Centralize language support and generalize agent packet retrieval

.agents/skills/codestory-grounding/SKILL.md

@@ -41,9 +41,11 @@ checkout is only the tool artifact unless the user is editing CodeStory itself.
 - When `packet` reports `sufficient` and `follow_up_commands` is empty, answer
   from the packet; budget truncation alone is not a gap. Preserve supported-claim
   wording and include a compact "Support files" list from `answer.citations` and
-  `sufficiency.avoid_opening`. Do not run ordinary source reads, `rg`, `grep`, or
-  `git show` only to verify packet citations; run more commands only for a named
-  unresolved gap, an edit target, or a user-requested worktree proof.
+  `sufficiency.avoid_opening_paths`. The older `sufficiency.avoid_opening` field
+  is human-readable compatibility prose, not the raw path contract. Do not run
+  ordinary source reads, `rg`, `grep`, or `git show` only to verify packet
+  citations; run more commands only for a named unresolved gap, an edit target,
+  or a user-requested worktree proof.
 - When `packet` reports `partial`, read `sufficiency.follow_up_commands` and run
   those commands in order. Prefer listed targeted `search --why` commands before
   escalating to a larger packet budget. As soon as a follow-up packet becomes
@@ -61,6 +63,13 @@ checkout is only the tool artifact unless the user is editing CodeStory itself.
   failed, treat product retrieval as unavailable until `retrieval_mode=full` is
   restored. Repo-text output is diagnostic only; do not use it as a substitute
   for mandatory sidecar evidence.
+- Under `graph_first_v1`, `retrieval_mode=full` means graph and lexical sidecars
+  are complete, generated `symbol_search_doc` and component-report virtual docs
+  are current, and Qdrant is complete only for selected dense anchors. A zero
+  dense-anchor manifest is valid only when reported explicitly; otherwise
+  Qdrant mismatch or unavailability is fail-closed. Search evidence should name
+  provenance such as `exact`, `lexical_source`, `symbol_doc`, `graph_neighbor`,
+  `component_report`, or `dense_anchor`.
 
 ## Command Routing
 

.agents/skills/codestory-grounding/references/doctor.md

@@ -22,7 +22,7 @@ Reads project/cache/index/retrieval health without mutating the index. Use it at
 | Path | Command | Expected result |
 |------|---------|-----------------|
 | Normal path | `<codestory-cli> doctor --project <target-workspace>` | Reports project root, cache path, indexed stats, retrieval state, sidecar embedding setup, environment hints, and next commands. |
-| Failure path | If cache or index checks warn, run `index --project <target-workspace> --refresh full`; if mandatory sidecars are missing or stale, run the setup/index commands surfaced by `doctor`; if semantic reports `semantic partial`, `semantic stale`, or `semantic failed`, rebuild before trusting broad packet/search evidence. | Separates missing index, stale semantic docs, partial semantic docs, and mandatory retrieval setup failures. |
+| Failure path | If cache or index checks warn, run `index --project <target-workspace> --refresh full`; if mandatory sidecars are missing or stale, run the setup/index commands surfaced by `doctor`; if symbol docs, dense anchors, policy version, Qdrant counts, or semantic health report partial/stale/failed state, rebuild before trusting broad packet/search evidence. | Separates missing index, stale symbol docs, partial dense anchors, and mandatory retrieval setup failures. |
 | Integration edge | Use doctor before `ground`, `search --why`, `explore`, `context`, or `serve`; its next commands are the safe follow-up loop. | Prevents read commands from silently querying the wrong or empty cache. |
 
 ## Notes
@@ -31,5 +31,5 @@ Reads project/cache/index/retrieval health without mutating the index. Use it at
 - The `attention:` block repeats warnings first so agents do not miss semantic partial/stale/failure messages buried in the full check list.
 - Environment rows report retrieval-related variables such as `CODESTORY_EMBED_BACKEND`, `CODESTORY_EMBED_LLAMACPP_URL`, and sidecar enablement flags.
 - The embedding checks distinguish product llama.cpp sidecar state from hash, ONNX, disabled, or stale diagnostic states.
-- Treat `semantic ok` plus `retrieval_mode=full` as the health state suitable for broad repository explanation prompts. Treat `semantic partial`, `semantic stale`, `semantic failed`, and non-`full` retrieval modes as instructions to repair setup or rebuild before trusting agent-facing evidence.
+- Treat `semantic ok` plus `retrieval_mode=full` as the health state suitable for broad repository explanation prompts. Under `graph_first_v1`, `full` may explicitly skip Qdrant only when dense-anchor count is zero and graph/lexical artifacts are current. Treat `semantic partial`, `semantic stale`, `semantic failed`, Qdrant count mismatch, and non-`full` retrieval modes as instructions to repair setup or rebuild before trusting agent-facing evidence.
 - Prefer JSON for CI or doc-contract checks.

.agents/skills/codestory-grounding/references/files.md

@@ -35,7 +35,7 @@ claims about what the graph can see.
 
 - `files` reads persisted `FileInfo`; it does not scan the repo live unless `--refresh` asks for an index refresh.
 - Treat `index usable` with incomplete or error counts as a partial-coverage signal, not a failure.
-- `summary.framework_route_coverage` is the support matrix for framework route extraction. It includes `status`, `fixture_status`, `confidence_floor`, `handler_link_support`, `unsupported_patterns`, `known_gaps`, and `promotable`. Treat `partial`, `heuristic`, text-only handler support, and `promotable=false` as review prompts, not proof of full framework parity.
+- `summary.framework_route_coverage` is the support matrix for framework route extraction. It includes `status`, `coverage_evidence`, `confidence_floor`, `handler_link_support`, `unsupported_patterns`, `known_gaps`, and `promotable`. Treat `partial`, `heuristic`, text-only handler support, and `promotable=false` as review prompts, not proof of full framework parity.
 - Route coverage statuses:
   - `supported`: fixture-backed behavior is passing and documented coverage is met.
   - `heuristic`: pattern-backed evidence that needs source review.

.agents/skills/codestory-grounding/references/index.md

@@ -1,7 +1,8 @@
 # `index` - Build or Refresh the Symbol Index
 
 Discovers project files, extracts symbols and edges, persists graph/search state
-to SQLite, and synchronizes semantic docs when embedding assets are available.
+to SQLite, writes graph-native symbol docs and component reports, and
+synchronizes selected dense anchors when embedding assets are available.
 
 ## Usage
 
@@ -15,7 +16,7 @@ to SQLite, and synchronizes semantic docs when embedding assets are available.
 |--------|---------|-----|
 | `--project <path>` / `--path <path>` | `.` | Target repository root. Always pass this explicitly. |
 | `--cache-dir <path>` | auto | Override the per-project cache root. |
-| `--refresh <auto|full|incremental|none>` | `auto` | Choose the graph/snapshot/semantic refresh mode. |
+| `--refresh <auto|full|incremental|none>` | `auto` | Choose the graph/snapshot/symbol-doc/dense-anchor refresh mode. |
 | `--format <markdown|json>` | `markdown` | Use JSON for automation and timing analysis. |
 | `--output-file <path>` | stdout | Write output to a file with an existing parent directory. |
 | `--dry-run` | off | Show workspace discovery and planned adds/removals without writing storage. |
@@ -28,19 +29,21 @@ to SQLite, and synchronizes semantic docs when embedding assets are available.
 | Mode | Behavior |
 |------|----------|
 | `auto` | Use `full` for an empty cache and `incremental` otherwise. |
-| `full` | Rebuild the project graph and semantic docs from the discovered workspace. |
-| `incremental` | Reindex changed/new/unindexed files, remove disappeared files, and prune touched semantic docs. |
+| `full` | Rebuild the project graph, symbol docs, component reports, and dense anchors from the discovered workspace. |
+| `incremental` | Reindex changed/new/unindexed files, remove disappeared files, and prune touched symbol docs or dense anchors. |
 | `none` | Inspect the existing cache without refreshing it. Use only after a known-good same-session index. |
 
 Use `--refresh full` for first-time indexes, cache/schema uncertainty, and fixes
 for historical indexing failures. Incremental runs can leave stale error rows
 when previously failing files are not touched.
 
-## Semantic Retrieval
+## Symbol Docs And Dense Anchors
 
-There is no `index --semantic off` flag. Semantic docs are part of the default
-index contract when embedding assets are ready. On a fresh machine, check the
-setup plan first:
+There is no `index --semantic off` flag. Graph-native `symbol_search_doc` rows
+are part of the default index contract. Under `graph_first_v1`, dense vectors
+are only written for selected anchors such as entrypoints, public APIs,
+documented nontrivial symbols, central graph nodes, component reports, and
+unstructured docs. On a fresh machine, check the setup plan first:
 
 ```text
 <codestory-cli> setup embeddings --project <target-workspace> --dry-run --format json
@@ -53,25 +56,27 @@ High-signal environment toggles:
 
 | Variable | Use |
 |----------|-----|
-| `CODESTORY_SEMANTIC_DOC_SCOPE=all` | Include all-symbol semantic docs. Accepted all-symbol aliases are `all`, `full`, `all-symbols`, and `all_symbols`; omitted or other values default to durable symbols. |
+| `CODESTORY_SEMANTIC_DOC_SCOPE=all` | Include the broader all-symbol symbol-doc scope for diagnostics. Accepted aliases are `all`, `full`, `all-symbols`, and `all_symbols`; omitted or other values default to durable symbols. |
 | `CODESTORY_EMBED_BACKEND=llamacpp` | Use the mandatory local llama.cpp embedding sidecar. |
 | `CODESTORY_EMBED_LLAMACPP_URL=http://127.0.0.1:8080/v1/embeddings` | Product embedding endpoint for bge-base sidecar vectors. |
 | `CODESTORY_SUMMARY_ENDPOINT=local` | Enable deterministic local summaries with `--summarize`. |
 
 Use other embedding, alias, batch-size, tokenizer, provider, hash, ONNX, and
 summary tuning variables only for focused diagnostics or historical comparisons.
 Agent packet/search readiness requires retrieval status to report
-`retrieval_mode=full`.
+`retrieval_mode=full`. A zero dense-anchor corpus is valid only when the
+manifest reports it explicitly; otherwise stale or unavailable Qdrant state
+fails closed.
 
 ## Output
 
 Markdown returns a compact index summary. JSON exposes the same data for tools:
 
 - project and storage path
 - refresh mode and discovered file/error counts
-- local navigation readiness notes and semantic doc counts
+- local navigation readiness notes, symbol-doc counts, dense-anchor counts, and policy reason counts
 - parse, flush, resolve, cleanup, cache, and semantic timing buckets
-- resolution counters and semantic reuse/embed/prune counts
+- resolution counters plus symbol-doc write and dense-anchor reuse/embed/skip/prune counts
 
 Important timing fields are `timings_ms.parse`, `timings_ms.flush`,
 `timings_ms.resolve`, `timings_ms.cleanup`, `cache_ms.search_index`,

.agents/skills/codestory-grounding/references/packet.md

@@ -31,7 +31,7 @@ tracing, ownership discovery, or change-impact analysis.
 |------|---------|-----------------|
 | Normal path | `<codestory-cli> packet --project <target-workspace> --question "How does indexing flow from CLI to storage?" --budget compact` | Markdown packet with cited claims, budget usage, gaps, and follow-up commands. |
 | Failure path | If the packet reports `partial` or `insufficient`, follow its `follow_up_commands`, usually deeper packet budget or concrete `search`, `context`, `trail`, or `snippet` calls. | Broad exploration is bounded by reported gaps instead of drifting into repeated file reads. |
-| Integration edge | Use JSON output for harnesses and stdio clients. If `sufficiency.status` is `sufficient` and `follow_up_commands` is empty, answer from packet supported claims and include a compact support-file list from `answer.citations` and `sufficiency.avoid_opening`; budget truncation alone is not a gap. | Makes benchmark traces and agent loops comparable across runs. |
+| Integration edge | Use JSON output for harnesses and stdio clients. If `sufficiency.status` is `sufficient` and `follow_up_commands` is empty, answer from packet supported claims and include a compact support-file list from `answer.citations` and `sufficiency.avoid_opening_paths`; budget truncation alone is not a gap. Treat `sufficiency.avoid_opening` as compatibility prose only. | Makes benchmark traces and agent loops comparable across runs. |
 
 ## Notes