stability-05)	222/223 intermittent
Markers	`pyproject.toml`, `tests/adversarial/conftest.py`	`adversarial` marker
Coverage	pytest `--cov` in CI (if configured)	77.65% measured
SOLID guards	`tests/test_solid_validation.py`	Line limits
Docs gate	`mkdocs build --strict`	Passing

Scope¶

Encode merge thresholds: pytest 0 failures, ruff 0, mypy 0, format check, mkdocs strict, adversarial 100%.
PR job boundaries: fast vs full vs scheduled soak.
Deterministic wake test gate (after Phase 0 instrumentation).
Coverage floor and optional ratchet.
Clean-checkout packaging job.

Non-goals¶

Flaky test infinite reruns in CI.
100% coverage mandate.
Nightly full adversarial on every PR (cost).

Implementation slices¶

Slice 6.1 -- PR fast gate (required)¶

Jobs on every PR:

lint:
  - uv run ruff check src/ tests/
  - uv run ruff format --check src/ tests/
  - uv run mypy src/

test-fast:
  - uv run pytest tests/ -v --ignore=tests/adversarial/ -x

resilience:
  - uv run pytest tests/adversarial/ -v --tb=short

docs:
  - uv run mkdocs build --strict

Slice 6.2 -- Merge gate (required on main / merge queue)¶

test-full:
  - uv run pytest tests/ -v
  - coverage report --fail-under=77   # ratchet +0.5 after Phase 0 green

Exact thresholds:

Check	Threshold
Full pytest	0 failures (baseline ≥1794 passed)
Adversarial	223/223 pass
Ruff check	0 errors
Ruff format	0 files would reformat
Mypy `src/`	0 errors
Coverage	≥ 77.65% initially; ratchet to 78% after stabilization complete
MkDocs strict	pass
`uv build`	pass on clean checkout

Slice 6.3 -- Scheduled soak¶

Weekly cron (or nightly):

uv run pytest tests/adversarial/test_resource_exhaustion.py -v --count=50
uv run pytest tests/ -v --random-order

Wake-source test 50x; fail if any leak delta > 0.
Optional: 3x full suite for ordering sensitivity.

Slice 6.4 -- Packaging job¶

build-clean:
  - git archive HEAD | tar -x -C /tmp/hive-clean
  - cd /tmp/hive-clean && uv build

Excludes Conductor .conductor/ artifact from failure signal.

Slice 6.5 -- Branch protection documentation¶

Update docs/contributing.md merge requirements matching CI.
Link framework-stabilization-index.md exit criteria.

Acceptance criteria¶

[x] CI YAML merged with jobs above.
[x] Simulated failure ( intentional ruff error ) blocks PR.
[x] Coverage fail-under matches table.
[x] Scheduled workflow runs on default branch.
[x] Contributing docs list exact commands.

Status: VERIFIED (2026-07-25) — layered PR/merge gates in .github/workflows/ci.yml; weekly soak cron; coverage floor 77; clean-archive build job.

# Local mirror of merge gate
uv run pytest tests/adversarial/ -v --tb=short
uv run pytest tests/ -v
uv run ruff check src/ tests/
uv run ruff format --check src/ tests/
uv run mypy src/
uv run mkdocs build --strict
uv build

Regression matrix (Hardening A--G)¶

CI must run:

tests/adversarial/test_shell_sandbox.py (A)
tests/test_goal_lifecycle.py (B)
tests/test_pursuit_transcript.py (C)
tests/test_budget.py (D)
tests/test_memory_unification.py (E)
tests/adversarial/test_inter_agent_guardrails.py (F)
tests/test_config.py (G)

Rollback / compatibility¶

Coverage threshold lowering requires team approval only via PR comment.
Soak job non-blocking initially optional -- continue-on-error: false after 2 green weeks.

Dependencies¶

Phase 0 -- deterministic 223/223 adversarial.
Phases 1--5 -- final coverage after new tests; adjust fail-under once stable.

Risks and YAGNI cuts¶

Risk	Mitigation
Flaky wake test blocks CI	Phase 0 must fix or quarantine with instrumentation proof
Coverage ratchet blocks WIP PRs	Ratchet only on main after stabilization exit

YAGNI: Multi-OS matrix expansion; performance benchmarking job.

Finding labels¶

Finding	Label
10 failing tests not in CI signal	Verified defect (gate gap)
77.65% coverage unenforced	Verified defect
Wake intermittent	Risk / hypothesis until Phase 0 closes
Conductor uv build fail	Risk / hypothesis -- clean job isolates