@parad0x_labs/openclaw-context-capsule 1.6.0 → 1.7.0

package/CHANGELOG.md

@@ -0,0 +1,21 @@
+# Changelog
+
+## 1.7.0
+
+- **Value-atom preservation.** A lone distinctive value — a bare port (`5433`), an
+  issue ref (`#4821`), a version (`v2.13.0`), an ISO date, or a hyphenated code
+  (`NEEDLE-ZX-7742`) — is now emitted as its own dense atom, so it survives
+  compression even when its surrounding sentence loses the budget race. Previously
+  such a value only survived when its whole line was selected. New regression:
+  `test/value-survival.test.mjs` (wired into `npm test`).
+- **Docs accuracy.** Corrected the documented config defaults (`maxCapsuleTokens`
+  `700` → `1400`, `capsuleTokenRatio` `0.08` → `0.14`) to match the shipped values.
+  Reframed the fidelity figures to state plainly that they are measured on the
+  maintainer's own private sessions via `test/fidelity-bench.mjs` (reproduce on your
+  own `~/.openclaw` sessions), not a repo fixture. The supersession clean rate (83%
+  floor) and value-survival remain CI-gated by `npm test`.
+
+## 1.6.0
+
+- Self-contained compression core bundled in `src/compression.ts` (no external
+  runtime dependency; no network / file-system / on-chain calls).

package/README.md

@@ -46,8 +46,8 @@ LLM is given the extractive capsule, not opaque compressed bytes.
 | --- | ---: | --- |
 | `minMessages` | `20` | Sessions shorter than this pass through unchanged. |
 | `keepRecentMessages` | `10` | Recent messages kept verbatim after compression. |
-| `maxCapsuleTokens` | `700` | Hard cap for the injected extractive capsule. |
-| `capsuleTokenRatio` | `0.08` | If OpenClaw provides a model token budget, cap the capsule to this fraction of the budget. |
+| `maxCapsuleTokens` | `1400` | Hard cap for the injected extractive capsule. |
+| `capsuleTokenRatio` | `0.14` | If OpenClaw provides a model token budget, cap the capsule to this fraction of the budget. |
 | `minCompressTokens` | `900` | Estimated transcript-token floor before compression activates. |
 
 ```jsonc
@@ -57,8 +57,8 @@ LLM is given the extractive capsule, not opaque compressed bytes.
       "context-capsule": {
         "minMessages": 20,
         "keepRecentMessages": 10,
-        "maxCapsuleTokens": 700,
-        "capsuleTokenRatio": 0.08,
+        "maxCapsuleTokens": 1400,
+        "capsuleTokenRatio": 0.14,
         "minCompressTokens": 900
       }
     }

package/SKILL.md

@@ -1,6 +1,9 @@
 ---
 name: context-capsule
 description: Compresses older OpenClaw agent session history into a bounded, lane-change-aware context capsule — keeps recent messages verbatim, flags abandoned directions, quarantines injected instructions, and redacts secrets. Local, deterministic, any model.
+license: MIT
+metadata:
+  author: Parad0x-Labs
 ---
 
 # Context Capsule
@@ -19,11 +22,17 @@ Claude, GPT, Ollama, Mistral, LM Studio.
   held-out pivot set it cleans 83% of abandoned subjects with **zero** wrongly
   flagged live choices.
 - **High fidelity per token.** Distinctive signals — file paths, IDs, ports,
-  URLs, commands, errors, decisions — are emitted as dense atoms, so the capsule
-  keeps **79% of key signals at ~5× reduction and 93% at ~3.4×** on real
-  sessions (measured, see `test/fidelity-bench.mjs`).
-
-> **Self-contained (v1.6.0):** The compression core is bundled directly in this
+  URLs, commands, errors, decisions — are emitted as dense atoms. As of **v1.7.0**,
+  a lone value (a bare port `5433`, an issue ref `#4821`, a version `v2.13.0`, an
+  ISO date, a hyphenated code like `NEEDLE-ZX-7742`) is emitted as its **own** atom,
+  so it survives compression even when its surrounding sentence loses the budget
+  race — CI-tested in `test/value-survival.test.mjs`. On the maintainer's own real
+  sessions the capsule kept ~79% of key signals at ~5× reduction and ~93% at ~3.4×;
+  `test/fidelity-bench.mjs` measures this against **your own** `~/.openclaw` sessions
+  (run it on your data to reproduce — the figures are from private sessions, not a
+  repo fixture).
+
+> **Self-contained (v1.7.0):** The compression core is bundled directly in this
 > skill (`src/compression.ts`). There is **no external runtime dependency**, and
 > the plugin makes **no network, file-system, or on-chain calls**. Everything
 > runs locally using only Node's built-in `zlib` and `crypto`, and is fully

package/dist/compression.js

@@ -906,6 +906,12 @@ const ATOM_URL_RE = /https?:\/\/[^\s)\]}>"']+/giu;
 const ATOM_PATH_RE = /(?:[.~]?\/)?[\w.-]+\/[\w./@+-]*\.[A-Za-z0-9]{1,8}/giu;
 const ATOM_CMD_RE = /\b(?:pnpm|npm|bun|node|git|gh|openclaw|launchctl|lsof|tail|cat|rg|jq|curl|ssh|docker|kubectl)\s+[\w./@:=-][^\n,;]*/giu;
 const ATOM_HASH_RE = /\b[A-Za-z0-9]{20,}\b/gu;
+// Bare distinctive VALUES the URL/path/command/hash passes miss: a standalone port or
+// code (3+ digits), an issue ref (#42), a version (v2.13), an ISO date, an sk- key, or a
+// hyphenated code carrying digits (NEEDLE-ZX-7742). These are the answer of a fact and are
+// short, so the length gate below would drop them — harvest them separately so a lone
+// `5433` / `#42` / `2026-07-15` survives as its own atom, not only when its line is picked.
+const ATOM_VALUE_RE = /\bsk-[A-Za-z0-9_-]{6,}\b|\b\d{4}-\d{2}-\d{2}\b|#\d{2,}\b|\bv\d+\.\d+(?:\.\d+)?\b|\b\w*-\w*\d{2,}[\w-]*\b|\b\d{3,}\b/giu;
 function extractAtoms(content) {
     const atoms = new Set();
     const harvest = (re) => {
@@ -925,6 +931,14 @@ function extractAtoms(content) {
         if (![...atoms].some((a) => a.includes(m)))
             atoms.add(m.slice(0, 60));
     }
+    // Bare values last, min length 2 (so `#42` survives), skipping any already contained in a
+    // URL/path/hash atom (a port inside a URL is not double-emitted).
+    ATOM_VALUE_RE.lastIndex = 0;
+    for (const m of content.match(ATOM_VALUE_RE) ?? []) {
+        const v = m.trim();
+        if (v.length >= 2 && ![...atoms].some((a) => a.includes(v)))
+            atoms.add(v.slice(0, 60));
+    }
     return [...atoms];
 }
 function extractFacts(messages, maxFacts) {

package/dist/index.js

@@ -190,7 +190,7 @@ class ContextCapsuleEngine {
     info = {
         id: "context-capsule",
         name: "Context Capsule",
-        version: "1.6.0",
+        version: "1.7.0",
         ownsCompaction: false,
         turnMaintenanceMode: "background",
     };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@parad0x_labs/openclaw-context-capsule",
-  "version": "1.6.0",
+  "version": "1.7.0",
   "description": "Self-contained OpenClaw context engine that compresses older agent session history into a bounded extractive capsule, preserving decisions, tasks, errors, files, links, and durable facts while keeping recent messages verbatim.",
   "type": "module",
   "main": "./dist/index.js",
@@ -8,7 +8,8 @@
     "dist/",
     "openclaw.plugin.json",
     "README.md",
-    "SKILL.md"
+    "SKILL.md",
+    "CHANGELOG.md"
   ],
   "keywords": [
     "openclaw",
@@ -38,7 +39,7 @@
     "typecheck": "tsc --noEmit -p tsconfig.json",
     "prepack": "npm run build",
     "prepublishOnly": "npm run build && npm test",
-    "test": "npm run build && node test/compression-smoke.mjs && node test/quality.test.mjs && node test/supersession-bench.mjs && node test/hardening-bench.mjs && node test/secret-bench.mjs && node test/platform-parity.test.mjs"
+    "test": "npm run build && node test/compression-smoke.mjs && node test/quality.test.mjs && node test/value-survival.test.mjs && node test/supersession-bench.mjs && node test/hardening-bench.mjs && node test/secret-bench.mjs && node test/platform-parity.test.mjs"
   },
   "engines": {
     "node": ">=22"