feat(maintain): add promote_check task to ontos maintain

[ohjonathan]

Summary

Adds promote_check (order 45) to the ontos maintain pipeline. Runs ontos promote --check non-interactively to report documents ready for promotion from L0/L1 to L2.

Changes

• maintain.py: New registered task _task_promote_check at order 45
• Ontos_Agent_Instructions.md: "Maintain Ontos" section updated (8 → 9 tasks)
• Ontos_Manual.md: Numbered task list updated (8 → 9 tasks)
• test_maintain.py: Updated registry order assertion + 2 new tests

Design Notes

• ontos scaffold was not added separately because migrate_untagged (order 10) already calls find_untagged_files + _run_scaffold_command(apply=True)
• promote_check uses the existing --check mode which is non-interactive and read-only

Tests

All 22 maintain tests pass.

---

Antigravity, powered by Claude

[ohjonathan]

Review: 3-agent team (Alignment, Code Quality, Adversarial)

Verdict: REQUEST CHANGES (test fixes only — production code is sound)

---

Alignment Review — APPROVE

PR is well-aligned with project goals. All conventions followed (commit style, branch naming, task registration pattern, test naming). Docs are consistent across Manual, Agent Instructions, and registry test. Design decisions verified in code:

• migrate_untagged already covers scaffold ✓
• promote --check is genuinely read-only (returns before any SessionContext or commit()) ✓

---

Code Quality Review — COMMENT (2 Major in test quality)

Production code is correct and follows established patterns perfectly. Issues are confined to test quality.

---

Adversarial Review — REQUEST CHANGES (1 High, 2 Medium)

Core safety claim verified: check=True is provably read-only. However both tests are fundamentally flawed.

---

Consolidated Findings

Sev	Finding
HIGH	test_promote_check_task_reports_promotable_docs scans the real project (cwd), not the test fixture — provides zero signal
MEDIUM	test_promote_check_dry_run_skips_scan monkeypatches __wrapped__ which doesn't exist; raising=False silently swallows the failure, making the safety net inert
MEDIUM	No failure-path test for exit_code != 0
LOW	Session log missing trailing newline
LOW	Pre-existing typo "documents find" → "documents found" in promote.py:216

---

Implementation Prompt

Three fixes needed, all in tests/commands/test_maintain.py:

Fix 1: test_promote_check_task_reports_promotable_docs scans real project, not fixture

_run_promote_command calls find_project_root() which resolves from cwd, not from the test's tmp_path. The test creates fixture docs but the task scans whatever project is at the current working directory.

Fix: Add monkeypatch to the test signature and call monkeypatch.chdir(tmp_path) before invoking the task. Then assert on the message content (e.g., "1 documents" or similar) to verify the fixture was actually scanned — not just result.status == "success".

Fix 2: test_promote_check_dry_run_skips_scan monkeypatch is inert

The monkeypatch targets _task_promote_check.__wrapped__, but register_maintain_task doesn't use functools.wraps, so __wrapped__ doesn't exist. The raising=False silently swallows this, making the safety net a no-op.

Fix: Replace the monkeypatch target with ontos.commands.promote._run_promote_command. Remove raising=False. Example:

monkeypatch.setattr(
    "ontos.commands.promote._run_promote_command",
    lambda _opts: (_ for _ in ()).throw(AssertionError("dry-run should not call promote")),
)

Fix 3: Add failure-path test

The _fail() branch (when exit_code != 0) is untested.

def test_promote_check_reports_failure(tmp_path, monkeypatch):
    _init_project(tmp_path)
    monkeypatch.chdir(tmp_path)
    ctx = _build_context(tmp_path, quiet=True)
    monkeypatch.setattr(
        "ontos.commands.promote._run_promote_command",
        lambda _opts: (1, "Document load failed"),
    )
    result = _task_promote_check(ctx)
    assert result.status == "failed"
    assert "Document load failed" in result.message

Verification

python3 -m pytest tests/commands/test_maintain.py -v --tb=short

---

🤖 3-agent review by Claude Code

[ohjonathan]

Round 2 Review: All 3 Agents APPROVE

Review	Round 1	Round 2
Alignment	APPROVE	APPROVE
Code Quality	COMMENT (2 Major)	APPROVE (4/4 resolved)
Adversarial	REQUEST CHANGES (1 High, 2 Medium)	APPROVE (all verified fixed)

All Round 1 findings resolved

Finding	Sev	Status
Test scans real project, not fixture	HIGH	Fixed — monkeypatch.chdir(tmp_path) + message assertion
Dry-run monkeypatch targets non-existent __wrapped__	MEDIUM	Fixed — targets _run_promote_command directly
No failure-path test	MEDIUM	Fixed — new test_promote_check_reports_failure added
Typo "documents find"	NIT	Fixed — "documents found"

Residual (non-blocking, cosmetic)

• LOW: "1 documents found" — no singular/plural handling (pre-existing in promote.py)
• LOW: Session log missing trailing newline

Test suite

23/23 maintain tests pass. No new issues introduced by the fix commit.

Consensus: APPROVE for merge.

🤖 3-agent review by Claude Code