pi-beyin 0.1.6 → 0.1.7

package/README.md

@@ -77,6 +77,21 @@ The extensions call `beyin.py` (read-only unless the agent runs a write command)
 A vault whose `beyin.py` does not accept `--harness pi` needs `tools/pi-harness-patch.py ensure
 --vault <vault>` (the extension does this on session start).
 
+## Relationship to beyin v3
+
+`pi-beyin` is the **pi-side** adapter; the vault, `beyin.py` and the canonical skill templates belong to
+beyin v3 and keep their own release cycle.
+
+- `tools/pi-harness-patch.py` is the only thing that touches beyin's files: it adds `pi` to the harness
+  list in two managed modules (plus three pi-specific fixes) and re-syncs the install manifest so
+  `beyin.py update` plans cleanly. **A beyin update reverts these edits**; the extension calls `ensure`
+  on session start, so they come back by themselves. `status` is read-only.
+- The skills here are **copies for pi**, not links into the vault. beyin still installs its own copies
+  under `.claude/skills/` for the other harnesses; pi does not read that directory.
+- Notifications stay in the adapter: the injected context, the receipt reminder, the distillation
+  suggestion and the `beyin_jev_*` tool descriptions ship with this package. They are written in Turkish
+  today (the author's language); anything a *skill* writes follows the user's language.
+
 ## Project memory skills
 
 Skill descriptions carry bilingual trigger phrases (English + Turkish) on purpose: a description is a

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-beyin",
-  "version": "0.1.6",
+  "version": "0.1.7",
   "description": "pi-native bridge for the beyin second brain: context injection, distillation reminders, and Jev advisor tools.",
   "keywords": [
     "pi-package",

package/tools/pi-harness-patch.py

@@ -57,29 +57,30 @@ EDITS = {
             "('codex', 'claude', 'antigravity', 'hermes', 'opencode', 'omp', 'pi')",
         ),
     ],
-    # Python Jev istemcisi urllib varsayılan UA'sı ('Python-urllib/x') ile CommandCode'a gidince
-    # Cloudflare 1010 -> HTTP 403 alıyor. openai-node UA'sı (pi-native yolun kullandığı) geçiyor.
-    # Update sonrası 'ensure' yeniden uygular. Ayrıntı: work staj/logs/0111.
+    # The Python Jev client sends urllib's default UA ('Python-urllib/x') to CommandCode and gets
+    # Cloudflare 1010 -> HTTP 403. The openai-node UA (what the pi-native path uses) passes.
+    # 'ensure' re-applies this after every update. Measured 2026-10-15.
     ".claude/scripts/beyin_v3_jev_client.py": [
         (
             "headers={'Authorization':'Bearer '+key,'Content-Type':'application/json'}",
             "headers={'Authorization':'Bearer '+key,'Content-Type':'application/json','User-Agent':'openai-node/4.0.0'}",
         ),
     ],
-    # CONTEXT_LIMIT (4000) Jev'in gerçek bağlam kapasitesinin çok altındaydı: 19943 kr bağlamla dahi
-    # kanıt Sonda olduğu hâlde doğru verdict + yüksek confidence verdi (kırpma yok, 2026-10-15 ölçümü);
-    # ~24424 kr'da transport sınırı devreye giriyor (`max_input_chars=24000`). 4000 iken >4000 kr notlar
-    # `source_context_incomplete` ile doğrulanamıyordu (log 0114). 12000: her mevcut notu kapsar, güvenli bölgede.
+    # CONTEXT_LIMIT (4000) sat far below Jev's real context capacity: even with 19943 chars of context
+    # and the evidence at the very end, it returned the right verdict with high confidence (no
+    # truncation, measured 2026-10-15). The transport limit starts around 24424 chars
+    # (`max_input_chars=24000`). At 4000, notes longer than that failed as
+    # `source_context_incomplete`. 12000 covers every current note and stays in the safe zone.
     ".claude/scripts/beyin_v3_jev.py": [
         (
             "CONTEXT_LIMIT = 4000  # Jev; a provider with a smaller window reports context_limit",
             "CONTEXT_LIMIT = 12000  # Jev; measured >=19943 kr safe (2026-10-15); transport cap max_input_chars=24000",
         ),
     ],
-    # memory_assessment tek istekte evidence + en fazla 4 prior gönderir; toplam gövde
-    # max_input_chars'ı (24000) aşınca `budget_exceeded` ile tüm değerlendirme düşüyordu (log 0114).
-    # Prior'lar opsiyonel, evidence zorunlu → bütçe yetmezse prior'ları sondan düşür. Bu,
-    # "complete context" invariant'ını bozmaz (hiçbir kaynak özetlenmez/kırpılmaz).
+    # memory_assessment sends evidence plus up to 4 priors in one request; when the body exceeded
+    # max_input_chars (24000) the whole assessment failed with `budget_exceeded`. Priors are optional,
+    # evidence is required -> when the budget is tight, drop priors from the end. This does not break
+    # the "complete context" invariant: no source is summarized or truncated.
     ".claude/scripts/beyin_v3_memory_assessment.py": [
         (
             "    try:\n"