composto-ai 0.7.1 → 0.8.2

package/README.md

@@ -1,23 +1,25 @@
 # Composto
 
-**Causal memory layer for coding agents. Catches the bug your agent is about to reintroduce.**
+**Token-efficient code context for AI agents. Your file's full structure in a fraction of the tokens, with its causal history baked in.**
 
-Composto is a repo-local graph of your git history that your AI coding agent consults before every edit. When a file was reverted recently, has a fix cluster in its history, or was last touched by someone who left the team, Composto surfaces that signal as in-context guidance before the agent writes the code. Hook-enforced on Claude Code, Cursor, and Gemini CLI. Local-first, MIT.
+> Send your agent the structure, not the noise.
+
+Composto compresses any source file into a Health-Aware IR that keeps exactly what your agent needs — signatures, types, control flow, dependencies — at 60-95% fewer tokens than raw code. On top of that, it surfaces the file's causal history (what historically changed and broke alongside the code you're touching) as advisory context. Local-first, MIT. Works with Claude Code, Cursor, and Gemini CLI.
 
 ```
-$ composto impact src/memory/signals/hotspot.ts
-
-verdict:    medium
-score:      0.52
-confidence: 0.50
-signals:
-  revert_match       ■■■■■■■■■■ strength=1.00 precision=1.00
-  hotspot            ■          strength=0.10 precision=0.54
-  fix_ratio          ■          strength=0.07 precision=0.54
-  author_churn       ·          strength=0.00 precision=0.16
-
-# This file was touched by a Revert commit in history.
-# blastradius remembers. Your LLM couldn't.
+$ composto ir src/memory/confidence.ts L1
+
+USE:./types.js
+OUT INTERFACE:ConfidenceContext
+OUT INTERFACE:ScoreAndConfidence
+FN:calibrationFactor(signals: Signal[])
+      GUARD:[firing.length === 0 → 1.0, avg < 20 → 0.3, avg < 100 → 0.6]
+FN:historyFactor(totalCommits: number)
+      GUARD:[totalCommits < 50 → 0.2, totalCommits < 200 → 0.5, totalCommits < 1000 → 0.8]
+OUT FN:computeScoreAndConfidence(signals: Signal[], ctx: ConfidenceContext)
+
+# 541 tokens of raw code → 230 tokens of IR (57% fewer). Structure intact:
+# every signature, dependency, and decision threshold survives.
 ```
 
 ---
@@ -45,9 +47,9 @@ composto impact src/auth/login.ts
 composto index --status     # diagnostics: schema, freshness, calibration
 ```
 
-### Also in the box: AST compression tools
+### The core: token-efficient structural context
 
-Composto also ships a tree-sitter based AST compressor (about 89% token savings) and a smart context packer for bug-fix tasks. These are separate from the causal layer but live in the same binary.
+Composto's spine is a tree-sitter based AST compressor and a smart context packer. Compress any file to IR, or pack a whole directory into a token budget:
 
 ```bash
 composto ir src/app.ts                 # compress a file to IR (L0/L1/L2/L3)
@@ -57,6 +59,10 @@ composto benchmark .                   # see compression stats
 
 See the [IR Layers](#ir-layers), [Health-Aware IR](#health-aware-ir), and [Context Budget](#context-budget) sections below for details.
 
+### On top: causal history as advisory context
+
+Composto also indexes your git history and surfaces what historically changed and broke alongside the file you're editing — advisory context the agent weighs, not a gate. See [Causal context](#causal-context) below.
+
 ### MCP plugin (Claude Code, Cursor, Claude Desktop)
 
 The MCP server is bundled inside `composto-ai`. Install the package globally first, then register the server with your client:
@@ -188,27 +194,25 @@ composto index --status        # diagnostics: schema, freshness, calibration
 
 ---
 
-## BlastRadius
+## Causal context
 
-Beyond compression, Composto indexes your repo's git history into a local SQLite graph and exposes it as a queryable risk surface. Before your agent edits a file, it can ask: *"has this region been reverted? who fixed the last similar bug? is the last author still around?"* — signals no LLM can infer from current code alone.
+On top of compression, Composto indexes your repo's git history into a local SQLite graph and surfaces what the current code can't tell you: *"has this region been reverted? does it have a fix cluster? what files historically changed and broke alongside it?"* — context no LLM can infer from the file alone. It's delivered as **advisory context the agent weighs**, not a hard gate.
 
-Four signals per query: `revert_match`, `hotspot`, `fix_ratio`, `author_churn`. Verdict is `low` / `medium` / `high` / `unknown`; when confidence is low the tool returns `unknown` rather than guessing. Precision is repo-calibrated (self-validation over the repo's own fix history).
+Signals per query: `revert_match`, `hotspot`, `fix_ratio`, `author_churn`, `cochange`. The tool returns `unknown` when confidence is low rather than guessing.
 
 ```
-verdict:    high
-score:      1.00
-confidence: 0.30
-signals:
-  revert_match       ■■■■■■■■■■ strength=1.00 precision=0.50
-  hotspot            ·          strength=0.00 precision=0.30
-  ...
+$ composto impact src/auth/login.ts
+
+revert_match   ■■■■■■■■■■ this file was touched by a Revert commit
+cochange       ■■■■■      historically co-changed with session.ts, token.ts in fixes
+hotspot        ■          14 changes in the last 90 days
 ```
 
-**Where we are, honestly.** The v2.1 time-travel backtest (rewinds the DB to each pre-fix snapshot) shows `revert_match` carrying most of the product's value — it clears the ship gate on picomatch (precision 0.65, recall 0.78 on the `medium|high` band). Signal-attributed precision (excluding `revert_match`) is weaker: the three non-revert signals are alive but need calibration work. See [docs/blastradius-proof-v2.md](docs/blastradius-proof-v2.md) for numbers on all four band combinations across two public repos + the per-signal diagnostic behind them.
+**Where we are, honestly.** A 4-repo time-travel backtest (fastify, express, got, flask — each rewound to pre-fix snapshots) shows the causal layer is a **high-recall, advisory-grade** signal: on mature repos it recovers 67-80% of the files a fix actually touches. Precision is modest (~0.55) — these signals point you at *candidates*, they don't certify them, which is exactly why Composto surfaces them as context for the agent to judge rather than as a blocking verdict. Recall scales with git history, so the value grows as your repo matures (a young repo gets little until it accumulates fix history).
 
-The honest framing: **BlastRadius is a bug-history memory layer that agents query before editing.** "This file was reverted three weeks ago" is the primary promise v1 delivers. The other signals expand the query surface — calibration work on `hotspot` and `fix_ratio` is the open follow-on.
+The honest framing: **causal context is a high-recall memory layer agents consult before editing** — "these files have a history of breaking together" — not a precision gate. The compression core works unconditionally; the causal layer adds repo-specific memory on top.
 
-Feature-flagged via `COMPOSTO_BLASTRADIUS=1` during the beta. Available as both CLI (`composto impact`, `composto index`) and MCP tool (`composto_blastradius`).
+Available as CLI (`composto impact`, `composto index`) and MCP tool (`composto_blastradius`, gated by `COMPOSTO_BLASTRADIUS=1` during beta).
 
 ---
 
@@ -307,14 +311,14 @@ Hotspot files get full detail. Everything else gets structure. Budget is never e
 ## Stats
 
 ```
-Overall compression: 89.2%
-L0 compression:      97.5%
-AST engine:          51/51 files (0 regex fallback)
+L1 compression:      ~81% fewer tokens (full IR, structure preserved)
+L0 compression:      ~97% fewer tokens (structure map)
+Token counts:        verified against a real BPE tokenizer, not estimates
+AST engine:          AST-parsed, 0 regex fallback
 Languages:           TypeScript, JavaScript, Python, Go, Rust
-Tests:               251 passing
-BlastRadius v2.1:    precision 0.65, recall 0.78 (picomatch, time-travel,
-                     medium|high band). Honest signal-attributed numbers
-                     in docs/blastradius-proof-v2.md.
+Causal layer:        high-recall advisory (0.67-0.80 recall on mature repos,
+                     time-travel backtest across 4 public repos); precision
+                     ~0.55, surfaced as context not a gate.
 ```
 
 ---

package/dist/index.js

@@ -1010,6 +1010,17 @@ function emitTier3(node) {
       if (expr.type === "await_expression") {
         return null;
       }
+      if (expr.type === "assignment_expression") {
+        if (node.parent?.type === "statement_block") return null;
+        const left = expr.childForFieldName("left");
+        const right = expr.childForFieldName("right");
+        if (left?.type === "member_expression" && right && (right.type === "function_expression" || right.type === "arrow_function")) {
+          const asyncPrefix = isAsync(right) ? "ASYNC " : "";
+          const params = right.childForFieldName("parameters")?.text ?? "()";
+          return `${asyncPrefix}FN:${left.text}${collapseText(params, 60)}`;
+        }
+        return null;
+      }
       if (expr.type === "call_expression") {
         const callee = expr.child(0)?.text ?? "";
         if (callee === "ObjectSetPrototypeOf" || callee === "Object.setPrototypeOf") {
@@ -2125,13 +2136,38 @@ function computeAuthorChurn(db, filePath) {
   return { ...base, strength };
 }
 
+// src/memory/signals/cochange.ts
+var SATURATION_DEGREE = 10;
+var FALLBACK_PRECISION5 = 0.3;
+function computeCochange(db, filePath) {
+  const row = db.prepare(`
+      SELECT COUNT(DISTINCT ft2.file_path) AS degree
+      FROM file_touches ft1
+      JOIN commits c ON c.sha = ft1.commit_sha AND c.is_fix = 1
+      JOIN file_touches ft2 ON ft2.commit_sha = ft1.commit_sha AND ft2.file_path != ft1.file_path
+      WHERE ft1.file_path = ?
+    `).get(filePath);
+  const degree = row?.degree ?? 0;
+  const strength = Math.min(1, degree / SATURATION_DEGREE);
+  const cal = getCalibration(db, "cochange", FALLBACK_PRECISION5);
+  return {
+    type: "cochange",
+    strength,
+    precision: cal.precision,
+    sample_size: cal.sampleSize,
+    evidence: [],
+    cochange_degree: degree
+  };
+}
+
 // src/memory/signals/index.ts
 function collectSignals(db, _repoPath, filePath) {
   return [
     computeRevertMatch(db, filePath),
     computeHotspot(db, filePath),
     computeFixRatio(db, filePath),
-    computeAuthorChurn(db, filePath)
+    computeAuthorChurn(db, filePath),
+    computeCochange(db, filePath)
   ];
 }
 
@@ -2170,18 +2206,29 @@ function historyFactor(totalCommits) {
   if (totalCommits < 1e3) return 0.8;
   return 1;
 }
+var DEFAULT_COCHANGE_FLOOR = 1;
+function cochangeFloor() {
+  const v = Number(process.env.COMPOSTO_COCHANGE_FLOOR);
+  return Number.isFinite(v) && v >= 0 && v <= 1 ? v : DEFAULT_COCHANGE_FLOOR;
+}
 function computeScoreAndConfidence(signals, ctx) {
+  const cochange = signals.find((s) => s.type === "cochange");
+  const averaged = signals.filter((s) => s.type !== "cochange");
   let num = 0;
   let den = 0;
-  for (const s of signals) {
+  for (const s of averaged) {
     if (s.strength <= 0 || s.precision <= 0) continue;
     num += s.strength * s.precision;
     den += s.precision;
   }
-  const score = den === 0 ? 0 : num / den;
+  let score = den === 0 ? 0 : num / den;
+  if (cochange) {
+    const floor = cochangeFloor();
+    score *= floor + (1 - floor) * cochange.strength;
+  }
   const confidence = Math.min(
-    coverageFactor(signals),
-    calibrationFactor(signals),
+    coverageFactor(averaged),
+    calibrationFactor(averaged),
     freshnessFactor(ctx),
     historyFactor(ctx.totalCommits)
   );
@@ -2752,6 +2799,25 @@ async function runBenchmark(projectPath) {
   console.log(`  L1 (full IR):        ${summary.totalRaw} \u2192 ${summary.totalIRL1} tokens (${summary.totalSavedPercent.toFixed(1)}% reduction)`);
   console.log(`  Files analyzed: ${summary.fileCount}`);
   console.log(`  Engine: ${summary.astCount} AST, ${summary.fpCount} FP`);
+  printSavingsCard(summary);
+}
+function printSavingsCard(summary) {
+  if (summary.totalRaw <= 0) return;
+  const saved = summary.totalRaw - summary.totalIRL1;
+  const pct2 = summary.totalSavedPercent.toFixed(1);
+  const SONNET_PER_MTOK = 3;
+  const LOADS_PER_DAY = 50;
+  const DAYS = 30;
+  const rawCost = summary.totalRaw / 1e6 * SONNET_PER_MTOK;
+  const irCost = summary.totalIRL1 / 1e6 * SONNET_PER_MTOK;
+  const monthly = (rawCost - irCost) * LOADS_PER_DAY * DAYS;
+  console.log("\n  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500");
+  console.log(`  \u{1F4B8} Every full-context load of this project: $${rawCost.toFixed(2)} raw \u2192 $${irCost.toFixed(2)} with Composto`);
+  console.log(`     (Claude Sonnet input, $3/Mtok). At ${LOADS_PER_DAY} loads/day that is ~$${monthly.toFixed(0)}/month saved.`);
+  console.log("\n  \u{1F4CB} Share your result:");
+  console.log(`     Composto compressed my ${summary.fileCount}-file project ${pct2}% (${summary.totalRaw.toLocaleString()} \u2192 ${summary.totalIRL1.toLocaleString()} tokens).`);
+  console.log("     Try yours: npm i -g composto-ai && composto benchmark .");
+  console.log("  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500");
 }
 async function runBenchmarkQuality(projectPath, filePath) {
   const apiKey = process.env.ANTHROPIC_API_KEY;

package/dist/mcp/server.js

@@ -818,6 +818,17 @@ function emitTier3(node) {
       if (expr.type === "await_expression") {
         return null;
       }
+      if (expr.type === "assignment_expression") {
+        if (node.parent?.type === "statement_block") return null;
+        const left = expr.childForFieldName("left");
+        const right = expr.childForFieldName("right");
+        if (left?.type === "member_expression" && right && (right.type === "function_expression" || right.type === "arrow_function")) {
+          const asyncPrefix = isAsync(right) ? "ASYNC " : "";
+          const params = right.childForFieldName("parameters")?.text ?? "()";
+          return `${asyncPrefix}FN:${left.text}${collapseText(params, 60)}`;
+        }
+        return null;
+      }
       if (expr.type === "call_expression") {
         const callee = expr.child(0)?.text ?? "";
         if (callee === "ObjectSetPrototypeOf" || callee === "Object.setPrototypeOf") {
@@ -1929,13 +1940,38 @@ function computeAuthorChurn(db, filePath) {
   return { ...base, strength };
 }
 
+// src/memory/signals/cochange.ts
+var SATURATION_DEGREE = 10;
+var FALLBACK_PRECISION5 = 0.3;
+function computeCochange(db, filePath) {
+  const row = db.prepare(`
+      SELECT COUNT(DISTINCT ft2.file_path) AS degree
+      FROM file_touches ft1
+      JOIN commits c ON c.sha = ft1.commit_sha AND c.is_fix = 1
+      JOIN file_touches ft2 ON ft2.commit_sha = ft1.commit_sha AND ft2.file_path != ft1.file_path
+      WHERE ft1.file_path = ?
+    `).get(filePath);
+  const degree = row?.degree ?? 0;
+  const strength = Math.min(1, degree / SATURATION_DEGREE);
+  const cal = getCalibration(db, "cochange", FALLBACK_PRECISION5);
+  return {
+    type: "cochange",
+    strength,
+    precision: cal.precision,
+    sample_size: cal.sampleSize,
+    evidence: [],
+    cochange_degree: degree
+  };
+}
+
 // src/memory/signals/index.ts
 function collectSignals(db, _repoPath, filePath) {
   return [
     computeRevertMatch(db, filePath),
     computeHotspot(db, filePath),
     computeFixRatio(db, filePath),
-    computeAuthorChurn(db, filePath)
+    computeAuthorChurn(db, filePath),
+    computeCochange(db, filePath)
   ];
 }
 
@@ -1974,18 +2010,29 @@ function historyFactor(totalCommits) {
   if (totalCommits < 1e3) return 0.8;
   return 1;
 }
+var DEFAULT_COCHANGE_FLOOR = 1;
+function cochangeFloor() {
+  const v = Number(process.env.COMPOSTO_COCHANGE_FLOOR);
+  return Number.isFinite(v) && v >= 0 && v <= 1 ? v : DEFAULT_COCHANGE_FLOOR;
+}
 function computeScoreAndConfidence(signals, ctx) {
+  const cochange = signals.find((s) => s.type === "cochange");
+  const averaged = signals.filter((s) => s.type !== "cochange");
   let num = 0;
   let den = 0;
-  for (const s of signals) {
+  for (const s of averaged) {
     if (s.strength <= 0 || s.precision <= 0) continue;
     num += s.strength * s.precision;
     den += s.precision;
   }
-  const score = den === 0 ? 0 : num / den;
+  let score = den === 0 ? 0 : num / den;
+  if (cochange) {
+    const floor = cochangeFloor();
+    score *= floor + (1 - floor) * cochange.strength;
+  }
   const confidence = Math.min(
-    coverageFactor(signals),
-    calibrationFactor(signals),
+    coverageFactor(averaged),
+    calibrationFactor(averaged),
     freshnessFactor(ctx),
     historyFactor(ctx.totalCommits)
   );

package/dist/memory/api.js

@@ -364,13 +364,38 @@ function computeAuthorChurn(db, filePath) {
   return { ...base, strength };
 }
 
+// src/memory/signals/cochange.ts
+var SATURATION_DEGREE = 10;
+var FALLBACK_PRECISION5 = 0.3;
+function computeCochange(db, filePath) {
+  const row = db.prepare(`
+      SELECT COUNT(DISTINCT ft2.file_path) AS degree
+      FROM file_touches ft1
+      JOIN commits c ON c.sha = ft1.commit_sha AND c.is_fix = 1
+      JOIN file_touches ft2 ON ft2.commit_sha = ft1.commit_sha AND ft2.file_path != ft1.file_path
+      WHERE ft1.file_path = ?
+    `).get(filePath);
+  const degree = row?.degree ?? 0;
+  const strength = Math.min(1, degree / SATURATION_DEGREE);
+  const cal = getCalibration(db, "cochange", FALLBACK_PRECISION5);
+  return {
+    type: "cochange",
+    strength,
+    precision: cal.precision,
+    sample_size: cal.sampleSize,
+    evidence: [],
+    cochange_degree: degree
+  };
+}
+
 // src/memory/signals/index.ts
 function collectSignals(db, _repoPath, filePath) {
   return [
     computeRevertMatch(db, filePath),
     computeHotspot(db, filePath),
     computeFixRatio(db, filePath),
-    computeAuthorChurn(db, filePath)
+    computeAuthorChurn(db, filePath),
+    computeCochange(db, filePath)
   ];
 }
 
@@ -409,18 +434,29 @@ function historyFactor(totalCommits) {
   if (totalCommits < 1e3) return 0.8;
   return 1;
 }
+var DEFAULT_COCHANGE_FLOOR = 1;
+function cochangeFloor() {
+  const v = Number(process.env.COMPOSTO_COCHANGE_FLOOR);
+  return Number.isFinite(v) && v >= 0 && v <= 1 ? v : DEFAULT_COCHANGE_FLOOR;
+}
 function computeScoreAndConfidence(signals, ctx) {
+  const cochange = signals.find((s) => s.type === "cochange");
+  const averaged = signals.filter((s) => s.type !== "cochange");
   let num = 0;
   let den = 0;
-  for (const s of signals) {
+  for (const s of averaged) {
     if (s.strength <= 0 || s.precision <= 0) continue;
     num += s.strength * s.precision;
     den += s.precision;
   }
-  const score = den === 0 ? 0 : num / den;
+  let score = den === 0 ? 0 : num / den;
+  if (cochange) {
+    const floor = cochangeFloor();
+    score *= floor + (1 - floor) * cochange.strength;
+  }
   const confidence = Math.min(
-    coverageFactor(signals),
-    calibrationFactor(signals),
+    coverageFactor(averaged),
+    calibrationFactor(averaged),
     freshnessFactor(ctx),
     historyFactor(ctx.totalCommits)
   );

package/dist/memory/worker.js

@@ -283,11 +283,31 @@ function validateAuthorChurn(db) {
   const hits = db.prepare(`SELECT COUNT(*) AS n FROM fix_links`).get().n;
   return { total, hits };
 }
+function validateCochange(db) {
+  const total = db.prepare(`
+    SELECT COUNT(DISTINCT ft1.file_path) AS n
+    FROM file_touches ft1
+    JOIN commits c ON c.sha = ft1.commit_sha AND c.is_fix = 1
+    JOIN file_touches ft2 ON ft2.commit_sha = ft1.commit_sha AND ft2.file_path != ft1.file_path
+  `).get().n;
+  const hits = db.prepare(`
+    SELECT COUNT(DISTINCT ft1.file_path) AS n
+    FROM file_touches ft1
+    JOIN commits c ON c.sha = ft1.commit_sha AND c.is_fix = 1
+    JOIN file_touches ft2 ON ft2.commit_sha = ft1.commit_sha AND ft2.file_path != ft1.file_path
+    WHERE ft1.file_path IN (
+      SELECT ft.file_path FROM file_touches ft
+      JOIN fix_links fl ON fl.suspected_break_sha = ft.commit_sha
+    )
+  `).get().n;
+  return { total, hits };
+}
 var VALIDATORS = {
   revert_match: validateRevertMatch,
   hotspot: validateHotspot,
   fix_ratio: validateFixRatio,
-  author_churn: validateAuthorChurn
+  author_churn: validateAuthorChurn,
+  cochange: validateCochange
 };
 function refreshCalibration(db, headSha) {
   const now = Math.floor(Date.now() / 1e3);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "composto-ai",
-  "version": "0.7.1",
+  "version": "0.8.2",
   "description": "Proactive AI team companion — less tokens, more insight",
   "type": "module",
   "bin": {
@@ -18,6 +18,7 @@
   },
   "scripts": {
     "build": "tsup",
+    "prepublishOnly": "npm run build",
     "test": "vitest run",
     "test:watch": "vitest",
     "dev": "tsx src/index.ts"