seathatflowsinourveins · Sep 20, 2026 · Sep 20, 2026
@@ -21,5 +21,7 @@ jobs:
         run: python3 scripts/validate.py
       - name: Validate catalog evidence and coverage joins
         run: python3 scripts/validate_catalogs.py
+      - name: Check offline ecosystem explorer rebuild
+        run: python3 scripts/build_ecosystem.py --check
       - name: Test validation failure modes
         run: python3 -m unittest
@@ -4,6 +4,11 @@ A current, evidence-backed reference for native **Codex + Claude Code**, with sc
 
 **Snapshot: September 20, 2026.** This repository records a tested selection and its limits. It is not a claim that every available framework is installed or that an independent universal SOTA benchmark has been won.
 
+Open the **[offline ecosystem explorer](docs/ecosystem/README.md)** for the full
+layer map, searchable public repository index, native skills and workers, scoped
+memory/retrieval, and token-efficiency policy. Its downloadable single HTML file
+keeps dated source reviews, recorded execution and current-host acceptance distinct.
+
 The new **[evidence-led convergence practice](blueprints/convergence-practice/README.md)**
 connects public own repositories, stars and curated-list discovery to pinned
 source review, frozen experiments and explicit adoption decisions. It adds an

@@ -39,7 +39,9 @@
     "current_corporate_action_readiness": "blueprints/us-equities/corporate-action-readiness/receipt.json",
     "current_authenticated_data": "blueprints/us-equities/authenticated-data/native-receipt.json",
     "current_wave": "blueprints/us-equities/identity-readiness/plan.json",
-    "current_identity_observation": "blueprints/us-equities/identity-readiness/native-receipt.json"
+    "current_identity_observation": "blueprints/us-equities/identity-readiness/native-receipt.json",
+    "ecosystem_explorer": "docs/ecosystem/index.html",
+    "ecosystem_explorer_guide": "docs/ecosystem/README.md"
   },
   "recipe_map": {
     "affaan-m/ECC": "recipes/README.md",

@@ -0,0 +1,101 @@
+# Offline ecosystem manifest
+
+Open [index.html](index.html) locally in a browser. It is one self-contained file:
+the styles, application script and public search data are embedded. The page
+makes no background requests and needs no server, account, package install or
+external font. Source links navigate only when selected. GitHub shows HTML source;
+download the file to view it locally. The same file can be served by an existing
+static host, but this change does not configure GitHub Pages or paid hosting.
+
+The four views connect a layered ecosystem map, the canonical repository explorer,
+token-efficiency policy, and evidence with dated source provenance. All 504 public
+index identities and the existing 342-star snapshot are retained. The separate
+current-integrations lane makes newly observed Tavily setup searchable without
+silently enlarging the canonical index or accepted component manifest. Stars and
+awesome lists remain discovery signals. Layer tags are navigation heuristics,
+not adoption decisions or quality scores.
+
+## Rebuild and check
+
+From the repository root, using Python 3.10 or newer and its standard library:
+
+```sh
+python3 scripts/build_ecosystem.py --write
+python3 scripts/build_ecosystem.py --check
+python3 scripts/validate.py
+python3 scripts/validate_catalogs.py
+python3 -m unittest
+```
+
+Some existing data tests intentionally refuse symlinked paths and private output
+inside Git. On macOS, set `TMPDIR` to a resolved temporary directory outside any
+checkout before running the full suite. Optional native SDK/data tests retain
+their normal dependency-based skips.
+
+The curated layer map, policy summaries and links live in [manifest.json](manifest.json).
+The presentation lives in [template.html](template.html). The generator reads only
+explicit public files inside this repository; source paths cannot escape it or
+follow symlinks. It resolves the decision index's typed pointers and embeds input
+hashes. Unknown receipts, unresolved source pointers, duplicate repository
+identities, noncanonical repository URLs and stale generated bytes fail the build.
+Source text is JSON-escaped and rendered with DOM text nodes; navigable source
+URLs use credential-free HTTPS. A restrictive content security policy permits
+only the exact embedded application script and denies background connections.
+
+The generated page is deterministic. It does not use current wall time, Git status,
+network responses, personal catalogs or live credentials. The evidence manifest's
+`/receipts` section has its own canonical JSON hash; its `files` hash list is not
+an input, avoiding a cycle when that list hashes the generated HTML. All other
+input hashes identify exact file bytes. Run `--check` after changing a source;
+update the existing publication hash manifest after regeneration.
+
+Existing source-record links use the recorded immutable public base commit.
+New files in `docs/ecosystem/` did not exist at that base; their links use public
+`main` after publication, while their exact bytes are embedded or hash-listed.
+The page makes this distinction explicit. This does not turn a moving branch
+link into an immutable source pin; upstream review records retain exact commits.
+
+## Evidence boundaries
+
+The explorer joins useful native execution only from explicitly linked
+`native_cli_e2e` or `native_model_e2e` receipts. A historical version inventory,
+catalog decision, source review or component pin does not become useful execution.
+The two retrieval replay highlights are separately identified as recorded
+evaluations, with their own denominators and limits. Current acceptance on the
+browser's host stays unknown: a static HTML file probes no accounts or services.
+
+The [current primary-source review](source-review.json) records native Claude Code
+features, existing retrieval and memory options, and local inference gates. It
+does not install anything or change accepted pins. In particular, ai-memory 2.3.2
+remains a candidate for an isolated migration/scope trial; native 2.3.1 state was
+not upgraded by this work.
+
+The [Tavily receipt](tavily-receipt.json) records CLI 0.1.8, eight official skills
+at a full source commit, native authentication and one three-result search.
+Subsequent fresh native Codex CLI prompt rendering and Claude initialization
+discovered all eight skills without model calls. Existing desktop hot reload remains
+unverified; extract, crawl and deep research were not accepted by that search. The [readiness observation](readiness-observation.json)
+records a separate existing WSL environment's native sign-ins and 75 offline worker
+checks. A subsequent [standalone native Claude receipt](legacy-worker-receipt.json)
+retains an initial turn-cap failure and a revised accepted task with 57 actual
+tests, zero source product edits and complete worker cleanup. The native client
+reported 11 turns under the revised 12-turn bound; effective effort is unknown.
+The changed prompt and bound prevent a matched efficiency comparison. Native
+usage and list-price estimates are retained without summing them into savings or
+account billing. This proves one standalone task, not delegated owner/peer
+integration, Codex inference or general acceptance on another host. These public
+records omit account identifiers, personal paths and private project topology.
+
+Token efficiency is a core operating policy across native clients, skills,
+subagents and workers: keep cache and compaction, retrieve scoped sources, disclose
+instructions progressively, retain paths and failures in bounded outputs, and
+write scoped memory only with explicit intent. Automatic history capture stays
+off. Avoid duplicate MCP services and bulk instruction packs. Claim token or cost
+savings only with matched complete usage and comparable task quality; native
+cache counts and selected-artifact reductions are distinct measurements.
+
+Tests cover source confinement, unresolved and duplicate identities, execution
+boundaries, source/URL escaping, new-integration counts, input hashes, changed
+documents and generated-file tampering. Browser review covers searching, combined
+filters, dialogs, keyboard exit, small screens and the absence of external asset
+requests. Repository validation remains an integrity check, not a new native run.