Flow fingerprints
Two-level hashed fingerprints (structural plus content) identify a function's flow across revisions. Content-addressed, so identical behavior hashes identically even as surrounding code moves.
Read the tool referenceThe features
New codebase?
See the decisions.
Old codebase? Read the accumulated intent — validated.
Thirteen tools, ranked by what only DesignGraph can do. A new codebase has no accumulated intent yet — the structural decision-path lens works immediately. An old codebase has history — the semantic layer reads it back, validated against evidence.
Full pages
Four features,
in depth.
- 01 · Moat
controller_chain— Controller chainsCall graph with controller awareness: the conditions and branches guarding every call, up and down.
Follow the path - 02 · Moat
groups_manifest— Session memorySession groups link straight to the session file — understanding that survives the session.
Continue the thread - 03 · Unique
sot_query— State ownershipCanonical write sites: who truly owns a state variable, with all writers and the rationale.
Find the source - 04 · Unique
groups_for_function— Intent that travelsThe composed intent of one function: its groups, claims, and zoom levels across sessions.
Read the claims
Brief entries
Nine more tools,
at a glance.
- 05
flow— Flow fingerprintsTwo-level hashed flow fingerprints for a function: structure plus content, addressable.
Read the brief - 06
groups_validate— Evidence and claimsClaims checked against evidence blocks — validated, never merely asserted.
Read the brief - 07
callers— CallersWho calls this function — every answer carrying its own honesty envelope.
Read the brief - 08
references— ReferencesEvery occurrence of a name: definition, callers, and callees in one view.
Read the brief - 09
- 10
status— StatusIndex status: freshness, counts, and diagnostics before you trust an answer.
Read the brief - 11
what_changed— What changedHermetic change audit between two states of the code — no silent drift.
Read the brief - 12
call_graph— Call graphThe summary-first call graph around a function — zoom in only where it matters.
Read the brief - 13 · Table-stakes
impact— ImpactBlast radius before you edit: the required pre-edit gate for a safe change.
Read the brief
Evidence and claims
Function groups carry claims, and claims resolve against evidence blocks. The validate report tells you which claims hold and which need attention — the honesty moat, folded into Intent that travels.
Read the tool referenceCallers
Who calls this function, upward. Every answer labels its own incompleteness with explicit reasons — see the live example on Session memory.
Read the tool referenceReferences
All occurrences of a name — definition plus callers plus callees — for the full structural picture. Pair with Controller chains to add the decisions.
Read the tool referenceSearch
Substring match across function names, callers, callees, and constants. The fastest on-ramp when you know a fragment but not its home.
Read the tool referenceStatus
Index freshness, function and edge counts, and diagnostics. Run it first: a stale index cannot support positive claims.
Read the tool referenceWhat changed
Hermetic audit of what changed between two states of the code, with the affected functions and the tests that exercise them.
Read the tool referenceCall graph
Summary-first call graph around a function: the shape up front, full detail on zoom. The structural sibling of Controller chains.
Read the tool referenceImpact
Blast radius: transitive callers, affected state, and callees — the required gate before any edit. Pair with State ownership to see who owns what you touch.
Read the tool referenceHonest labels
What is real,
what is next.
- Every live example on the full pages is real MCP tool output captured against the DesignGraph codebase — never pseudo-output.
- The intent data layer is unpopulated in this build: intent labels require the labeling pass, which needs an LLM client. The machinery is source-verified and fixture-proven; the labels are not yet filled in.
- Session internals (chat references) and claim internals (verification pins) appear only as labeled internal-mechanism illustrations — never presented as tool output.
Start exploring