v0.5.0

Changed

• map and graph are merged into one MCP tool. Pass path for a
  structural outline or symbol for relationships (callers, callees,
  importers, impact, path, explore) — never both. read and status are no
  longer separate MCP tools (an agent's own file reader and the automatic
  per-call refresh already cover them); both remain available from the shell
  as cgraph read and cgraph stats. MCP responses are no longer truncated
  to a default token budget — every tool still accepts an explicit budget
  to cap output on request.

Removed

• The docs tool and dependency-doc extraction (src/deps/). Reading
  node_modules/site-packages/NuGet/Maven and fetching registry artifacts to
  build a usage-ranked API reference was a large, separately-maintained
  subsystem better served by a dedicated docs MCP server (e.g.
  Context7). cgraph docs, the MCP
  docs tool, and openProject(...).dependencies() are gone; map/graph
  still show external call edges from the existing import graph.

Added

• Images are recognized by extension and never read. .png, .ico,
  .jpg, and similar formats used to be read fully into memory and hashed
  before the binary sniff threw them away, which made index/update look
  stuck on large icon sets or design assets for no reason. They now take the
  same stat-only fast path as oversized files, tagged with a distinct
  image skip reason (visible in update's "skipped by reason" and in
  doctor) instead of being folded into the generic binary count.
• Unparsed files are findable again. find only ever searched symbols,
  so a file with none — an image, a binary, anything skipped as too-large,
  minified, or generated — was invisible to it even though the index knew it
  existed; map already listed such files but didn't say why they weren't
  parsed. find <name> now also matches on the path of unparsed files, map
  annotates each with its skip reason (e.g. (image), (too-large)) instead
  of a bare (not parsed), and graph gives a specific error when a target
  is an indexed file with no symbols rather than reporting it as unknown.

---

Install

npm install -g cgraph

Published to npm as cgraph@0.5.0 with build provenance.

v0.4.0

Added

• JSON, CSS, HTML, and Bash language packs. JSON object keys and CSS
  rules/keyframes become nested symbols the same way code does; HTML surfaces
  landmark tags, <script>/<style> blocks, and any element with an id
  (narrow by design — every <div> as a symbol would be noise); Bash extracts
  function definitions and resolves calls between them. A YAML pack ships
  alongside these but is not yet registered: the bundled tree-sitter grammar
  fails to parse under the pinned web-tree-sitter version, so YAML files
  still fall back to being indexed as unparsed stubs until that combination is
  fixed upstream.
• packs.strict config option. Off by default. A project can turn off the
  "load an unpredicted language on first sight" fallback, so only packs for
  languages its own manifests actually named are ever used — a vendored
  script in an unexpected language becomes a stub instead of silently pulling
  in a grammar+pack for it.
• cgraph init pre-approves its own MCP tools for Claude Code. Previously
  init registered the MCP server and wrote the tool-usage instructions, but
  the tools still sat behind a permission prompt in the checked-in project
  config — fine for an interactive session where a human clears it once, but
  a subagent spawned non-interactively has nobody to click "allow", so a
  denied call looked exactly like a missing tool and the model fell back to
  grep. init now also writes the approval into .claude/settings.json
  (shared, not the gitignored settings.local.json), additively and without
  touching anything a user already set there.

---

Install

npm install -g cgraph

Published to npm as cgraph@0.4.0 with build provenance.

v0.3.0

Added

• Sub-project detection. A root that bundles several independently-cloned
  repos — a frontend/backend/desktop fleet of microservices, each its own git
  checkout, opened as one workspace — now has every file labeled with which
  nested repo it belongs to. cgraph init and cgraph map report each
  detected sub-project alongside its own stack and frameworks, so an agent
  working across the fleet can tell an Angular file from an ASP.NET file at a
  glance instead of treating the whole tree as one undifferentiated project.
  A nested repo's own local .git/info/exclude is now honored too, the same
  way .gitignore and .cgraphignore already were. Off switch:
  detectSubprojects: false.

---

Install

npm install -g cgraph

Published to npm as cgraph@0.3.0 with build provenance.

v0.2.5

• Release v0.2.5 (dfc9e14)
• fix: ambiguous resolution ignoring rank, false staleness warning, hooks corruption, overstated caller confidence (551be54)

---

Install

npm install -g cgraph

Published to npm as cgraph@0.2.5 with build provenance.

v0.2.4

• Release v0.2.4 (bfb1d10)
• fix: correct stale command/tool-count references in error messages (26c502a)

---

Install

npm install -g cgraph

Published to npm as cgraph@0.2.4 with build provenance.

v0.2.3

• Release v0.2.3 (e19df1e)
• fix: generated agent instructions pointed to a nonexistent status command (a8e58da)

---

Install

npm install -g cgraph

Published to npm as cgraph@0.2.3 with build provenance.

v0.2.2

• Release v0.2.2 (c86f0b5)
• Merge branch 'main' of https://github.com/Ramachandrajoshi/Code-Graph (dac09c1)
• Release v0.2.1 (f2256a3)
• Merge pull request #1 from Ramachandrajoshi/claude/ci-build-failures-hrcnsg (96bde5b)
• test: skip flaky Maven integration assertion on transient fetch miss (d0cf71a)
• test: skip Maven network test when repo endpoint is unreachable (09dd682)
• Stabilize network-dependent archive integration tests (648e63e)
• Fix MCP shutdown resolve value and add in-flight close regression test (66192cc)
• Fix MCP server exiting before in-flight tool calls finish responding (303e55b)

---

Install

npm install -g cgraph

Published to npm as cgraph@0.2.2 with build provenance.

v0.2.0

Added

• Dependency docs are fetched when not installed locally. Every registry
  publishes the same artifact cgraph reads off disk, so it is downloaded and run
  through the same parsers: npm .tgz (falling back to @types/<pkg>, since a
  large share of npm ships no types of its own), NuGet .nupkg, Maven
  -sources.jar, PyPI wheels. No scraping and no third-party service — the
  result is version-exact because it is the published artifact.
• Maven dependency docs, local and remote, now that a zip reader exists.
• cgraph docs --refresh --guides fetches llms.txt
  where a project publishes one. Package archives carry API reference because
  that is what is in the source; they cannot carry setup guides, which were
  never in the source. This is the only prose source, and coverage is partial.
• tar.gz and zip readers built on node:zlib, with no new dependencies.
• C# language pack. Types, members, properties, namespaces, XML doc
  summaries and access modifiers, with using resolved to the file declaring
  that namespace. LINQ and BCL calls are classified as runtime built-ins rather
  than swamping the unresolved list.
• Project technology discovery. Manifests (*.csproj, package.json,
  pom.xml, pyproject.toml, go.mod, Cargo.toml, …) are read before any
  parsing, so only the packs a project actually needs are loaded and only their
  grammars are fetched. Frameworks — Angular, React, ASP.NET, Entity Framework,
  Spring, Django, FastAPI — are detected from every manifest in the tree, not
  just the root. A file in an unpredicted language still loads its pack on
  sight: discovery decides what loads eagerly, never what gets ignored.
• .NET dependency documentation. NuGet ships compiled assemblies, so there
  is no source to parse — but packages carry an XML documentation file beside
  the DLL with every public member and its summary. docs now reads it,
  preferring the newest target framework present.
• Local embedding models. provider: "local" targets any OpenAI-compatible
  endpoint (Ollama, LM Studio, llama.cpp, vLLM) with baseUrl, model and
  optional dimensions. Nothing leaves the machine and nothing costs money. The
  returned vector width is verified against dimensions on the first batch,
  because a silent mismatch corrupts every similarity score in a way that is
  close to untraceable.
• The index refreshes itself. The MCP server checks for changes before
  answering, so an agent never reads a graph that disagrees with the working
  tree — after an editor save, a git checkout, a rebase, or another agent's
  edit. No watcher, no hook, nothing to remember. Throttled (3s) so a burst of
  tool calls costs one scan, and disabled with
  autoRefresh.enabled: false.
• cgraph hooks install — pre-warms the index after checkout, merge and
  rebase, the operations that change hundreds of files at once. Appends between
  markers and never replaces an existing hook; backgrounded so it cannot delay
  a git command. post-commit is excluded by default because committing does
  not change the working tree.
• cgraph init --agent for Claude Code, GitHub Copilot, opencode, Cursor and
  Windsurf, writing both the MCP registration and an instruction block telling
  the agent to use these tools instead of grep.

Changed

• No token-savings multiplier is published any more. The previous figures
  compared cgraph against an invented grep-then-read baseline: an agent opens
  every file grep matched, capped at ten, and reads each whole. No agent was
  ever run. Change the cap, assume line windows instead of whole files, or
  assume two probes instead of one, and the ratio moves by an order of
  magnitude — the assumption determined the number, not the measurement.
  Responses now report their own size, and where a concrete comparable exists
  (the file an outline describes) both numbers are shown as facts with no
  arithmetic between them. bench/ remains as a regression guard on response
  sizes, comparable only to itself.
• Freshness checks no longer read the repository. A file whose size and
  mtime match the index is skipped without being opened; only files that fail
  that check are read and hashed. On llama.cpp a no-op pass drops from ~520ms
  to ~170ms, which is what makes automatic refresh affordable. The content hash
  still decides, so a file touched but not edited is read once and correctly not
  re-parsed.

Breaking

• SavingsLedger is renamed UsageLedger in the programmatic API, and records
  tokens_returned / tokens_source instead of tokens_baseline /
  tokens_saved. The old name asserted a saving the tool cannot observe. Only
  affects code importing it from cgraph; the CLI and MCP surfaces are
  unchanged.
• cgraph stats no longer prints a reduction factor, and cgraph doctor no
  longer prints "tokens saved". Anything parsing that output needs updating.

---

Install

npm install -g cgraph

Published to npm as cgraph@0.2.0 with build provenance.

v0.1.1

Fixed

• Indexing no longer fails on Node builds without SQLite FTS5. FTS5 is a
  compile-time option and Node's bundled SQLite omits it in many builds
  (22.14 and 23.11 among them), so creating the full-text tables in a migration
  aborted startup entirely with no such module: fts5. Availability is now
  probed at runtime: with FTS5 the index uses it, without it search falls back
  to exact, prefix, trigram and substring matching over names, signatures and
  docs. cgraph doctor reports which mode is active, and a later upgrade to a
  Node that has FTS5 is picked up automatically and backfilled — no re-index.

---

Install

npm install -g cgraph

Published to npm as cgraph@0.1.1 with build provenance.