Module usage 

Cost/token usage ingestion for interactive claude-code sessions.

Ninox launches the real claude CLI interactively inside a tmux pane (see harness::HarnessRegistry::interactive_cmd/worker_cmd) rather than driving it through a scripted API, so there is no request/response boundary Ninox controls where a cost or token count could be captured directly. What does exist is the CLI’s own on-disk transcript: every turn is appended as a JSON line to

~/.claude/projects/<escaped-workspace-path>/<claude-session-uuid>.jsonl

where the directory name is the session’s working directory with every non-alphanumeric character replaced by -, and each assistant-type line carries a message.usage object (input_tokens, output_tokens, cache_creation_input_tokens, cache_read_input_tokens) plus message.model. Since each Ninox session (orchestrator subdirectory, standalone worktree, or CLI worker worktree) runs in its own dedicated workspace directory, that directory is a reliable 1:1 key back to the session — no ATHENE_SESSION-style attribution needed for this part.

There is no on-disk USD figure to read, so cost is estimated from token counts against a small built-in pricing table. These are rough priors, not live pricing — good enough to answer “is this session burning money” and to seed the spawn-modal’s data-driven estimate, not to reconcile an invoice.

Structs

UsageSnapshot

A snapshot of a workspace’s accumulated claude usage, computed by summing every turn recorded in its transcript(s).

Functions

claude_project_slug

The directory-name claude derives from a working-directory path: every byte that isn’t an ASCII letter or digit becomes -.

claude_projects_dir

Directory Ninox expects claude’s own per-project transcripts to live under. Honors NINOX_CLAUDE_PROJECTS_DIR as a test/override seam (mirrors AppConfig::resolved_brain_path’s NINOX_BRAIN pattern); otherwise ~/.claude/projects.

ingest_usage_for_workspace

Compute the current usage snapshot for a session’s workspace directory, by locating and summing claude’s own transcript(s) for that directory. Returns None when no transcripts exist yet (e.g. the agent hasn’t sent its first turn) or the workspace can’t be resolved.