ccg-namespace

ccg-namespace — Graph Namespace Isolation

Use namespaces to isolate graph rows for different services or repositories. This skill reads existing namespace state and never writes it.

Core Pattern

ccg search --namespace payment "checkout"
ccg status --namespace users

Through MCP:

list_namespaces()
search(namespace: "payment", query: "checkout")
search(namespace: "payment", query: "why does checkout retry a failed capture")

list_namespaces reports every namespace holding graph data with its node count — the first call before interpreting any namespace-scoped result. Both search query shapes work namespace-scoped: keywords and plain-language questions answered from recorded reasons.

Write Boundary

Do not plan or execute a namespace build, update, migration, postprocess, or scoped replacement from this skill. A request that needs one of those operations is not enough to invoke ccg-build: report that the user must explicitly name ccg-build in the current request.

Operational Guidance

• Use one namespace per service or repository when graph state must remain isolated.
• Use the default namespace for ordinary single-repository local work.
• Pass the same namespace consistently to search, docs, and analysis tools.
• Namespace deletion and file upload are not MCP capabilities.
• A graph namespace does not generate or copy Markdown files. Generate and place docs separately before expecting namespace-scoped get_doc_content reads to succeed.

Federation Boundaries

Only search, query_graph, and list_graph_stats accept namespaces: []. Treat get_node, get_annotation, get_doc_content, list_flows, and list_cross_refs as single-namespace operations.

A federated search labels every hit with its namespace, and candidates with no visible evidence are cut before the page is bounded, so a namespace's files are hits it can justify rather than padding.

limit and offset are per namespace here, not shared across them. limit: 5 over three namespaces means up to five files from each, and every namespace with a hit appears however small the limit is — a limit below the namespace count no longer silences the ones at the back. truncated counts the files left off across all of them, and annotation_coverage adds up across all of them too — one fraction over every namespace searched, not one namespace's. Page with the offset the response's next gives you: it moves every namespace through its own list at once, so it is not this page's offset plus the file count you were handed.