weavatrix 0.3.14 → 1.0.0

package/SECURITY.md

@@ -1,101 +0,0 @@
-# Security Policy
-
-## Supported versions
-
-| Version | Supported |
-|---|---|
-| `0.3.x` | Yes |
-| `0.2.x` and older | Upgrade to the latest release |
-
-Security fixes are provided for the latest published version of Weavatrix. Upgrade before reporting
-an issue that may already be resolved.
-
-## Reporting a vulnerability
-
-Please do not open a public issue for a suspected vulnerability. Use GitHub's private vulnerability
-reporting flow:
-
-https://github.com/sergii-ziborov/weavatrix/security/advisories/new
-
-Include the affected Weavatrix version, operating system, MCP client, enabled capability groups,
-reproduction steps, expected boundary, observed behavior, and potential impact. Avoid including real
-credentials, private source code, or other sensitive data in the report.
-
-Reports will be investigated privately. Public disclosure should be coordinated after a fix or
-mitigation is available so users have a reasonable opportunity to update.
-
-## Security boundaries
-
-The default `offline` MCP profile exposes no HTTP tools. Every standard source, manifest,
-configuration and coverage read is canonical-path-contained within the active repository and rejects
-traversal plus symlink or junction escapes. The optional malware dependency scan may inspect installed
-dependency caches such as GOPATH. `offline` permits repository switching only through an explicit
-local `open_repo` call; select `pinned` to remove that tool, the global repository listing, and
-cross-repository API tracing, holding a hard startup-repository boundary.
-
-Installed-package malware pattern matching is static heuristic evidence. It cannot confirm execution,
-package compromise, or credential exposure and therefore cannot emit `critical`; heuristic severity
-is capped at `high` with explicit `NOT_VERIFIED` runtime/origin/lockfile/exposure fields. Independently
-confirmed malicious-package advisories remain separate vulnerability evidence and may be critical.
-
-The JS/TS precision overlay is enabled by default for new graphs and runs the package-pinned `typescript-language-server` and
-TypeScript runtime as local child processes. It never resolves a repository executable, invokes
-`npx`, runs package scripts, or installs dependencies. Automatic type acquisition is disabled, and
-the provider receives an allowlisted OS/temp/locale environment with a constrained executable path;
-registry, proxy, cloud, token and `NODE_OPTIONS` values are excluded. The provider makes no
-Weavatrix HTTP request and Weavatrix transmits no source or evidence. TypeScript may still read
-locally declared project configuration, dependencies and type declarations; returned evidence is
-accepted only after repository realpath containment. Before spawning the provider, Weavatrix parses
-the applicable repo-contained TypeScript/JavaScript config chains with its bundled TypeScript API and
-refuses semantic mode when a configured language-service plugin, unresolved/outside config, or input
-safety limit is encountered. Config and configured-project inputs are fingerprinted and rechecked
-before cached evidence is used and after each provider run.
-This is an application/process boundary rather than an operating-system
-network sandbox. On stdio EOF, SIGTERM or SIGINT, the MCP server stops new providers, performs a
-bounded graph-work drain, and closes or tree-terminates TLS/tsserver before exiting.
-Set `WEAVATRIX_PRECISION=off` before server startup (or select `off` in the MCPB semantic-precision
-setting) to keep new graphs parser-only from their first build.
-
-The parser runtime and language grammars are also exact package-pinned dependencies. Before compiling
-WASM, Weavatrix requires the runtime and every selected grammar to resolve inside their dependency
-directories and match a release-pinned SHA-256 allowlist. Repository paths, URLs, symlinks and custom
-`runtimeWasm`/`wasmDir` overrides cannot enter this path. This specifically bounds the generated
-Emscripten dynamic-module loader in `web-tree-sitter`: although upstream contains `eval` glue for
-EM_ASM/EM_JS side-module exports, Weavatrix supplies only its known local runtime and grammar artifacts.
-An integrity mismatch fails the graph build instead of silently falling back to an unverified parser.
-
-Dependency scanners should distinguish a package's declared license from component notices.
-`typescript@5.9.3` declares `Apache-2.0`; its `ThirdPartyNoticeText.txt` also reproduces the W3C Community
-Final Specification Agreement for the Web Background Synchronization specification. That attribution
-does not change the npm package's declared license, but conservative scanners may report the embedded
-notice separately. `web-tree-sitter@0.25.10` declares MIT; its generated loader is covered by the
-artifact boundary above. These observations are security evidence, not a request to suppress third-party
-scanner warnings or legal advice.
-
-`verified_change` is read-only by default. Its optional targeted-test step can execute only an
-existing `package.json` script whose name matches the bounded test/check/verify allowlist. The call
-must set `run_tests:true` and the server operator must separately set
-`WEAVATRIX_ALLOW_TEST_RUNS=1`; otherwise no repository script runs. The tool accepts no command or
-shell string, caps scripts/arguments/time, rejects shell-sensitive argument characters, and launches
-the selected package-manager script with the credential-stripped child environment. Repository test
-scripts are repository-controlled code and may have arbitrary side effects, so enable this flag only
-for repositories and scripts you trust.
-
-The 0.3 MIT artifact contains no HTTP tool, sync endpoint/token setting, remote advisory refresher or
-compatibility alias that can enable one. `offline` and `pinned` are both network-free. Advisory
-matching reads only a local validated cache; a missing or stale cache remains `NOT_CHECKED` or
-`PARTIAL`, never a clean zero.
-
-The public extension API allows a dependent package to add MCP tools, packaged skills and local
-audit providers. Extension tools cannot replace a core tool, extension profiles cannot replace a
-core profile, and audit providers must declare `network:"none"`; failures remain explicit
-`ERROR/PARTIAL`. The separate `weavatrix-online` superset owns endpoint policy, consent,
-authentication, capability negotiation and every outbound request.
-
-Every MCP response includes transient local `_meta["weavatrix/metrics"]` timing, output-size/token
-estimate, graph freshness/revision/update and graph-cache status. Weavatrix does not persist, aggregate
-or transmit these measurements; they are response evidence for the caller, not product telemetry.
-
-Core exposes local-only services that build and validate the bounded source-free payload, validate
-connector-provided advisory records, create opaque repository identity and cache a validated
-architecture contract. These services read no connector credential and perform no network I/O.

package/docs/adr/0001-v0.3-offline-online-split.md

@@ -1,60 +0,0 @@
-# ADR 0001: Weavatrix 0.3 offline/online split
-
-Status: accepted; release candidate implemented
-
-Date: 2026-07-19
-
-## Decision
-
-The public `weavatrix` package remains MIT licensed and becomes a strictly
-offline MCP in 0.3. Its analyzers, graph, local LSP provider, cache, repository
-registry, Health, dependency, duplicate, impact, history and architecture tools
-remain available without a network capability.
-
-All outbound HTTP tools move to the separate public-source `weavatrix-online`
-package, starting at 0.1.0. It uses the permission-required Weavatrix Online
-Source License 1.0 and remains unpublished until the paired release gates pass.
-It extends the MIT engine rather than copying or relicensing it.
-
-The managed Cloud and licensed self-hosted Enterprise products are private
-server implementations of the same source-free wire contract. Neither is
-bundled into the offline package.
-
-## Tool ownership after extraction
-
-Offline owns local advisory matching but not remote refresh. Online owns:
-
-- `refresh_advisories`;
-- `preview_sync`;
-- `sync_graph`;
-- `pull_architecture_contract`;
-- endpoint capability negotiation and authentication.
-
-The exact final catalog is locked by tests before release. There is no legacy
-`online`, `hosted` or `full` alias in the offline artifact that can re-enable an
-HTTP implementation.
-
-## Update model
-
-`weavatrix-online` depends on a compatible `weavatrix` release and uses a
-supported extension API. Parser, graph schema, LSP and baseline Health fixes
-are therefore updated once in the MIT core. Online is an expanded superset: it
-may add proprietary MCP tools, packaged skills and local analyzer providers as
-well as transport, endpoint/authentication and consent workflows. It cannot
-replace or carry patched copies of core implementations.
-
-## Release gates
-
-The two public packages are released as a compatible pair only after:
-
-1. offline npm, MCPB, skill, server metadata and docs expose no HTTP tools or
-   hosted URL/token inputs;
-2. network action modules and fetch paths are absent from the offline package;
-3. the online connector passes the same local graph/Health regression suite
-   through its core dependency;
-4. Cloud and Enterprise pass one wire-contract suite;
-5. license metadata clearly distinguishes MIT core from permission-required,
-   source-available connector code;
-6. stale 0.2 processes fail clearly rather than advertising mixed catalogs.
-
-Existing MIT releases are unchanged and remain usable under MIT.

package/docs/agent-task-benchmark.md

@@ -1,58 +0,0 @@
-# Independent agent-change benchmark
-
-The 0.2.7 release includes two deliberately separate measurements.
-
-`npm run benchmark:agent` is a deterministic local routing microbenchmark. It measures whether the
-task retriever selects the expected symbol, plus false-positive selections, estimated output tokens,
-and local latency. It is a regression test, not evidence that an autonomous code change succeeded.
-
-The release comparison is an independently executed, blind agent-change benchmark. The evaluator
-must run the same pinned tasks and repository commits three times: with Weavatrix, with Codebase
-Memory, and with Serena. The product author must not adjudicate or repair a run while it is active.
-
-## Required task protocol
-
-Each task must provide a repository commit, natural-language request, allowed commands, timeout, and
-machine-verifiable acceptance tests. Keep task prompts and acceptance tests identical across systems;
-randomize system order. Record the entire agent input/output token count and wall-clock duration.
-
-A run is successful only when the requested behavior is implemented, all acceptance tests pass, and
-the final diff stays inside the task scope. A false positive is an evidenced warning, impacted symbol,
-or required edit that an evaluator confirms is unrelated to the task. Record a count, including zero;
-do not silently discard noisy output.
-
-Use a mix of at least:
-
-- behavior changes with indirect callers;
-- API route/client changes;
-- refactors with a duplicate or architecture regression trap;
-- changes whose nearest relevant test is not in the edited file;
-- one negative task where the correct action is to refuse or return unknown.
-
-## Result contract
-
-Save the blind evaluator output as JSON:
-
-```json
-{
-  "schemaVersion": "weavatrix.agent-change-results.v1",
-  "evaluator": "independent organization or person",
-  "systems": {
-    "weavatrix": {"runs": [{"taskId": "change-auth", "success": true, "falsePositives": 0, "tokens": 1234, "durationMs": 4567}]},
-    "codebase-memory": {"runs": [{"taskId": "change-auth", "success": true, "falsePositives": 1, "tokens": 1400, "durationMs": 5000}]},
-    "serena": {"runs": [{"taskId": "change-auth", "success": false, "falsePositives": 2, "tokens": 1600, "durationMs": 7000}]}
-  }
-}
-```
-
-Every system must contain the same unique task IDs. Every run must report success, false positives,
-tokens, and duration. Then run:
-
-```sh
-npm run benchmark:agent -- --independent-results path/to/results.json
-npm run benchmark:agent:release -- --independent-results path/to/results.json
-```
-
-The first command reports change-success rate, false positives per task, median tokens, and median
-duration for all three systems. The release command fails unless the independent comparison is
-complete and valid. Missing or malformed competitor data is `INCOMPLETE`, never a green comparison.

package/docs/releases/v0.2.19.md

@@ -1,33 +0,0 @@
-# Weavatrix 0.2.19
-
-0.2.19 is a supply-chain signal-precision patch. It was produced after the complete Hosted
-installed-package scan exposed false positives in otherwise ordinary dependency content.
-
-## Malware evidence precision
-
-- A hardcoded external URL is no longer sufficient for a malware finding.
-- A standalone `fetch`, Axios, request, curl or wget call to an ordinary external URL is also weak
-  evidence by itself. It remains attached as bounded co-evidence when the same package has a
-  separate security signal, such as whole-environment serialization.
-- Strong URL and execution evidence is unchanged: known exfil/webhook hosts, public raw-IP URLs,
-  cloud metadata endpoints, download-plus-exec lifecycle scripts, reverse shells and miners still
-  produce review findings.
-- Unicode variation selectors (`U+FE00`–`U+FE0F`) are presentation controls commonly used with
-  emoji and no longer trigger hidden-Unicode review. Bidirectional override/isolate controls remain
-  covered because they can conceal source order.
-
-## Verification
-
-- Synthetic end-to-end fixtures prove ordinary URL literals, standalone network calls and variation
-  selectors stay quiet while multi-signal and strong malicious families remain detectable.
-- Real Weavatrix Hosted dogfood checked 651/651 pinned package versions through OSV with no known
-  advisories and scanned 535 installed npm packages with zero malware findings.
-- The same audit reports complete npm declaration/import verification, zero missing dependencies,
-  zero unused dependencies, zero duplicate declarations and zero runtime cycles.
-- Full release tests, release-manifest checks, npm/MCPB pack/runtime smoke and supported Node CI remain
-  mandatory before publication.
-
-## Product boundary
-
-The offline package remains MIT. The accepted 0.3 split and the separate permission-licensed online
-connector are unchanged by this patch.

package/docs/releases/v0.3.0.md

@@ -1,43 +0,0 @@
-# Weavatrix 0.3.0
-
-0.3.0 makes the public MIT package a strictly network-free core and introduces
-the supported extension boundary used by the separately versioned
-`weavatrix-online` product.
-
-## Trust boundary
-
-- The 34 core tools are local. Remote OSV refresh, endpoint authentication,
-  architecture pull and graph sync were removed from this package.
-- Legacy `online`, `osv`, `hosted` and `full` profile names fail loudly. They
-  cannot re-enable an outbound request implementation.
-- The offline artifact gate scans the packed source boundary and rejects an
-  HTTP implementation or Hosted credential/config surface.
-- Extensions compose in-process through a stable API. They may add tools,
-  skills and local audit providers without copying or patching the core.
-
-## Critical precision fixes
-
-- Fresh dogfood of this repository produced 58 persisted `EXACT_LSP` edges.
-  An on-demand `get_dependents(startMcpServer)` query returned its direct caller
-  as `EXACT_LSP`; zero-edge `UNAVAILABLE` is no longer the expected JS/TS path.
-- `Cargo.lock` v3/v4 registry records become exact crates.io pins for offline
-  OSV matching. A saved `cargo audit --json` report is also imported as native
-  RustSec vulnerability/warning evidence; missing, invalid or stale reports
-  remain visibly incomplete.
-- Maven/Gradle import-to-artifact evidence now reads exact class ownership from
-  already installed JAR central directories without executing Java or a build.
-  A missing JAR, unresolved property/catalog expression, ambiguity or heuristic
-  fallback keeps the ecosystem `PARTIAL` instead of falsely claiming `COMPLETE`.
-- Cross-repository contracts combine static HTTP/GraphQL/gRPC/event models with
-  source-free `weavatrix.transport-runtime.v1` reports. Normalized or OTLP JSON
-  observations resolve an exact dynamic GraphQL field, gRPC method or
-  Kafka/event destination only when the report is fresh, revision-matched,
-  capture-complete and source-correlated. Missing or unobserved runtime paths
-  remain explicit `UNKNOWN`, never false matches.
-
-## Product family
-
-`weavatrix` remains the MIT base of the whole product family. The public-source,
-separately licensed `weavatrix-online` 0.1.0 package depends on this release and
-may add Online-only tools, skills and analyzers. Managed Cloud and licensed
-self-hosted servers implement the same source-free wire contract.

package/docs/releases/v0.3.1.md

@@ -1,30 +0,0 @@
-# Weavatrix 0.3.1
-
-0.3.1 fixes a dogfood-discovered blind spot in symbol-aware change impact when
-one generated or snapshot asset makes the aggregate Git diff exceed its byte
-budget.
-
-## Bounded large-diff recovery
-
-- `change_impact` first keeps its normal bounded zero-context diff.
-- If that aggregate diff is oversized, it obtains the complete bounded changed-
-  file list and recovers zero-context diffs one file at a time under the same
-  global byte and time budgets.
-- Source files that fit retain line and symbol classification. Only individual
-  oversized, timed-out or otherwise unreadable files become conservative
-  `unknown` seeds.
-- The result remains `PARTIAL/HIGH` while any file lacks textual evidence; the
-  fallback improves useful evidence without turning unknown into clean.
-
-## Exact semantic evidence
-
-The global TypeScript/JavaScript LSP overlay is a bounded prewarm and may report
-`PARTIAL` on large repositories. Exact workflows are not limited by that global
-cap: `get_dependents` and `inspect_symbol` use revision-bound point queries, and
-`change_impact` uses one bounded exact batch for changed symbols.
-
-On the Hosted 0.3 release commit, the repaired fallback recovered 27/27 changed
-paths and 57 mapped symbol changes, reduced the conservative seed set from the
-1,000 cap to 72, and produced `DIRECT_EXACT_TRANSITIVE_GRAPH` with 16/16 selected
-JS/TS symbols verified by `typescript-language-server`. The remaining large
-snapshot assets stayed explicit `unknown`; transitive hops stayed graph-backed.

package/docs/releases/v0.3.10.md

@@ -1,76 +0,0 @@
-# Weavatrix 0.3.10
-
-0.3.10 closes the honesty gaps a production evaluation surfaced: Rust inline
-tests are no longer reported as production code, endpoint tools are
-production-first everywhere, and semantic-precision statuses say exactly how
-much was actually proven. Every fix landed with regression tests and survived
-an adversarial multi-agent review of the full changeset.
-
-## Node-level Rust test classification
-
-Rust puts unit tests inside production files, so path classification alone
-misreported them everywhere. The extractor now proves test-only compilation —
-`#[cfg(test)]` / `#[cfg(all(test, ...))]` on items, impl blocks and modules,
-`#[test]`/`#[tokio::test]`-style harness attributes and `#[bench]` functions,
-inner `#![cfg(test)]` attributes at file and module-body scope, and attributes
-separated from their item by comments — and stamps a `test_surface` flag on the
-node. Deliberately conservative: `#[cfg(any(test, ...))]` also ships in
-production builds and stays unclassified.
-
-Every consumer honors the flag like a path-classified test: `find_dead_code`
-(candidates and test-only consumer resolution), `query_graph` seeds and
-traversal, `hot_path_review`, `god_nodes`, `verified_change` task anchors,
-`no-tests`/`tests-only` build modes, hosted sync payloads and evidence
-snapshots. A test-term question or `include_tests`/`include_classified` opts
-back in. `extractorSchemaV` 5 → 6; cached graphs rebuild once, automatically.
-
-## Production-first endpoint tools
-
-- `list_endpoints` suppresses endpoints declared in classified
-  test/e2e/generated/mock/story/docs/benchmark/temp paths by default, with a
-  count note and `include_classified` opt-in (auto-enabled on tests-only
-  graphs). Opted-in rows carry `[classified:...]` tags.
-- `trace_endpoint` filters classified handler twins before its ambiguity
-  check, ranks remaining same-named candidates by the import binding at the
-  route file (below same-directory proximity — a file-level edge proves
-  coupling, not identity), and reports `importBoundAtRoute`/`pathClasses` per
-  candidate when it still fails closed.
-- New `handler_file` parameter resolves genuinely symmetric candidates: it
-  narrows the full candidate pool before scoring, prefers exact repo-relative
-  matches over path-suffix matches, and compares case-insensitively.
-
-## Honest semantic-precision statuses
-
-`COMPLETE` previously meant only "the bounded request finished", producing
-`COMPLETE (0 EXACT_LSP edges)` on Rust repositories and `COMPLETE; exact
-absence was not proven` in `get_dependents`. Now:
-
-- A new `NONE` state covers the zero-eligible-targets outcome; `UNAVAILABLE`
-  stays reserved for provider failure. `PRECISION_OVERLAY_V` 4 → 5 discards
-  old-vocabulary sidecars.
-- `COMPLETE` requires every selected target actually probed.
-- `open_repo` and the build log print the `graph_stats`-style coverage
-  fraction: `PARTIAL — 32/1373 bounded target(s) queried (truncated)`.
-- `get_dependents` never pairs a state word with a contradicting note, and
-  keeps `NONE`/`UNAVAILABLE` truthful instead of collapsing them into
-  `PARTIAL`.
-
-## Output fidelity
-
-- `trace_api_contract` records per-endpoint method mismatches with sample
-  callers and always shows a mismatch block, so `*_WITH_METHOD_MISMATCHES`
-  verdicts stay backed by visible rows under any `top_n`.
-- Structure findings and hosted evidence snapshots stop reporting idiomatic
-  Rust `mod.rs`/`lib.rs`/`main.rs` parent-child module-tree cycles as
-  compile-time coupling (counted as `rustModuleTreeCouplings`; genuine
-  cross-directory cycles still report).
-- `query_graph` edge lines and `shortest_path` hops widen repeated or generic
-  file basenames (`mod.rs`, `index.*`, `__init__.py`) to their last two path
-  segments.
-
-## Documentation
-
-The always-fresh graph contract — debounced Git freshness probe plus bounded
-incremental refresh on every graph/health call, no watcher process required —
-is now documented in the README and the agent skill, together with the Rust
-test-surface classification rules.

package/docs/releases/v0.3.11.md

@@ -1,55 +0,0 @@
-# Weavatrix 0.3.11
-
-0.3.11 teaches the graph two new surfaces. SQL schemas become first-class graph
-citizens with honest, evidence-gated dead-code verdicts, and Solidity joins the
-parsed languages at zero dependency cost. The weavatrix.com site moved to its
-own repository, so the engine repo now ships engine only.
-
-## SQL: the graph reaches the database schema
-
-No SQL grammar ships in the pinned tree-sitter-wasms, so `.sql` indexing runs
-on a dependency-free statement scanner (a new `textOnly` language-module class)
-built for the graph's needs — symbols and references, not SQL semantics.
-String literals and comments are blanked before structural scanning, so quoted
-text can never fabricate a reference.
-
-- `CREATE TABLE`/`VIEW`/`FUNCTION`/`PROCEDURE`/`INDEX`/`TRIGGER` become graph
-  symbols; table columns (including `ALTER TABLE ADD COLUMN`) are member
-  symbols with `member_of`, and tables carry their columns as `field_types`.
-- References resolve inside `.sql` files (a view's `SELECT`, an index's `ON`
-  table, a trigger's `EXECUTE FUNCTION`, foreign-key `REFERENCES`) — and, the
-  real point, from application code: literal SQL found in string literals of
-  any indexed language links the enclosing function to the table it queries.
-  `change_impact` and `get_dependents` on a table now answer "which code
-  touches this?"
-- Dead-code verdicts stay honest. Schema objects are judged only when the
-  repository demonstrably uses literal SQL the scanner can read — an
-  ORM-driven repo produces silence, not guesses. Columns of a
-  `SELECT *`-consumed table are never judged by name, and indexes/triggers
-  (DB-engine surface) are never judged at all. A flagged column reports
-  "no SQL statement in the indexed sources references it".
-
-## Solidity
-
-The tree-sitter-solidity grammar already shipped inside the pinned
-tree-sitter-wasms package — 0.3.11 registers it (and pins its SHA-256 in the
-parser-artifact allowlist), so Solidity support adds zero new dependencies.
-Contracts, interfaces, libraries, functions, constructors, modifiers, events,
-errors, structs, enums and state variables are extracted with visibility and
-membership; `is` inheritance, modifier invocations, `emit`, `new Contract()`
-and `using X for Y` all feed the call/heritage resolvers. Imports resolve
-through relative paths, Foundry remappings (`remappings.txt` / `foundry.toml`)
-and root-anchored specs; npm-style specs (`@openzeppelin/...`) are recorded as
-external dependencies. `.sol` files share directory scope like Go/C#/Rust,
-which makes Solidity's plain `import "./Base.sol"` — a statement that names no
-symbols — resolvable for siblings without guessing.
-
-`extractorSchemaV` 6 → 7; cached graphs rebuild once, automatically.
-
-## Repository
-
-The weavatrix.com landing site (pages, Cloudflare Worker config, deploy
-workflow and its asset checks) moved byte-identically to the separate
-`weavatrix-site` repository, removing ~1.2 MB of page assets from the engine
-repo. The MCPB icon the site used to own is vendored at `mcpb/icon.png`; the
-published npm package contents are unchanged.

package/docs/releases/v0.3.12.md

@@ -1,44 +0,0 @@
-# Weavatrix 0.3.12
-
-0.3.12 makes the read-only core the provably-pure base of a layered stack. The
-core stays exactly what it was for its users — the same 34 offline analysis tools
-— but it now exposes a small read-only analysis surface for the packages built on
-top of it, and machine gates prove it can neither modify code nor reach anything
-external. Rust cross-crate call resolution and a `query_graph` seeding fix round
-out the release.
-
-## The read-only base of the stack
-
-Weavatrix now anchors a stack of growing responsibility:
-`weavatrix-online ⊃ weavatrix-refactor ⊃ weavatrix`. You install one package and
-get everything below it. This release adds only what the base owes the layers
-above; it registers no new tools and changes no existing tool.
-
-- New `weavatrix/analysis-kit` export: a narrow, read-only surface (graph
-  identity, cycle detection, architecture verification, the bundled tree-sitter
-  grammars, the read-only TypeScript language-server client, dead-code signals)
-  that upper-layer packages compose. Nothing here writes, applies, or describes
-  an edit.
-- Two release gates make the base's guarantees machine-checked, alongside the
-  existing ADR 0001 no-network gate: the core source contains **no edit-plan,
-  applier, or refactoring-builder marker**, and **no literal URL** at all. A
-  vulnerability finding references the advisory id rather than an external link;
-  the URL-matching malware heuristics are matchers over scanned code, not links
-  the core follows. Installing the base alone can neither modify your code nor
-  reach the network.
-
-## Rust cross-crate call resolution
-
-Rust call edges now resolve across crates, inline-test coverage is reported
-honestly, and same-crate call noise is reduced — cross-crate `references` edges
-land where a caller reaches a definition in another crate.
-
-## query_graph seeding
-
-A `query_graph` question that reduced entirely to stop words (a bare
-`architecture`), or inferred a language absent from the repository (`contract`
-inferring Solidity where there are no `.sol` files), previously returned a
-misleading "No nodes matched". Both now fall back only when they would otherwise
-return nothing: the stop filter relaxes when every token is a stop word, and a
-requested language that matches zero nodes is dropped. Queries that already
-worked are unchanged.

package/docs/releases/v0.3.13.md

@@ -1,34 +0,0 @@
-# Weavatrix 0.3.13
-
-0.3.13 fixes two misleading verification and graph-path signals found while
-dogfooding a real duplicate-removal refactor. It also bounds cross-repository
-contract responses with revision-bound pagination.
-
-## Honest refactor verification
-
-- `verified_change` no longer reports an enabled duplicate ratchet as
-  `SKIPPED: duplicate ratchet disabled` during the planning phase. An enabled
-  plan now reports `PLANNED`, `enabled: true`, and states that the ratchet runs
-  during `phase="verify"`. An explicit `duplicate_ratchet: false` remains the
-  only disabled state.
-- `shortest_path` remains an undirected connectivity query, but every rendered
-  hop now preserves the stored graph-edge direction. Reverse traversal renders
-  `<--imports--` or `<--calls--` instead of inventing a forward dependency.
-  Repeated and generic file names are widened in both the header and hop lines.
-
-## Bounded cross-repository contract evidence
-
-- `trace_api_contract` response schema v4 returns compact, paginated evidence
-  by default. `page_size`, `per_item_limit`, and a revision/filter-bound opaque
-  `cursor` keep responses bounded; `response_detail="full"` remains an explicit
-  opt-in and is still paginated.
-- Invalid, stale, or cross-query cursors fail closed with `INVALID_CURSOR`.
-- Repository graph reconciliation now reuses a persisted fresh graph before
-  invoking a rebuild, while stale or failed refreshes retain their explicit
-  completeness status.
-
-## Layer compatibility
-
-The read-only core keeps its 34-tool surface and `weavatrix.edit-plan.v1`
-contract unchanged. `weavatrix-refactor` 0.1.1 and `weavatrix-online` 0.2.0
-are the paired upper-layer releases for this core version.

package/docs/releases/v0.3.14.md

@@ -1,37 +0,0 @@
-# Weavatrix 0.3.14
-
-0.3.14 is a Repo Lens dogfood hardening release. It keeps the 34-tool,
-network-free Core surface while making large JavaScript graphs practical for
-UI consumers and reducing production-analysis noise.
-
-## Structured results and performance
-
-- `find_duplicates` and ordinary `run_audit` results now include bounded
-  structured groups/findings alongside their existing text summaries, so MCP
-  clients can render native views without parsing prose.
-- Local symbol references are indexed once per file and declaration ranges are
-  subtracted with binary search instead of rescanning the whole file for every
-  symbol. On the 8,117-node Repo Lens graph, aggregation fell from more than
-  64 seconds to about 0.93 seconds while preserving graph totals.
-
-## Production-first precision
-
-- `vendor`, `vendored`, and `third-party` paths are classified explicitly and
-  suppressed consistently across duplicates, audit, endpoints, context,
-  architecture, dead-code, hot-path, and graph-query surfaces.
-- Electron preload files are treated as entry points only when another source
-  contains a bounded `BrowserWindow` `preload:` option.
-- Endpoint extraction masks JavaScript regex literals and comments correctly
-  and recognizes bounded `node:http` method plus `url.pathname` guards. Repo
-  Lens now reports its four production control routes without comment/test
-  false positives in the production view.
-
-## Coverage interoperability
-
-- `coverage_map` reads Cargo Tarpaulin's official JSON report shape, including
-  component paths and `traces[].stats.Line`, and advertises both Tarpaulin JSON
-  and `cargo llvm-cov` LCOV workflows for Rust repositories.
-
-This Core release is paired with `weavatrix-refactor` 0.1.2 and
-`weavatrix-online` 0.2.0. Each layer keeps its own package, license, version,
-and release provenance.

package/docs/releases/v0.3.2.md

@@ -1,15 +0,0 @@
-# Weavatrix 0.3.2
-
-0.3.2 carries the bounded large-diff recovery introduced in 0.3.1 and makes
-its byte-limit detection consistent across supported Node.js platforms.
-
-Some `spawnSync` implementations can return a successful process status while
-the captured Git diff already exceeds the configured byte budget. Weavatrix now
-checks the measured output length before accepting the aggregate diff, rather
-than treating process status alone as proof that the evidence is complete.
-
-The same fail-closed behavior remains in force: per-file diffs that fit keep
-line/symbol classification, individually oversized files remain `unknown`, and
-the overall result stays `PARTIAL/HIGH` until every changed file has bounded
-textual evidence. Hosted dogfood continues to verify 16/16 selected changed
-JS/TS symbols with exact LSP direct-reference evidence.

package/docs/releases/v0.3.3.md

@@ -1,25 +0,0 @@
-# Weavatrix 0.3.3
-
-0.3.3 hardens the portable parser boundary after reviewing two conservative
-third-party supply-chain alerts.
-
-`web-tree-sitter@0.25.10` is MIT-licensed, but its generated Emscripten loader
-contains `eval` support for EM_ASM/EM_JS exports in dynamic side modules. The
-same generated path remains in `web-tree-sitter@0.26.11`; changing the version
-would not remove the signal. Weavatrix now verifies the exact SHA-256 of the
-package-owned runtime and every selected `tree-sitter-wasms@0.1.13` grammar
-before compilation. Parser artifacts must resolve inside their pinned package
-directories. Repository paths, URLs, symlink escapes and custom runtime or
-grammar locations are rejected, and an integrity mismatch fails closed.
-
-The release also documents a separate license-classification alert.
-`typescript@5.9.3` declares `Apache-2.0`. Its third-party notice reproduces the
-W3C Community Final Specification Agreement for the Web Background
-Synchronization specification; conservative scanners can identify that
-component notice as non-permissive even though it is not the package's declared
-license. Weavatrix keeps the notice visible and does not suppress the external
-scanner finding.
-
-The public site now presents the product boundary directly: the 34-tool MIT,
-network-free Core and the separately licensed 39-tool Online overlay, including
-managed Cloud and customer-hosted Enterprise availability.