feat(curator): umbrella-first prompt, inherit parent config, unbounded iterations

Based on three live test runs against 346 agent-created skills on the
author's own setup (~6.5 min, opus-4.7, 86 API calls), the curator
prompt needed three sharpenings before it consistently produced real
umbrella consolidation instead of passive audit output:

**Umbrella-first framing.** The original 'decide keep/patch/archive/
consolidate' framing lets opus default to 'keep' whenever two skills
aren't byte-identical. The new prompt explicitly tells the reviewer
that pairwise distinctness is the wrong bar — the right question is
'would a human maintainer write this as N separate skills, or one
skill with N labeled subsections?' Expect 10-25 prefix clusters; merge
each into an umbrella via one of three methods.

**Three concrete consolidation methods.** (a) Merge into an existing
umbrella (patch the broadest skill, archive siblings); (b) Create a
new umbrella SKILL.md (skill_manage action=create); (c) Demote
session-specific detail into references/, templates/, or scripts/
under the umbrella via skill_manage action=write_file, then archive
the narrow sibling. This matches the support-file vocabulary the
review-prompt side already uses (PR #17213).

**Two observed bailouts pre-empted:** 'usage counters are zero so I
can't judge' (rule 4: judge on content, not use_count) and 'each has
a distinct trigger' (rule 5: pairwise distinctness is the wrong bar).

**Config-aware parent inheritance.** _run_llm_review() was building
AIAgent() without explicit provider/model, hitting an auto-resolve
path that returned empty credentials → HTTP 400 'No models provided'
against OpenRouter. Fork now inherits the user's main provider and
model (via load_config + resolve_runtime_provider) before spawning —
runs on whatever the user is currently on, OAuth-backed or
pool-backed included.

**Unbounded iteration ceiling.** max_iterations=8 was way too low for
an umbrella-build pass over hundreds of skills. A live pass takes
50-100 API calls (scanning, clustering, skill_view'ing candidates,
patching umbrellas, mv'ing siblings). Raised to 9999 — the natural
stopping criterion is 'no more clusters worth processing', not an
arbitrary tool-call budget.

**Tests updated:** test_curator_review_prompt_has_invariants accepts
DO NOT / MUST NOT and drops 'keep' from the required-verb set (the
umbrella-first prompt correctly deemphasizes 'keep' as a first-class
decision label since passive keep-everything is the failure mode
being prevented). Added test_curator_review_prompt_is_umbrella_first
asserting the umbrella framing, class-level thinking, references/
+ templates/ + scripts/ support-file mentions, and the 'use_count
is not evidence of value' pre-emption. Added
test_curator_review_prompt_offers_support_file_actions asserting
skill_manage action=create and action=write_file are both named.

**Live validation on author's setup:**
- Run 1 (old prompt): 3 archives, stopped after surveying — typical passive outcome
- Run 2 (consolidation prompt): 44 archives, 3 patches, surfaced the 50-skill mlops reorg duplicate bug but didn't umbrella
- Run 3 (this prompt): 249 archives + 18 new class-level umbrellas created, reducing agent-created skills from 346 → 118 with every archived skill's content preserved as references/ under its umbrella. Pinned skill untouched. Full report in PR description.

tests/agent/test_curator.py

@@ -359,13 +359,19 @@ def test_state_atomic_write_no_tmp_leftovers(curator_env):
 def test_curator_review_prompt_has_invariants():
     """Core invariants must be in the review prompt text."""
     from agent.curator import CURATOR_REVIEW_PROMPT
-    assert "MUST NOT" in CURATOR_REVIEW_PROMPT
+    assert "MUST NOT" in CURATOR_REVIEW_PROMPT or "DO NOT" in CURATOR_REVIEW_PROMPT
     assert "bundled" in CURATOR_REVIEW_PROMPT.lower()
     assert "delete" in CURATOR_REVIEW_PROMPT.lower()
     assert "pinned" in CURATOR_REVIEW_PROMPT.lower()
-    # Must mention the decisions the reviewer can make
-    for verb in ("keep", "patch", "archive", "consolidate"):
+    # Must describe the actions the reviewer can take. The exact vocabulary
+    # has tightened over time (the umbrella-first prompt drops 'keep' as a
+    # first-class decision verb, since passive keep-everything is the
+    # failure mode the prompt is trying to avoid), but the core merge /
+    # archive / patch trio must remain callable.
+    for verb in ("patch", "archive"):
         assert verb in CURATOR_REVIEW_PROMPT.lower()
+    # Must mention consolidation (possibly via "merge" or "consolidat")
+    assert "consolidat" in CURATOR_REVIEW_PROMPT.lower() or "merge" in CURATOR_REVIEW_PROMPT.lower()
 
 
 def test_curator_review_prompt_points_at_existing_tools_only():
@@ -400,6 +406,44 @@ def test_curator_does_not_instruct_model_to_pin():
     )
 
 
+def test_curator_review_prompt_is_umbrella_first():
+    """The curator prompt must push umbrella-building / class-level thinking,
+    not pair-level 'are these two the same?' analysis."""
+    from agent.curator import CURATOR_REVIEW_PROMPT
+    lower = CURATOR_REVIEW_PROMPT.lower()
+    # Must frame the task as active umbrella-building, not a passive audit.
+    assert "umbrella" in lower, (
+        "must use UMBRELLA framing — the class-first abstraction the curator "
+        "is designed to produce"
+    )
+    # Must tell the reviewer not to stop at pair-level distinctness.
+    assert "class" in lower, "must reference class-level thinking"
+    # Must cover the three consolidation methods explicitly
+    assert "references/" in CURATOR_REVIEW_PROMPT, (
+        "must name references/ as a demotion target for session-specific content"
+    )
+    # templates/ and scripts/ make the umbrella a real class-level skill
+    assert "templates/" in CURATOR_REVIEW_PROMPT
+    assert "scripts/" in CURATOR_REVIEW_PROMPT
+    # Must say the counter argument: usage=0 is not a reason to skip
+    assert "use_count" in CURATOR_REVIEW_PROMPT or "counter" in lower, (
+        "must pre-empt the 'usage counters are zero, I can't judge' bailout"
+    )
+
+
+def test_curator_review_prompt_offers_support_file_actions():
+    """Support-file demotion (references/templates/scripts) must be one of
+    the three consolidation methods, alongside merge-into-existing and
+    create-new-umbrella."""
+    from agent.curator import CURATOR_REVIEW_PROMPT
+    # skill_manage action=write_file is how references/ are added to an
+    # existing skill — this is the create-adjacent action the curator needs
+    # to demote narrow siblings without touching their SKILL.md.
+    assert "write_file" in CURATOR_REVIEW_PROMPT
+    # Must offer creating a brand-new umbrella when no existing one fits
+    assert "action=create" in CURATOR_REVIEW_PROMPT or "create a new umbrella" in CURATOR_REVIEW_PROMPT.lower()
+
+
 
 def test_cli_unpin_refuses_bundled_skill(curator_env, capsys):
     """hermes curator unpin must refuse bundled/hub skills too (matches pin)."""