Skill: kill the lookup-loop time-sink (3 fixes from Codex transcript)

Root cause observed in a real Codex session: the agent burned ~3min
on `make venv` + 4 sequential `make diagram-lookup` queries trying
to find a non-existent ref-arch (OCI PostgreSQL accessed FROM GCP
via Cross-Cloud Interconnect). The catalog has 123 entries; if two
honest queries can't surface a topology, none exists — but nothing
in the skill said "stop." Three coordinated fixes:

1) Makefile: `make venv` is now idempotent.
   Marker file `.venv/.deps-installed` keyed off `requirements.txt`
   mtime — the second `make venv` in a row is a 26ms no-op (was
   1m53s on a Codex sandbox, paid every turn). Force a rebuild with
   `rm .venv/.deps-installed`.

2) kb/architecture-center/catalog.yaml: new `known_gaps` block.
   Surfaced by the lookup tool as a top-of-output banner whenever
   the (synonym-expanded) query tokens match any one of a gap's
   `triggers` token sets. Two gaps seeded from real engagements:

     - gcp-to-oci-native-services — OCI PostgreSQL/OKE/Cache/etc.
       reached FROM GCP via Cross-Cloud Interconnect (the
       Database@Google Cloud entries are the OPPOSITE direction).

     - newer-oci-services-without-icon — OCI Cache (Redis/Valkey)
       and OCI PostgreSQL post-v24.2-toolkit; no ref-arch, no
       toolkit icon, fall back to generic stencil + explicit label.

   Each gap's `notice:` is a copy-paste recommended composition so
   the agent doesn't have to invent a strategy.

3) tools/archcenter_pattern_lookup.py: detects + surfaces gaps.
   `lookup()` now returns `{"matches": ..., "gaps": ...}`. The
   text printer emits a "⚠ KNOWN GAP — <id>" banner with the
   notice block ABOVE the score-based top-K, so the agent sees
   the stop-and-compose signal before ever scrolling to results.
   YAML output mode mirrors the same shape under `gaps:`.

4) SKILL.md (option 2 step 1) + Codex copy via sync-skill.
   New explicit rule: "Lookup budget: max 2 queries — never loop."
   Documents the gap banner, the closest-3-then-ask fallback, and
   the prohibition against grasping at a 3rd refinement.

Verified manually:
  - GCP+postgresql query → gap fires
  - "redis ha multi-az" → newer-OCI gap fires
  - "exadata cross region data guard" → no false-positive gap

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.agents/skills/oci-deal-accelerator/SKILL.md

@@ -137,6 +137,7 @@ Pick a number, or just describe what you need.
      - **The cached `_template.yaml` is your YAML SCAFFOLD** — auto-extracted from the canonical `.drawio` by `tools/archcenter_drawio_to_template.py`. It carries Oracle's container geometry (region/vcn/ad/subnet bboxes), service positions, and edge waypoints in the `absolute_layout:` shape your spec uses. Copy it as the starting point, rename ids, fill in `type:` for services (the extractor leaves `type: TODO_identify` because the canonical icon TYPE is encoded in stencil bytes — pick the right alias from the renderer's TYPE_TO_ICON), and adapt to the customer's components. Lookup result surfaces it as `cached: yaml=<path>`.
      - **The cached `.drawio` is the visual source of truth** if the template extractor missed something — open it (drawio.exe, the SVG companion, or the XML) for the canonical look.
      - **Do NOT use `examples/` as a geometry source.** `examples/` is previous user output, not authoritative. The `_template.yaml` covers everything `examples/` was being abused for.
+     - **Lookup budget: max 2 queries — never loop.** If query 1 returns nothing convincing, run ONE refinement (broader phrasing or different keyword angle), then STOP. Re-querying the catalog 3+ times to find a "better" match is the dominant time-sink in this phase — the catalog has 123 entries indexed once; if two passes can't surface a topology match, none exists. Output of the lookup tool also surfaces a **`⚠ KNOWN GAP`** banner at the top whenever the query touches a documented gap (e.g. OCI-native services accessed FROM GCP via Cross-Cloud Interconnect, or OCI Cache/Redis/PostgreSQL as primary subject). When that banner appears, follow its **Recommended composition** verbatim instead of running more queries — it lists the primitives to combine. When NO gap banner appears AND no top-3 entry matches the topology, present the closest 3 to the user with: *"No exact ref-arch matches. Closest hits are X / Y / Z. I can either base the diagram on the closest of these (geometry transfers, service names don't), or compose from `<list specific slugs / patterns>`. Which?"* — let the user pick the chassis instead of guessing.
   2. **Pre-generation review.** Confirm the component list with the user (REQUESTED + TECHNICAL DEPENDENCIES per the whitelist).
   3. **Author the spec in `absolute_layout` shape** — this is REQUIRED. Do NOT use the legacy workload-driven shape (`tenancy → regions → compartments → services`); it does not resolve OCI icon stencils, does not run through the spec validator, and produces wireframe-looking output (rectangles with text instead of real OCI icons). Every container, service, label, and connection needs explicit `(x, y, w, h)`.
 

Makefile

@@ -19,11 +19,20 @@ help: ## Show this help
 
 VENV_PYTHON := $(shell command -v python3.12 2>/dev/null || command -v python3.11 2>/dev/null || command -v python3.10 2>/dev/null || echo python3)
 
-venv: ## Create virtual environment and install dependencies
+venv: .venv/.deps-installed ## Create or update venv (idempotent — no-op if requirements.txt unchanged)
-	@echo "Using $(VENV_PYTHON) to create venv..."
+
-	$(VENV_PYTHON) -m venv .venv
+# Marker file pattern: the venv only rebuilds when requirements.txt is
+# newer than the marker. A fresh `make venv` after a successful install
+# is an instant no-op (was 1m53s on Codex sandboxes — every turn).
+# To force a rebuild: `rm .venv/.deps-installed` (or `rm -rf .venv`).
+.venv/.deps-installed: requirements.txt
+	@if [ ! -x .venv/bin/python ]; then \
+	  echo "Using $(VENV_PYTHON) to create venv..."; \
+	  $(VENV_PYTHON) -m venv .venv; \
+	fi
 	.venv/bin/pip install --upgrade pip
 	.venv/bin/pip install -r requirements.txt
+	@touch .venv/.deps-installed
 	@echo "Virtual environment ready. Run: source .venv/bin/activate"
 	@echo "Tip: run 'make install-hooks' once to enable the pre-commit"
 	@echo "     hook that keeps SKILL.md in sync with the Codex copy."

SKILL.md

@@ -135,6 +135,7 @@ Pick a number, or just describe what you need.
      - **The cached `_template.yaml` is your YAML SCAFFOLD** — auto-extracted from the canonical `.drawio` by `tools/archcenter_drawio_to_template.py`. It carries Oracle's container geometry (region/vcn/ad/subnet bboxes), service positions, and edge waypoints in the `absolute_layout:` shape your spec uses. Copy it as the starting point, rename ids, fill in `type:` for services (the extractor leaves `type: TODO_identify` because the canonical icon TYPE is encoded in stencil bytes — pick the right alias from the renderer's TYPE_TO_ICON), and adapt to the customer's components. Lookup result surfaces it as `cached: yaml=<path>`.
      - **The cached `.drawio` is the visual source of truth** if the template extractor missed something — open it (drawio.exe, the SVG companion, or the XML) for the canonical look.
      - **Do NOT use `examples/` as a geometry source.** `examples/` is previous user output, not authoritative. The `_template.yaml` covers everything `examples/` was being abused for.
+     - **Lookup budget: max 2 queries — never loop.** If query 1 returns nothing convincing, run ONE refinement (broader phrasing or different keyword angle), then STOP. Re-querying the catalog 3+ times to find a "better" match is the dominant time-sink in this phase — the catalog has 123 entries indexed once; if two passes can't surface a topology match, none exists. Output of the lookup tool also surfaces a **`⚠ KNOWN GAP`** banner at the top whenever the query touches a documented gap (e.g. OCI-native services accessed FROM GCP via Cross-Cloud Interconnect, or OCI Cache/Redis/PostgreSQL as primary subject). When that banner appears, follow its **Recommended composition** verbatim instead of running more queries — it lists the primitives to combine. When NO gap banner appears AND no top-3 entry matches the topology, present the closest 3 to the user with: *"No exact ref-arch matches. Closest hits are X / Y / Z. I can either base the diagram on the closest of these (geometry transfers, service names don't), or compose from `<list specific slugs / patterns>`. Which?"* — let the user pick the chassis instead of guessing.
   2. **Pre-generation review.** Confirm the component list with the user (REQUESTED + TECHNICAL DEPENDENCIES per the whitelist).
   3. **Author the spec in `absolute_layout` shape** — this is REQUIRED. Do NOT use the legacy workload-driven shape (`tenancy → regions → compartments → services`); it does not resolve OCI icon stencils, does not run through the spec validator, and produces wireframe-looking output (rectangles with text instead of real OCI icons). Every container, service, label, and connection needs explicit `(x, y, w, h)`.
 

kb/architecture-center/catalog.yaml

@@ -18,6 +18,78 @@ last_verified: "2026-03-16"
 source: "https://docs.oracle.com/en/solutions/oracle-architecture-center/"
 entry_count: 123
 
+# =============================================================================
+# KNOWN GAPS — topologies the catalog does NOT cover well
+# =============================================================================
+#
+# Surfaced by tools/archcenter_pattern_lookup.py at the top of its output
+# whenever the user's query tokens match any one of the `triggers` token
+# sets below. Purpose: stop the lookup loop early when no canonical Oracle
+# ref-arch exists for what the user asked, and tell the agent how to
+# COMPOSE the topology from primitives instead of refining query phrasing.
+#
+# Schema per gap:
+#   id:        unique slug
+#   triggers:  list of token sets — fires when the expanded query contains
+#              EVERY token in at least ONE set (OR across sets, AND inside
+#              each set). Tokens are lowercased and synonym-expanded the
+#              same way as scoring tokens.
+#   notice:    multi-line string surfaced verbatim above the score-based
+#              top-K. Should explain WHY this is a gap and HOW to compose
+#              from existing primitives. Be concrete (slugs, alias names).
+#
+# Add a gap entry only after a real lookup loop wasted user time on a
+# topology with no canonical match. Speculation belongs elsewhere.
+
+known_gaps:
+
+  - id: gcp-to-oci-native-services
+    triggers:
+      - [gcp, postgresql]
+      - [gcp, redis]
+      - [gcp, mysql]
+      - [gcp, oke]
+      - [gcp, cache]
+      - [gcp, compute]
+      - [google, postgresql]
+      - [google, redis]
+      - [google, mysql]
+      - [google, oke]
+      - [google, cache]
+      - [google, compute]
+    notice: |
+      No canonical Oracle ref-arch covers OCI-native services (PostgreSQL,
+      OKE, Cache, etc.) accessed FROM Google Cloud via Cross-Cloud
+      Interconnect. The `Database@Google Cloud` entries below are the
+      OPPOSITE direction (Oracle DB running ON GCP, not GCP apps reaching
+      OCI-native services).
+
+      Recommended composition — STOP refining the lookup query, build from:
+        1. A simple VCN baseline (e.g. `start-with-an-oci-virtual-cloud-network`).
+        2. The OCI service in a private subnet behind DRG + FastConnect VC.
+        3. A `GCP <region>` container on the left with `cloud_icon: gcp_compute`
+           (or the relevant GCP service — the renderer maps to mxgraph.gcp2.*
+           the same way it does for AWS).
+
+  - id: newer-oci-services-without-icon
+    triggers:
+      - [redis]
+      - [valkey]
+      - [oci_cache]
+    notice: |
+      OCI Cache with Redis/Valkey is GA after the v24.2 toolkit shipped, so
+      no dedicated ref-arch exists in the catalog and no toolkit icon ships
+      either. Oracle's own multi-service ref archs that DO mention Redis
+      embed inline SVG over the generic database/cache stencil.
+
+      Recommended path — don't keep searching:
+        1. Pick the closest topology BY SHAPE (HA pair, primary+replica,
+           multi-AZ) from the catalog — geometry transfers, names don't.
+        2. Use `type: cache` (or `type: database`) with explicit
+           `label: "OCI Cache (Redis)"`. The renderer's alias table already
+           routes `type: redis` → generic stencil.
+        3. Reference: SKILL.md option 2 step 3 (newer-OCI-services convention).
+
 entries:
 
   # ===== DATABASE — MULTICLOUD (Database@Azure) =====

tools/archcenter_pattern_lookup.py

@@ -327,12 +327,42 @@ def _patterns_for(entry: dict) -> list[str]:
     return _build_patterns_index().get(entry.get("url", ""), [])
 
 
-def lookup(query: str, top: int = 5, expand_synonyms: bool = True) -> list[dict]:
+def _matched_gaps(query_tokens: set[str], gaps: list[dict]) -> list[dict]:
+    """Return every known gap whose triggers match the (expanded) query.
+
+    A gap fires when ANY one of its `triggers` token sets is a subset of
+    `query_tokens` — i.e. AND inside each set, OR across sets. This
+    matches the "stop searching, compose instead" intent: if even one
+    narrow trigger pair (e.g. `[gcp, postgresql]`) is in the query, the
+    catalog has nothing canonical to offer.
+    """
+    matched: list[dict] = []
+    for gap in gaps or []:
+        triggers = gap.get("triggers") or []
+        for trigger_set in triggers:
+            tokens_required = {t.lower() for t in trigger_set}
+            if tokens_required and tokens_required.issubset(query_tokens):
+                matched.append(gap)
+                break
+    return matched
+
+
+def lookup(query: str, top: int = 5, expand_synonyms: bool = True) -> dict:
+    """Score the catalog against `query` and return the top-K plus any
+    known-gap notices the query triggers.
+
+    Returns a dict shaped as:
+        {"matches": [<top-K entry dicts>], "gaps": [<matched gap dicts>]}
+
+    Callers that only care about matches can read `result["matches"]`.
+    """
     catalog = yaml.safe_load(CATALOG.read_text(encoding="utf-8"))
     entries = catalog.get("entries", [])
+    gaps_def = catalog.get("known_gaps", []) or []
     qt = _tokens(query)
     if expand_synonyms:
         qt = _expand_query_tokens(qt)
+    matched_gaps = _matched_gaps(qt, gaps_def)
     scored = []
     # Two-pass scoring: cheap fields first (title/tags/services/summary),
     # then enrich the top-K with expensive fields (description text from
@@ -372,10 +402,19 @@ def lookup(query: str, top: int = 5, expand_synonyms: bool = True) -> list[dict]
             "visual_patterns": _patterns_for(e),
             "has_description": bool(c.get("_description")),
         })
-    return scored
+    return {"matches": scored, "gaps": matched_gaps}
 
 
-def _print_results(query: str, results: list[dict]) -> None:
+def _print_results(query: str, payload: dict) -> None:
+    gaps = payload.get("gaps") or []
+    results = payload.get("matches") or []
+    for gap in gaps:
+        notice = (gap.get("notice") or "").rstrip()
+        gap_id = gap.get("id", "unknown")
+        print(f"⚠ KNOWN GAP — {gap_id}")
+        for line in notice.splitlines():
+            print(f"   {line}" if line else "")
+        print()
     if not results:
         print(f"No matches for: {query}")
         return