- 0033: Unified Entity State Model (lifecycle × health × delivery)
- 0034: Frontend Data & State Architecture (TanStack + SSE)
- 0035: Design System & Component Library (shadcn/Radix)
- 0039: Knowledge Substrate Minimal Interface (5th Branch manager)

Foundation for the governance ADR series (0041-0052).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Twelve governance ADRs establishing the governed-orchestration
substrate for lionagi:

- 0041: Immutable Evidence Nodes
- 0042: Task Certificate
- 0043: Governed Tool Declaration
- 0044: Tool Gates
- 0045: Break-Glass Protocol
- 0046: JIT Tool Grant
- 0047: Agent Charter
- 0048: Agent Segregation of Duties
- 0049: Log Tier Governance
- 0050: Operation Context
- 0051: Tool Registry Allowlists
- 0052: Policy Resolution

All ADRs use lionagi primitives (Element, Pile, Tool, Branch
managers) as integration points. Status: Proposed — pending
validation via the governance show.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Navigation aids for the ADR corpus — glossary of governance terms
and a reader's guide for navigating the ADR dependency graph.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Automated fix of MD032 (blanks-around-lists), MD040 (code-fence
language), MD058 (blanks-around-tables) across ADRs 0003, 0007,
0010, 0015, 0016, 0018, 0020-0023, 0027. Pure formatting.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…1189

CRIT-1: Delete 4 broken test files in tests/test_work/ that imported
non-existent classes (Form, WorkItemStatus, WorkerEngine, StepDef,
ChoiceSet, NumericRange, StringPattern). Keep only test_worker.py which
imports real classes from lionagi.work.worker. Canonical tests live in
tests/work/test_work_system.py (67 tests, all passing).

CRIT-2: Replace signal.SIGALRM timeout in engine._run_task with
concurrent.futures.ThreadPoolExecutor.future.result(timeout=...) which
is safe to call from any thread (SIGALRM raises ValueError in non-main
threads and is process-global, causing concurrent timeouts to clobber
each other).

CRIT-3: Add HANDLER_ALLOWED_MODULES allowlist to definition.py.
resolve_handler() now validates the module prefix against the allowlist
before importing, preventing arbitrary code execution (os.system,
subprocess.call, etc.) from YAML/JSON config.

MAJ-1: Add VALID_TRANSITIONS dict and WorkForm.transition_to() method
that validates status transitions and raises ValueError for illegal
moves (e.g., completed→draft). Add status guard in WorkEngine.submit()
to reject forms in submitted/completed/error status.

MAJ-2: Add ReDoS protection to Rule._check_pattern(). Enforce
REGEX_MAX_INPUT_LENGTH (10,000 chars) and run the match in a thread
with REGEX_MATCH_TIMEOUT (1s) via concurrent.futures, preventing
catastrophic backtracking from nested quantifier patterns.

MAJ-3: Replace _resolve_id's direct access to engine._lock and
engine._items with new public methods engine.get_item(id) and
engine.find_by_prefix(prefix).

MAJ-4: Guard the bare `from .mcp import ...` in cli/main.py with
try/except ImportError, assigning None fallbacks and guarding all
call sites.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- WorkEngine.submit_async(): mirror the _submittable status guard already
  present in submit() so async callers cannot enqueue completed/errored forms
- SchedulerEngine.find_by_prefix(): add missing method used by the CLI
  `li schedule pause/remove/resume <prefix>` path; without it those commands
  raise AttributeError at runtime

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…t re-fire bug

MAJ-5: Replace ThreadPoolExecutor context manager with manual lifecycle so
shutdown(wait=False, cancel_futures=True) is called on timeout — the old
`with` block called shutdown(wait=True) on __exit__, blocking the caller
for the full duration of catastrophic backtracking.

MAJ-6: In SchedulerEngine.mark_completed(), transition one-shot items
(cron_expr=None, interval_seconds=None) to STATUS_COMPLETED when
_compute_next_run() returns None, instead of leaving them STATUS_ACTIVE
and letting them re-fire immediately on every tick.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ThreadPoolExecutor cannot reliably timeout Python `re` catastrophic
backtracking because the GIL prevents the main thread from regaining
control. ProcessPoolExecutor runs the match in a separate process with
its own GIL, making kill-on-timeout effective.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…ut + kill

ProcessPoolExecutor cannot pickle re.Match results and cannot terminate
running workers. Switch to explicit multiprocessing.Process with Queue
for result passing (bool only, no Match object) and terminate()/kill()
for reliable cleanup on timeout.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>