feat(cross-repo): extend project='*' to get_impact and trace_call_path + channel dedup

Three fixes:

1. Cross-repo get_impact (project='*'):
   Searches _cross_repo.db for the target symbol across all repos.
   For each repo containing it, opens the per-project DB, resolves
   Class→Method, runs BFS, and collects depth-grouped impact results.
   Returns per-repo risk assessment with combined risk level.

2. Cross-repo trace_call_path (project='*'):
   Same pattern — finds the function in all repos, traces both
   inbound (callers) and outbound (callees) at depth 2 per repo.
   Uses cbm_cross_repo_trace_in_project() for BFS with Class→Method
   and file-level resolution.

3. Channel dedup safety net:
   After channel detection Pass 2 (file-level constants), deletes
   file-level ghost entries when a named function entry already exists
   for the same (channel_name, file_path, project, direction).
   Applied in both per-project detect_channels (store.c) and
   cross-repo build (cross_repo.c). Currently deletes 0 rows
   (no actual ghosts in current data), but prevents future duplicates
   if both extractors detect the same pattern.

All three tools now support project='*' for cross-repo queries:
- search_graph(project='*') — cross-repo search
- get_impact(project='*') — cross-repo blast radius
- trace_call_path(project='*') — cross-repo call chain trace

Author: Your Name

src/mcp/mcp.c

@@ -1002,6 +1002,7 @@ static char *handle_get_graph_schema(cbm_mcp_server_t *srv, const char *args) {
 
 /* Forward declarations — defined in cross-repo handler section */
 static char *handle_cross_repo_search(cbm_mcp_server_t *srv, const char *args);
+static char *handle_cross_repo_trace_call(cbm_mcp_server_t *srv, const char *args, char *func_name);
 static char *derive_short_project(const char *full_project);
 static void add_trace_steps(yyjson_mut_doc *doc, yyjson_mut_val *parent,
                             const char *key, cbm_cross_trace_step_t *steps, int count);
@@ -1430,8 +1431,18 @@ static char *handle_get_process_steps(cbm_mcp_server_t *srv, const char *args) {
     return result;
 }
 
+/* Forward declaration */
+static char *handle_cross_repo_impact(cbm_mcp_server_t *srv, const char *args);
+
 static char *handle_get_impact(cbm_mcp_server_t *srv, const char *args) {
     char *project = cbm_mcp_get_string_arg(args, "project");
+
+    /* Cross-repo impact: project="*" searches all repos for the target */
+    if (project && strcmp(project, "*") == 0) {
+        free(project);
+        return handle_cross_repo_impact(srv, args);
+    }
+
     char *target = cbm_mcp_get_string_arg(args, "target");
     char *direction = cbm_mcp_get_string_arg(args, "direction");
     int max_depth = cbm_mcp_get_int_arg(args, "max_depth", 3);
@@ -2066,6 +2077,13 @@ static char *handle_get_architecture(cbm_mcp_server_t *srv, const char *args) {
 static char *handle_trace_call_path(cbm_mcp_server_t *srv, const char *args) {
     char *func_name = cbm_mcp_get_string_arg(args, "function_name");
     char *project = cbm_mcp_get_string_arg(args, "project");
+
+    /* Cross-repo trace: project="*" searches all repos for the function */
+    if (project && strcmp(project, "*") == 0) {
+        free(project);
+        return handle_cross_repo_trace_call(srv, args, func_name);
+    }
+
     cbm_store_t *store = resolve_store(srv, project);
     char *direction = cbm_mcp_get_string_arg(args, "direction");
     int depth = cbm_mcp_get_int_arg(args, "depth", 3);
@@ -3979,7 +3997,232 @@ static char *handle_generate_embeddings(cbm_mcp_server_t *srv, const char *args)
     return result;
 }
 
-/* ── build_cross_repo_index ──────────────────────────────────── */
+/* ── Cross-repo get_impact (project="*") ─────────────────────── */
+
+static char *handle_cross_repo_impact(cbm_mcp_server_t *srv, const char *args) {
+    (void)srv;
+    char *target = cbm_mcp_get_string_arg(args, "target");
+    char *direction = cbm_mcp_get_string_arg(args, "direction");
+    int max_depth = cbm_mcp_get_int_arg(args, "max_depth", 3);
+    if (!direction) direction = heap_strdup("upstream");
+    if (max_depth <= 0) max_depth = 3;
+
+    if (!target || !target[0]) {
+        free(target); free(direction);
+        return cbm_mcp_text_result("{\"error\":\"target is required\"}", true);
+    }
+
+    cbm_cross_repo_t *cr = cbm_cross_repo_open();
+    if (!cr) {
+        free(target); free(direction);
+        return cbm_mcp_text_result(
+            "{\"error\":\"Cross-repo index not built. Run build_cross_repo_index first.\"}", true);
+    }
+
+    /* Find all repos containing the target symbol via BM25 search */
+    cbm_cross_search_output_t search_out = {0};
+    cbm_cross_repo_search(cr, target, NULL, 0, 20, &search_out);
+
+    yyjson_mut_doc *doc = yyjson_mut_doc_new(NULL);
+    yyjson_mut_val *root = yyjson_mut_obj(doc);
+    yyjson_mut_doc_set_root(doc, root);
+
+    yyjson_mut_obj_add_strcpy(doc, root, "target", target);
+    yyjson_mut_obj_add_strcpy(doc, root, "direction", direction);
+    yyjson_mut_obj_add_bool(doc, root, "cross_repo", true);
+
+    bool is_upstream = strcmp(direction, "upstream") == 0;
+    const char *bfs_dir = is_upstream ? "inbound" : "outbound";
+    const char *edge_types[] = {"CALLS"};
+
+    char *seen_projects[50] = {0};
+    int seen_count = 0;
+
+    yyjson_mut_val *repos_arr = yyjson_mut_arr(doc);
+    int total_affected = 0;
+    const char *max_risk = "LOW";
+
+    for (int si = 0; si < search_out.count && seen_count < 50; si++) {
+        cbm_cross_search_result_t *sr = &search_out.results[si];
+        if (!sr->project || !sr->name) continue;
+        if (strcmp(sr->name, target) != 0) continue;
+        if (!sr->label) continue;
+        if (strcmp(sr->label, "Function") != 0 && strcmp(sr->label, "Method") != 0 &&
+            strcmp(sr->label, "Class") != 0) continue;
+
+        bool seen = false;
+        for (int j = 0; j < seen_count; j++) {
+            if (seen_projects[j] && strcmp(seen_projects[j], sr->project) == 0) { seen = true; break; }
+        }
+        if (seen) continue;
+        seen_projects[seen_count++] = heap_strdup(sr->project);
+
+        char db_path_buf[2048];
+        project_db_path(sr->project, db_path_buf, sizeof(db_path_buf));
+        cbm_store_t *pstore = cbm_store_open_path_query(db_path_buf);
+        if (!pstore) continue;
+
+        cbm_node_t *nodes = NULL; int ncount = 0;
+        cbm_store_find_nodes_by_name(pstore, sr->project, target, &nodes, &ncount);
+
+        if (ncount > 0) {
+            int64_t start_id = nodes[0].id;
+            if (nodes[0].label && strcmp(nodes[0].label, "Class") == 0) {
+                sqlite3 *pdb = cbm_store_get_db(pstore);
+                sqlite3_stmt *ms = NULL;
+                sqlite3_prepare_v2(pdb,
+                    "SELECT target_id FROM edges WHERE source_id=?1 AND type='DEFINES_METHOD' LIMIT 1",
+                    -1, &ms, NULL);
+                if (ms) {
+                    sqlite3_bind_int64(ms, 1, start_id);
+                    if (sqlite3_step(ms) == SQLITE_ROW) start_id = sqlite3_column_int64(ms, 0);
+                    sqlite3_finalize(ms);
+                }
+            }
+
+            cbm_traverse_result_t tr = {0};
+            cbm_store_bfs(pstore, start_id, bfs_dir, edge_types, 1, max_depth, 50, &tr);
+
+            yyjson_mut_val *repo_item = yyjson_mut_obj(doc);
+            char *short_proj = derive_short_project(sr->project);
+            yyjson_mut_obj_add_strcpy(doc, repo_item, "project", short_proj ? short_proj : "");
+            yyjson_mut_obj_add_strcpy(doc, repo_item, "project_id", sr->project);
+            free(short_proj);
+            yyjson_mut_obj_add_strcpy(doc, repo_item, "file_path", sr->file_path ? sr->file_path : "");
+            yyjson_mut_obj_add_int(doc, repo_item, "total_affected", tr.visited_count);
+
+            int d1 = 0;
+            yyjson_mut_val *d1_arr = yyjson_mut_arr(doc);
+            for (int vi = 0; vi < tr.visited_count; vi++) {
+                if (tr.visited[vi].hop == 1) {
+                    d1++;
+                    yyjson_mut_val *h = yyjson_mut_obj(doc);
+                    yyjson_mut_obj_add_strcpy(doc, h, "name",
+                        tr.visited[vi].node.name ? tr.visited[vi].node.name : "");
+                    yyjson_mut_obj_add_strcpy(doc, h, "label",
+                        tr.visited[vi].node.label ? tr.visited[vi].node.label : "");
+                    yyjson_mut_obj_add_strcpy(doc, h, "file_path",
+                        tr.visited[vi].node.file_path ? tr.visited[vi].node.file_path : "");
+                    yyjson_mut_arr_add_val(d1_arr, h);
+                }
+            }
+            yyjson_mut_obj_add_val(doc, repo_item, "d1_will_break", d1_arr);
+
+            const char *risk = d1 >= 20 ? "CRITICAL" : d1 >= 10 ? "HIGH" : d1 >= 3 ? "MEDIUM" : "LOW";
+            yyjson_mut_obj_add_str(doc, repo_item, "risk", risk);
+            if (strcmp(risk, "CRITICAL") == 0 ||
+                (strcmp(risk, "HIGH") == 0 && strcmp(max_risk, "CRITICAL") != 0) ||
+                (strcmp(risk, "MEDIUM") == 0 && strcmp(max_risk, "LOW") == 0))
+                max_risk = risk;
+            total_affected += tr.visited_count;
+
+            yyjson_mut_arr_add_val(repos_arr, repo_item);
+            cbm_store_traverse_free(&tr);
+        }
+        cbm_store_free_nodes(nodes, ncount);
+        cbm_store_close(pstore);
+    }
+
+    yyjson_mut_obj_add_str(doc, root, "risk", max_risk);
+    yyjson_mut_obj_add_int(doc, root, "total_affected", total_affected);
+    yyjson_mut_obj_add_int(doc, root, "repos_with_target", seen_count);
+    yyjson_mut_obj_add_val(doc, root, "repos", repos_arr);
+
+    char *json = yy_doc_to_str(doc);
+    yyjson_mut_doc_free(doc);
+    cbm_cross_search_free(&search_out);
+    cbm_cross_repo_close(cr);
+    for (int i = 0; i < seen_count; i++) free(seen_projects[i]);
+    free(target); free(direction);
+    char *result = cbm_mcp_text_result(json, false);
+    free(json);
+    return result;
+}
+
+/* ── Cross-repo trace_call_path (project="*") ────────────────── */
+
+static char *handle_cross_repo_trace_call(cbm_mcp_server_t *srv, const char *args,
+                                          char *func_name) {
+    (void)srv;
+    if (!func_name || !func_name[0]) {
+        free(func_name);
+        return cbm_mcp_text_result("{\"error\":\"function_name is required\"}", true);
+    }
+
+    cbm_cross_repo_t *cr = cbm_cross_repo_open();
+    if (!cr) {
+        free(func_name);
+        return cbm_mcp_text_result(
+            "{\"error\":\"Cross-repo index not built. Run build_cross_repo_index first.\"}", true);
+    }
+
+    cbm_cross_search_output_t search_out = {0};
+    cbm_cross_repo_search(cr, func_name, NULL, 0, 30, &search_out);
+
+    yyjson_mut_doc *doc = yyjson_mut_doc_new(NULL);
+    yyjson_mut_val *root = yyjson_mut_obj(doc);
+    yyjson_mut_doc_set_root(doc, root);
+    yyjson_mut_obj_add_strcpy(doc, root, "function_name", func_name);
+    yyjson_mut_obj_add_bool(doc, root, "cross_repo", true);
+
+    char *seen_projects[50] = {0};
+    int seen_count = 0;
+    yyjson_mut_val *repos_arr = yyjson_mut_arr(doc);
+
+    for (int si = 0; si < search_out.count && seen_count < 50; si++) {
+        cbm_cross_search_result_t *sr = &search_out.results[si];
+        if (!sr->project || !sr->name) continue;
+        if (strcmp(sr->name, func_name) != 0) continue;
+        if (!sr->label) continue;
+        if (strcmp(sr->label, "Function") != 0 && strcmp(sr->label, "Method") != 0 &&
+            strcmp(sr->label, "Class") != 0) continue;
+
+        bool seen = false;
+        for (int j = 0; j < seen_count; j++) {
+            if (seen_projects[j] && strcmp(seen_projects[j], sr->project) == 0) { seen = true; break; }
+        }
+        if (seen) continue;
+        seen_projects[seen_count++] = heap_strdup(sr->project);
+
+        char db_path_buf[2048];
+        project_db_path(sr->project, db_path_buf, sizeof(db_path_buf));
+
+        yyjson_mut_val *repo_item = yyjson_mut_obj(doc);
+        char *short_proj = derive_short_project(sr->project);
+        yyjson_mut_obj_add_strcpy(doc, repo_item, "project", short_proj ? short_proj : "");
+        yyjson_mut_obj_add_strcpy(doc, repo_item, "project_id", sr->project);
+        free(short_proj);
+        yyjson_mut_obj_add_strcpy(doc, repo_item, "file_path", sr->file_path ? sr->file_path : "");
+        yyjson_mut_obj_add_strcpy(doc, repo_item, "label", sr->label);
+
+        cbm_cross_trace_step_t *in_steps = NULL; int in_count = 0;
+        cbm_cross_repo_trace_in_project(db_path_buf, func_name, sr->file_path,
+                                        NULL, "inbound", 2, &in_steps, &in_count);
+        if (in_count > 0) add_trace_steps(doc, repo_item, "incoming_calls", in_steps, in_count);
+        cbm_cross_trace_free(in_steps, in_count);
+
+        cbm_cross_trace_step_t *out_steps = NULL; int out_count = 0;
+        cbm_cross_repo_trace_in_project(db_path_buf, func_name, sr->file_path,
+                                        NULL, "outbound", 2, &out_steps, &out_count);
+        if (out_count > 0) add_trace_steps(doc, repo_item, "outgoing_calls", out_steps, out_count);
+        cbm_cross_trace_free(out_steps, out_count);
+
+        yyjson_mut_arr_add_val(repos_arr, repo_item);
+    }
+
+    yyjson_mut_obj_add_int(doc, root, "repos_with_function", seen_count);
+    yyjson_mut_obj_add_val(doc, root, "repos", repos_arr);
+
+    char *json = yy_doc_to_str(doc);
+    yyjson_mut_doc_free(doc);
+    cbm_cross_search_free(&search_out);
+    cbm_cross_repo_close(cr);
+    for (int i = 0; i < seen_count; i++) free(seen_projects[i]);
+    free(func_name);
+    char *result = cbm_mcp_text_result(json, false);
+    free(json);
+    return result;
+}
 
 /* ── Cross-repo search (search_graph with project="*") ───────── */
 

src/store/cross_repo.c

@@ -364,6 +364,16 @@ cbm_cross_repo_stats_t cbm_cross_repo_build(void) {
     if (ins_node) sqlite3_finalize(ins_node);
     if (ins_emb) sqlite3_finalize(ins_emb);
 
+    /* Suppress file-level ghost channel entries when named entries exist */
+    sqlite3_exec(db,
+        "DELETE FROM cross_channels WHERE function_name = '(file-level)' "
+        "AND EXISTS (SELECT 1 FROM cross_channels c2 "
+        "WHERE c2.channel_name = cross_channels.channel_name "
+        "AND c2.file_path = cross_channels.file_path "
+        "AND c2.project = cross_channels.project "
+        "AND c2.direction = cross_channels.direction "
+        "AND c2.function_name != '(file-level)')", NULL, NULL, NULL);
+
     /* Build FTS5 index with camelCase splitting */
     sqlite3_exec(db, "DELETE FROM cross_nodes_fts", NULL, NULL, NULL);
     sqlite3_exec(db,

src/store/store.c

@@ -5400,6 +5400,19 @@ int cbm_store_detect_channels(cbm_store_t *s, const char *project, const char *r
     }
 
     if (ins) sqlite3_finalize(ins);
+
+    /* Suppress file-level ghost entries when a named function entry exists
+     * for the same channel+file+direction. This prevents duplicate channel
+     * entries when both per-node (Pass 1) and file-level (Pass 2) extractors
+     * detect the same channel pattern in the same file. */
+    exec_sql(s, "DELETE FROM channels WHERE function_name = '(file-level)' "
+                "AND EXISTS (SELECT 1 FROM channels c2 "
+                "WHERE c2.channel_name = channels.channel_name "
+                "AND c2.file_path = channels.file_path "
+                "AND c2.project = channels.project "
+                "AND c2.direction = channels.direction "
+                "AND c2.function_name != '(file-level)')");
+
     return total;
 }