Named context "Repos Linked: 0" desync: index_helper early-return skips register_repo_in_context for already-populated graphs #1317

Summary

In named context mode, a context's repos: list in config.yaml can permanently desync from the actual graph: cgc context list shows "Repos Linked: 0" even though the Neo4j graph is fully populated (Repository node + hundreds of thousands of nodes) and cgc find / cgc query / the MCP tools all resolve correctly against it.

The root cause is an ordering bug in index_helper (src/codegraphcontext/cli/cli_helpers.py): the "already indexed, skipping" early-return sits above the register_repo_in_context(...) call. Once a context's graph has been populated by any path that did not register the repo into config.yaml (e.g. the MCP add_code_to_graph tool, or a CLI run interrupted/killed mid-resolution), every subsequent cgc index . --context <ctx> short-circuits at the skip guard before reaching the registration line — so the two stores can never reconcile through the normal command path. It is self-perpetuating.

Environment

• CodeGraphContext: 0.5.1 (pipx)
• Neo4j: 2026.05.0 (Homebrew), bolt://localhost:7687
• OS: macOS
• Mode: named (single context, set as default_context)

Reproduction

1. Populate a named context's graph via a path that does not register the repo in config.yaml — e.g. the MCP add_code_to_graph tool, or start cgc index . --context X and kill it during the CALLS-resolution phase (graph nodes are written incrementally; the repo registration is not reached).
2. Confirm the graph is real:

   cgc query "MATCH (n) RETURN count(n)"                 # -> hundreds of thousands
   cgc query "MATCH (r:Repository) RETURN r.path, r.name" # -> your repo
   

3. Observe the desync:

   cgc context list      # -> "Repos Linked: 0"
   cat ~/.codegraphcontext/config.yaml   # -> contexts.X.repos: []
   

4. Try to repair via the normal command:

   cgc index . --context X
   # -> "Repository '.' is already indexed with N files. Skipping."
   

   It returns before registering the repo, so repos: stays [] forever.

Root cause

In cli/cli_helpers.py, index_helper (around the lines below in 0.5.1):

# repo already in graph WITH files -> early return
if file_count > 0:
    console.print(f"...already indexed with {file_count} files. Skipping.")
    console.print("...Use 'cgc index --force' to re-index")
    db_manager.close_driver()
    return                                   # <-- returns here (~line 300)

# ...registration happens AFTER, so it is never reached on the skip path:
if context and ctx.mode == "named":
    if not register_repo_in_context(context, str(path_obj), auto_create=False):  # ~line 306-309
        db_manager.close_driver()
        raise typer.Exit(code=1)

register_repo_in_context (in cli/config_manager.py) is the only writer of ctx.repos. Because the early return is above it, a context whose graph was populated without registration can never be repaired by cgc index — the skip guard fires first.

This is closely related to #701 (cgc index . says repo "exists but has no files"), which lives in the same repo-state-detection block (MATCH (r:Repository)-[:CONTAINS*]->(f:File) file-count check).

Impact

• The graph works fine, so it's not data loss — but cgc context list misreports Repos Linked: 0, which reads as "nothing is indexed" and is alarming/confusing.
• Any feature that scopes off ctx.repos (e.g. watch, re-index targeting) sees an empty list for a fully-indexed context.

Suggested fix

Register the repo before the "already indexed, skip" early-return (registration is idempotent), or register inside the skip branch before returning. Minimal change — move the if context and ctx.mode == "named": register_repo_in_context(...) block above the file_count > 0 early return, so a known/indexed repo still gets linked into the named context's repos: list.

Workaround

Call the idempotent registration writer directly (no re-index needed):

cd "$(pipx environment --value PIPX_LOCAL_VENVS)/codegraphcontext"  # or your venv
./bin/python -c "from codegraphcontext.cli.config_manager import register_repo_in_context as r; r('X', '/abs/path/to/repo')"
cgc context list   # -> Repos Linked: 1

(cgc index --force also works but triggers a full re-index, which is expensive on large repos — see the separate CALLS-resolution perf work in #1304.)