Skip to content

CLI reference

All commands are subcommands of tokengraph (or python tokengraph_all.py). Add --help to any command for its full flags. Run against the current repo, or point elsewhere with the global --path PATH.

Retrieval

CommandPurpose
context "task" -b 6000Token-budgeted context pack for a task
ask "task"Focused pack + intent / coverage / risk / cost metadata
semantic "query" -n 10Find symbols by meaning (embeddings)
federated "task" --root ../a --root ../bRetrieve across multiple repos, merged
diff-contextBudgeted pack for exactly what the git diff touches

Graph navigation

CommandPurpose
modulesToken-count table of top-level dirs — call first
archWhole-repo overview: modules, hubs, cycles, languages, routes
explain FILESignatures + imports + external callers for a file
skeleton FILEEvery signature in a file, no bodies
summary FILECompact module summary
callers / callees QNAMETrace the call graph (full symbol source: MCP get_symbol)
impact QNAMEBlast radius: callers / subclasses / tests
method-impact QNAMEFunction-level: who breaks / deps / overrides / call sites
test-map [TARGET]Map implementations ↔ tests (+ coverage %)
lines FILE 40 80Surgical line range (secret-scanned, sandboxed)
map imports|hierarchy|routes|hubsProject graph views

Trust gates

CommandPurpose
validate "task" --min-coverage 60Coverage gate (percent, default 60); exit 1 if insufficient
judge --answer-file A --context-file CScore if an answer is grounded (0–100%)
verify --answer-file AFlag fabricated files / symbols; exit 1 if any
groundingQuantify the guard: fabrications caught vs real refs flagged

Grounded generation

CommandPurpose
conventionsDetect naming / layout / test / export style + conformance
scaffold NAME --kind module --applyPropose (or write) a convention-matched file
verify-planCheck a plan's file/symbol refs + blast radius before acting
verify-outputAudit AI-generated code for fabricated files / symbols / imports
reviewAudit the working/staged diff for scope drift, hub edits, missing tests
create "task" --applyOrchestrate scaffold → plan → verify-output → review

Cost, savings & optimization

CommandPurpose
costEstimate USD of an API call before sending it
prompt-scoreScore a prompt's clarity / specificity / context / action
summarize-chatCompress a long transcript into a token-cheap brief
dedupeRemove near-duplicate context blocks
squeezeShrink a pasted stacktrace / CI log / JSON blob
measure "task" / reportToken savings for a task / aggregated
gainRealized savings from the ledger (tokens + $), dashboards
statusOne-line repo snapshot (branch / index / savings)

Setup & operations

CommandPurpose
indexBuild / rebuild the graph (MCP tool: reindex)
serveRun the MCP server (stdio or HTTP)
ide-setupWire the MCP server + rules (default: Claude Code + VS Code/Copilot; --all for every editor)
ide-pluginScaffold installable VS Code / Neovim / JetBrains plugins
langsList parseable languages by extraction tier
evidenceDeterministic, hash-grounded evidence pack (audit/CI)
hallucinationMulti-repo, reproducible grounding benchmark
doctor / healthDiagnose the install / index

See also

Released under the terms in the repository LICENSE.