sigmap 6.5.0 → 6.5.2

package/AGENTS.md

@@ -47,6 +47,26 @@ Use this marker block for all appendable context files:
 <!-- Updated by gen-context.js -->
 # Code signatures
 
+## SigMap commands
+
+| When | Command |
+|------|---------|
+| Before answering a question | `sigmap ask "<your question>"` |
+| After code changes | `sigmap validate` |
+| To query by topic | `sigmap --query "<topic>"` |
+
+Always run `sigmap ask` or `sigmap --query` before searching for files relevant to a task.
+## changes (last 5 commits — 2 days ago)
+```
+src/config/loader.js                          +_legacyDetectAutoSrcDirs  ~detectAutoSrcDirs
+src/discovery/source-root-resolver.js         +resolveSourceRoots  +_detectMonorepo  +_enumerateCandidates  +_applySpecialRules
+src/discovery/language-detector.js            +detectLanguages  +_walkDepth
+src/discovery/source-root-scorer.js           +getRecentlyChangedDirs  +scoreCandidate  +_countSourceFiles
+src/discovery/framework-detector.js           +detectFrameworks  +_readDeps  +_readFile  +_existsAnywhere
+src/discovery/sigmapignore.js                 +loadIgnorePatterns  +matchesIgnorePattern
+src/retrieval/ranker.js                       +_computePenalty  ~scoreFile  ~rank  ~buildSigIndex
+```
+
 ## packages
 
 ### packages/cli/index.js
@@ -104,15 +124,7 @@ function score(cwd) → { * score: number, * grad
 function adapt(context, adapterName, opts = {}) → string
 ```
 
-### packages/adapters/codex.js
-```
-module.exports = { name, format, outputPath, write }
-function format(context, opts = {}) → string
-function outputPath(cwd) → string
-function write(context, cwd, opts = {})
-```
-
-### packages/adapters/claude.js
+### packages/adapters/copilot.js
 ```
 module.exports = { name, format, outputPath, write }
 function format(context, opts = {}) → string
@@ -121,13 +133,12 @@ function outputPath(cwd) → string
 function write(context, cwd, opts = {})
 ```
 
-### packages/adapters/copilot.js
+### packages/adapters/cursor.js
 ```
-module.exports = { name, format, outputPath, write }
+module.exports = { name, format, outputPath }
 function format(context, opts = {}) → string
 function _confidenceMeta(opts)
 function outputPath(cwd) → string
-function write(context, cwd, opts = {})
 ```
 
 ### packages/adapters/gemini.js
@@ -139,14 +150,6 @@ function write(context, cwd, opts = {})
 function _confidenceMeta(opts)
 ```
 
-### packages/adapters/cursor.js
-```
-module.exports = { name, format, outputPath }
-function format(context, opts = {}) → string
-function _confidenceMeta(opts)
-function outputPath(cwd) → string
-```
-
 ### packages/adapters/openai.js
 ```
 module.exports = { name, format, outputPath }
@@ -163,74 +166,24 @@ function _confidenceMeta(opts)
 function outputPath(cwd) → string
 ```
 
-## src
-
-### src/security/patterns.js
-```
-module.exports = { PATTERNS }
-```
-
-### src/security/scanner.js
-```
-module.exports = { scan }
-function scan(signatures, filePath) → { safe: string[], redacte
-```
-
-### src/extractors/cpp.js
-```
-module.exports = { extract }
-function extract(src) → string[]
-function extractBlock(src, startIndex)
-function extractMembers(block)
-function normalizeParams(params)
-function normalizeType(type)
-```
-
-### src/extractors/csharp.js
-```
-module.exports = { extract }
-function extract(src) → string[]
-function extractBlock(src, startIndex)
-function extractMembers(block)
-function normalizeParams(params)
-function normalizeType(type)
-```
-
-### src/extractors/dart.js
-```
-module.exports = { extract }
-function extract(src) → string[]
-function extractBlock(src, startIndex)
-function extractMembers(block)
-function normalizeParams(params)
-```
-
-### src/extractors/deps.js
+### packages/adapters/claude.js
 ```
-module.exports = { extractPythonDeps, extractTSDeps, buildReverseDepMap }
-function extractPythonDeps(src) → string[]
-function extractTSDeps(src) → string[]
-function buildReverseDepMap(forwardMap) → Map<string, string[]>
+module.exports = { name, format, outputPath, write }
+function format(context, opts = {}) → string
+function _confidenceMeta(opts)
+function outputPath(cwd) → string
+function write(context, cwd, opts = {})
 ```
 
-### src/extractors/go.js
+### packages/adapters/codex.js
 ```
-module.exports = { extract }
-function extract(src) → string[]
-function extractBlock(src, startIndex)
-function extractInterfaceMethods(block)
-function normalizeParams(params)
+module.exports = { name, format, outputPath, write }
+function format(context, opts = {}) → string
+function outputPath(cwd) → string
+function write(context, cwd, opts = {})
 ```
 
-### src/extractors/java.js
-```
-module.exports = { extract }
-function extract(src) → string[]
-function extractBlock(src, startIndex)
-function extractMembers(block)
-function normalizeParams(params)
-function normalizeType(type)
-```
+## src
 
 ### src/extractors/javascript.js
 ```
@@ -530,11 +483,6 @@ function formatAnalysisTable(stats, showSlow) → string
 function formatAnalysisJSON(stats) → object
 ```
 
-### src/config/defaults.js
-```
-module.exports = { DEFAULTS }
-```
-
 ### src/format/dashboard.js
 ```
 module.exports = { generateDashboardHtml, renderHistoryCharts, computeExtractorCoverage, percentile, overBudgetStreak }
@@ -622,18 +570,6 @@ function queryContext(args, cwd)
 function getImpact(args, cwd)
 ```
 
-### src/retrieval/ranker.js
-```
-module.exports = { rank, buildSigIndex, scoreFile, formatRankTable, formatRankJSON, DEFAULT_WEIGHTS, detectIntent }
-function scoreFile(filePath, sigs, queryTokens, weights) → number
-function rank(query, sigIndex, opts) → { file: string, score: nu
-function _parseContextFile(contextPath) → Map<string, string[]>
-function buildSigIndex(cwd, opts) → Map<string, string[]>
-function formatRankTable(results, query) → string
-function formatRankJSON(results, query) → object
-function detectIntent(query)
-```
-
 ### src/tracking/logger.js
 ```
 module.exports = { logRun, readLog, summarize }
@@ -652,15 +588,6 @@ function extractClassMembers(block)
 function normalizeParams(params)
 ```
 
-### src/config/loader.js
-```
-module.exports = { loadConfig, loadBaseConfig }
-function loadBaseConfig(extendsVal, cwd)
-function detectAutoSrcDirs(cwd, excludeList) → string[]
-function loadConfig(cwd) → object
-function deepClone(obj)
-```
-
 ### src/learning/weights.js
 ```
 module.exports = { BASELINE, DECAY, MAX_MULT, MIN_MULT, weightsPath, clampMultiplier, normalizeFile, loadWeights, saveWeights, updateWeights, boostFiles, penalizeFiles, resetWeights, exportWeights, importWeights }
@@ -678,6 +605,84 @@ function exportWeights(cwd, outputPath)
 function importWeights(cwd, importPath, replace)
 ```
 
+### src/config/defaults.js
+```
+module.exports = { DEFAULTS }
+```
+
+### src/config/loader.js
+```
+module.exports = { loadConfig, loadBaseConfig }
+function loadBaseConfig(extendsVal, cwd)
+function detectAutoSrcDirs(cwd, excludeList) → string[]
+function _legacyDetectAutoSrcDirs(cwd, excludeList) → string[]
+function loadConfig(cwd) → object
+function deepClone(obj)
+```
+
+### src/discovery/source-root-registry.js
+```
+module.exports = { REGISTRY }
+```
+
+### src/discovery/source-root-resolver.js
+```
+module.exports = { resolveSourceRoots }
+function resolveSourceRoots(cwd, opts = {})
+function _detectMonorepo(cwd)
+function _enumerateCandidates(cwd, isMonorepo, ignorePatterns, excludeList)
+function _applySpecialRules(scored, cwd, primaryFw, fwEntry, frameworks)
+function _dedupeNested(scored)
+function _computeConfidence(frameworks, languages, scoredCount)
+```
+
+### src/discovery/language-detector.js
+```
+module.exports = { detectLanguages }
+function detectLanguages(cwd)
+function _walkDepth(dir, depth, extCount)
+```
+
+### src/discovery/source-root-scorer.js
+```
+module.exports = { scoreCandidate, getRecentlyChangedDirs, ROOT_ENTRYPOINTS }
+function getRecentlyChangedDirs(cwd)
+function scoreCandidate(dirName, fullPath, context)
+function _countSourceFiles(dir, depth)
+```
+
+### src/discovery/framework-detector.js
+```
+module.exports = { detectFrameworks }
+function detectFrameworks(cwd)
+function _readDeps(cwd)
+function _readFile(p)
+function _existsAnywhere(cwd, filename, maxDepth)
+function _walkFind(dir, name, depth)
+```
+
+### src/discovery/sigmapignore.js
+```
+module.exports = { loadIgnorePatterns, matchesIgnorePattern }
+function loadIgnorePatterns(cwd)
+function matchesIgnorePattern(dirName, patterns)
+```
+
+### src/retrieval/ranker.js
+```
+module.exports = { rank, buildSigIndex, scoreFile, formatRankTable, formatRankJSON, DEFAULT_WEIGHTS, GRAPH_BOOST_AMOUNTS, detectIntent }
+function _computePenalty(filePath)
+function _computeHubs(graph)
+function _isHub(filePath)
+function scoreFile(filePath, sigs, queryTokens, weights) → { score: number, signals:
+function rank(query, sigIndex, opts) → { file: string, score: nu
+function _parseContextFile(contextPath) → Map<string, string[]>
+function buildSigIndex(cwd, opts) → Map<string, string[]>
+function formatRankTable(results, query) → string
+function formatRankJSON(results, query) → object
+function detectIntent(query)
+```
+
 ### src/mcp/server.js
 ```
 module.exports = { start }

package/CHANGELOG.md

@@ -10,6 +10,32 @@ Format: [Semantic Versioning](https://semver.org/)
 
 ---
 
+## [6.5.2] — 2026-04-27
+
+### Added
+
+- **2-hop graph boost with decay** — `rank()` now traverses 2 hops in the dependency graph instead of 1. Direct imports (+0.40) and second-order imports (+0.15 with decay) receive score boosts for better context relevance in multi-layer dependency scenarios.
+- **Hub suppression** — shared utility files (detected by >20% fanout threshold or static patterns like `util/`, `helper/`, `common/`) are now excluded from graph boosts to prevent over-boosting generic utilities.
+- **Incremental signature cache (`sigCache`)** — new opt-in `sigCache: true` config key enables mtime-based caching of extracted signatures. Cache is automatically busted on version changes, and unchanged files skip re-extraction for faster subsequent runs.
+- **Cache health statistics** — `--health` output now includes cache stats: entry count and disk size in KB. `--health --json` includes `cacheStats` field with `entries` and `sizeKb` when cache exists.
+
+---
+
+## [6.5.1] — 2026-04-25
+
+### Added
+
+- **Retrieval explain** — `rank()` and `scoreFile()` now return detailed signal breakdown (exactToken, symbolMatch, prefixMatch, pathMatch, penalty) for transparency in ranking decisions
+- **7-intent ranking** — expanded intent detection from 4 to 7 patterns (debug, explain, refactor, review, test, integrate, navigate). Each intent applies tuned weights to prioritize relevant signals.
+- **Negative-signal penalty layer** — formalized penalties for test files (0.4x), generated code (0.3x), documentation (0.2x), and node_modules (0.0x) to deprioritize non-source content
+
+### Changed
+
+- `formatRankTable` now shows penalty column and signals breakdown for top 3 results
+- `formatRankJSON` now includes `intent` and `signals` fields in output for API consumers
+
+---
+
 ## [6.5.0] — 2026-04-25
 
 ### Added

package/README.md

@@ -38,7 +38,7 @@ Works with Copilot, Claude, Cursor, Windsurf, and any LLM.
 
 ## Why SigMap?
 
-- **78.9% hit@5** — right file found in top 5 results (vs 13.6% baseline)
+- **81.1% hit@5** — right file found in top 5 results (vs 13.6% baseline)
 - **40–98% token reduction** — 2K–4K tokens instead of 80K+
 - **52.2% task success rate** — up from 10% without context
 - **1.69 prompts per task** — down from 2.84
@@ -51,7 +51,7 @@ Works with Copilot, Claude, Cursor, Windsurf, and any LLM.
 
 | Without SigMap | With SigMap |
 |---|---|
-| ❌ Guessing which files are relevant | ✅ Right file in context — 79% of the time |
+| ❌ Guessing which files are relevant | ✅ Right file in context — 81% of the time |
 | ❌ Sending the full repo to your AI | ✅ Minimal context — only what matters |
 | ❌ Embeddings / vector DB required | ✅ Grounded answers, no infra needed |
 
@@ -75,11 +75,11 @@ Ask → Rank → Context → Validate → Judge → Learn
 ## Benchmark
 
 ```
-Benchmark : sigmap-v6.4-main
-Date      : 2026-04-23
+Benchmark : sigmap-v6.5-main
+Date      : 2026-04-25
 
-Hit@5          : 78.9%   (baseline 13.6%  — 5.8× lift)
-Prompt reduction : 40.6%
+Hit@5          : 81.1%   (baseline 13.6%  — 6.0× lift)
+Prompt reduction : 41.4%
 Task success   : 52.2%   (baseline 10%)
 Prompts / task : 1.69    (baseline 2.84)
 Token reduction: 40–98%  (avg 96.9% across 18 real repos)