pipulate 2.76__tar.gz → 2.77__tar.gz

{pipulate-2.76 → pipulate-2.77}/.agents/skills/journal/references/index.md

@@ -1,8 +1,8 @@
 # The Pipulate journal, indexed
 
-> Auto-generated on 2026-10-02 by `scripts/articles/generate_ai_context.py` and
+> Auto-generated on 2026-10-04 by `scripts/articles/generate_ai_context.py` and
 > rewritten from scratch on every release. If this date looks stale, assume
-> the rest of the repo is newer than this map. 1510 entries indexed.
+> the rest of the repo is newer than this map. 1515 entries indexed.
 
 This repository holds the *machinery*. The *reasoning*, the running journal
 that explains why every piece exists, lives on a separate website and not in
@@ -27,6 +27,11 @@ The first 20 entries include full `index.md` URLs to establish
 the link pattern. All remaining entries are bare slugs. Reconstruct any full
 URL as: `https://mikelev.in/futureproof/{slug}/index.md`
 
+- [2026-10-04] [The Terminal House Style: Quieting the Console Without Silencing the Evidence](https://mikelev.in/futureproof/terminal-house-style-quiet-console-evidence/index.md)
+- [2026-10-03] [Mechanical Advantage and the Active Root: Authoritative Prefixes for Replayable AI Pipelines](https://mikelev.in/futureproof/mechanical-advantage-and-the-active-root/index.md)
+- [2026-10-03] [The Single-Owner Rule and the Verifiable Deed: Quieting Pipeline Noise in the Age of AI](https://mikelev.in/futureproof/single-owner-rule-and-verifiable-deeds/index.md)
+- [2026-10-03] [The Render Canary and the Quiet Compiler: Distinguishing Exposure from Corruption in the Age of AI](https://mikelev.in/futureproof/render-canary-and-the-quiet-compiler/index.md)
+- [2026-10-02] [The Enter Fence and the Bell Cue: Building Replayable Browser Walks in the Age of AI](https://mikelev.in/futureproof/the-enter-fence-and-the-bell-cue/index.md)
 - [2026-10-02] [The Clear Cups Protocol: Auditing MCP Calls with Replayable Receipts](https://mikelev.in/futureproof/clear-cups-protocol-auditing-mcp-calls/index.md)
 - [2026-10-02] [The Door Table and the Moving Twig: Replayable Installs and the Stick Bug Secret](https://mikelev.in/futureproof/the-door-table-and-the-moving-twig/index.md)
 - [2026-10-01] [The Scrollback Rule: Engineering Full-Width Terminal Dividers for Replayable Logs](https://mikelev.in/futureproof/scrollback-rule-clear-x-receipts/index.md)
@@ -42,16 +47,16 @@ URL as: `https://mikelev.in/futureproof/{slug}/index.md`
 - [2026-09-28] [Pinball, Pachinko, and the Art of Replayable AI Workflows](https://mikelev.in/futureproof/pinball-pachinko-replayable-ai-workflows/index.md)
 - [2026-09-28] [The Three-Folder Boundary: Ergonomic Workspaces and the Blank Stare in the Age of AI](https://mikelev.in/futureproof/three-folder-boundary-and-the-blank-stare/index.md)
 - [2026-09-26] [Closing the Loop: Fixing Patch Parser Failures and Actuator Surprises](https://mikelev.in/futureproof/fixing-patch-parser-failures-and-actuator-surprises/index.md)
-- [2026-09-26] [The Three Plain Words: Unifying Context and Building Verifiable Receipts](https://mikelev.in/futureproof/plain-words-verifiable-ai-context/index.md)
-- [2026-09-24] [When the Power Returns: Reconciling Diverged Git Branches and Single-Point Failures](https://mikelev.in/futureproof/when-the-power-returns-verifiable-git-resync/index.md)
-- [2026-09-24] [Local Chrome Profile Inventory and Typed Matrix Resolution](https://mikelev.in/futureproof/local-chrome-profile-inventory-and-typed-matrix-resolution/index.md)
-- [2026-09-21] [Local-First Development and Resilience in the Age of AI](https://mikelev.in/futureproof/local-first-resilience-age-of-ai/index.md)
-- [2026-09-20] [The Letting-Go Moment: Building Vim Training Wheels for the Age of AI](https://mikelev.in/futureproof/vim-training-wheels-and-the-letting-go-moment/index.md)
 
 ## Compact slug index -- pattern: https://mikelev.in/futureproof/{slug}/index.md
 
 Format: `[date] [size] slug` -- fetch any entry as `https://mikelev.in/futureproof/{slug}/index.md`
 
+- [2026-09-26] [460k] plain-words-verifiable-ai-context
+- [2026-09-24] [88k] when-the-power-returns-verifiable-git-resync
+- [2026-09-24] [159k] local-chrome-profile-inventory-and-typed-matrix-resolution
+- [2026-09-21] [184k] local-first-resilience-age-of-ai
+- [2026-09-20] [219k] vim-training-wheels-and-the-letting-go-moment
 - [2026-09-20] [327k] disciplined-forgetting-and-the-memory-hierarchy
 - [2026-09-19] [175k] four-models-one-matchbook-prompt-fanout
 - [2026-09-19] [48k] dual-model-failover-independent-backoff

{pipulate-2.76 → pipulate-2.77}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pipulate
-Version: 2.76
+Version: 2.77
 Summary: The context compiler that helps you prompt well -- local-first, Nix-reproducible, replayable. The successor to AI SEO software.
 Author-email: Mike Levin <pipulate@gmail.com>
 License-Expression: AGPL-3.0-or-later

{pipulate-2.76 → pipulate-2.77}/__init__.py

@@ -12,7 +12,7 @@ Usage:
     pipulate
 """
 
-__version__ = "2.76"
+__version__ = "2.77"
 # APOSTROPHES RESTORED (2026-08-04). They were stripped as a workaround for
 # flake.nix's descMatch regex, whose character class excluded ' from the
 # CAPTURE and truncated the banner to "(So)". That regex was fixed in the same
@@ -23,7 +23,7 @@ __version__ = "2.76"
 # is not a property of the system; the regex is. Blast radius is one banner:
 # nothing but flake.nix reads this name -- version_sync.py syncs __version__
 # and __description__, never this. So'wI' chu' -- "engage the cloaking device."
-__version_description__ = "Release Robustness"
+__version_description__ = "Learning to Walk"
 # SPDX expression, single source of truth, synced into pyproject.toml by
 # scripts/release/version_sync.py. "-or-later" (not bare AGPL-3.0, which is
 # deprecated SPDX) because the header below grants "any later version".

{pipulate-2.76 → pipulate-2.77}/pipulate.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pipulate
-Version: 2.76
+Version: 2.77
 Summary: The context compiler that helps you prompt well -- local-first, Nix-reproducible, replayable. The successor to AI SEO software.
 Author-email: Mike Levin <pipulate@gmail.com>
 License-Expression: AGPL-3.0-or-later

{pipulate-2.76 → pipulate-2.77}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "pipulate"
 # VERSION NOTE: Synced from pipulate/__init__.py.__version__ via python version_sync.py
-version = "2.76"
+version = "2.77"
 authors = [
   { name="Mike Levin", email="pipulate@gmail.com" },
 ]

{pipulate-2.76 → pipulate-2.77}/tests/test_mck_rep2.py

@@ -10,43 +10,70 @@ REPO_ROOT = Path(__file__).resolve().parents[1]
 sys.path.insert(0, str(REPO_ROOT))
 sys.path.insert(0, str(REPO_ROOT / "scripts"))
 
-import prompt_foo
+# prompt_foo is imported inside the two resolver tests (2026-10-04), so the
+# trail tests below load neither the compiler nor the browser stack.
 import walk
 
 
 class MotherCatRep2Tests(unittest.TestCase):
-    def test_pageworkers_trail_satisfies_strict_car_a_schema(self):
-        trail_path = REPO_ROOT / "assets" / "trails" / "botify_pageworkers.yaml"
-        trail = walk.load_trail(trail_path)
+    def test_public_walk_yaml_is_the_default_trail(self):
+        # THREE WALKS (2026-10-04): the PageWorkers trail this test once read
+        # was purged with five others; it pinned a trail's shape, never the
+        # Rep 2 witness. The bundled trail is public_walk, YAML since that day.
+        # Its JSON twin stays out of this test, so removing the twin moves no line.
+        self.assertEqual(walk.DEFAULT_TRAIL.name, "public_walk.yaml")
+        trail = walk.load_trail(walk.DEFAULT_TRAIL)
 
         self.assertEqual(trail["schema_version"], 1)
-        self.assertEqual(trail["name"], "botify_pageworkers")
+        self.assertEqual(trail["name"], "public_walk")
         self.assertEqual(
-            [stop["url_env"] for stop in trail["stops"]],
-            [
-                "PIPULATE_TRAIL_BOTIFY_OPTIMIZATION_URL",
-                "PIPULATE_TRAIL_BOTIFY_MONITORING_URL",
-                "PIPULATE_TRAIL_BOTIFY_REPORTING_URL",
-            ],
+            [stop["name"] for stop in trail["stops"]],
+            ["the_word", "the_receipt", "the_two_pages"],
         )
+        self.assertNotIn("introduction", trail)
         self.assertEqual(
             {
                 "headless": trail["defaults"]["headless"],
                 "persistent": trail["defaults"]["persistent"],
                 "override_cache": trail["defaults"]["override_cache"],
-                "is_notebook_context": trail["defaults"]["is_notebook_context"],
                 "profile_name": trail["defaults"]["profile_name"],
             },
             {
                 "headless": False,
                 "persistent": True,
                 "override_cache": True,
-                "is_notebook_context": False,
-                "profile_name": "botify",
+                "profile_name": "default",
             },
         )
 
+    def test_introduction_is_optional_and_shares_the_walk_namespace(self):
+        # THE INTRODUCTION IS SKIPPED BY DEFAULT (2026-10-04): an optional list
+        # of stops, validated like the walk's own; a name is unique across both
+        # lists, because both land in one capture archive.
+        import yaml
+
+        raw = yaml.safe_load(walk.DEFAULT_TRAIL.read_text(encoding="utf-8"))
+        first = raw["stops"][0]
+        fresh = dict(first, name="first_look", target_slot="first_slot")
+        with tempfile.TemporaryDirectory() as temp_dir:
+            path = Path(temp_dir) / "intro.yaml"
+            path.write_text(
+                yaml.safe_dump(dict(raw, introduction=[fresh])), encoding="utf-8"
+            )
+            trail = walk.load_trail(path)
+            self.assertEqual(
+                [stop["name"] for stop in trail["introduction"]], ["first_look"]
+            )
+            self.assertEqual(len(trail["stops"]), 3)
+            path.write_text(
+                yaml.safe_dump(dict(raw, introduction=[first])), encoding="utf-8"
+            )
+            with self.assertRaisesRegex(walk.TrailError, "must be unique"):
+                walk.load_trail(path)
+
     def test_resolver_matches_requested_and_final_guided_urls(self):
+        import prompt_foo
+
         requested_url = "https://app.example.com/project/page?analysis=1"
         final_url = requested_url + "&context=24h"
 
@@ -111,6 +138,8 @@ class MotherCatRep2Tests(unittest.TestCase):
             )
 
     def test_resolver_falls_back_to_legacy_prompt_foo_cache(self):
+        import prompt_foo
+
         target_url = "https://example.com/a/b?variant=1"
 
         with tempfile.TemporaryDirectory() as temp_dir:

{pipulate-2.76 → pipulate-2.77}/tools/scraper_tools.py

@@ -415,6 +415,7 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
     output_dir = base_dir / domain / url_path_slug
     artifacts = {}
     final_url = None
+    seconds_to_enter = None
 
     def failure(message):
         return {
@@ -663,6 +664,18 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
             driver.execute_cdp_cmd("Network.setCacheDisabled", {"cacheDisabled": True})
         except Exception as exc:
             logger.warning(f"⚠️ Could not turn off the browser cache; this page may come from it: {exc}")
+        # A BLANK START (2026-10-02): the window opened on the browser's
+        # new-tab page, and its chrome:// and chrome-untrusted:// loads sat at
+        # the top of every network log. Open about:blank, then read the
+        # performance log once and drop what it holds. The page under test
+        # has not been requested yet, so this read can take only the start
+        # page's traffic; the one read of the flight is still the drain
+        # further down.
+        try:
+            driver.get("about:blank")
+            driver.get_log("performance")
+        except Exception as exc:
+            logger.warning(f"⚠️ Could not clear the start page from the network log: {exc}")
         logger.info(f"Navigating to: {url}")
         driver.get(url)
 
@@ -686,6 +699,7 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
         # supplied one, play the ready sound exactly once.
         _stop_scrape_music(music_proc, ding=True)
         music_proc = None
+        bell_at = time.monotonic()
 
         if checkpoint is not None:
             try:
@@ -702,6 +716,7 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
                     checkpoint_result.get("error")
                     or "interactive checkpoint was not confirmed"
                 )
+            seconds_to_enter = round(time.monotonic() - bell_at, 1)
 
         try:
             final_url = driver.current_url
@@ -752,9 +767,12 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
         # Silent, plausible, and it inverts the one measurement the triptych
         # exists to take.
         #
-        # get_log() DRAINS: this is the one and only read. A second
-        # get_log("performance") anywhere returns [], so any debug drain added
-        # above this line steals the whole flight and leaves the ledger empty.
+        # get_log() DRAINS: this is the one read of the flight. A second
+        # get_log("performance") returns only what came after the first, so
+        # any debug drain added between driver.get(url) and here steals the
+        # flight and leaves the ledger empty. The start-page read just before
+        # driver.get(url) is the one exception (2026-10-02): it runs before
+        # the page under test is requested.
         if verbose: logger.info("🛜 Draining CDP performance log (network flight recorder)...")
         cdp_events = []
         try:
@@ -901,6 +919,10 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
             # not as bad -- fail-open on purpose, because marking years of
             # legacy captures suspect is noise wearing caution's hat.
             "source_provenance": source_provenance,
+            # SECONDS FROM THE BELL TO ENTER (2026-10-02): how long the human
+            # spent on the page after it loaded, so a glance can be told from
+            # two clicks and a wait. None when no human was asked.
+            "seconds_bell_to_enter": seconds_to_enter,
             "headers": actual_headers
         }
         headers_path = output_dir / "headers.json"
@@ -909,9 +931,99 @@ async def _selenium_capture(params: dict, checkpoint=None) -> dict:
 
         if take_screenshot:
             screenshot_path = output_dir / "screenshot.png"
-            driver.save_screenshot(str(screenshot_path))
+            # THE WHOLE PAGE, NOT THE WINDOW (2026-10-02, walk 3: the operator
+            # had scrolled down, and the screenshot held only the lower part).
+            # save_screenshot takes the viewport as it stands. CDP takes the
+            # page: scroll to the top, read the page's CSS size, and capture
+            # beyond the viewport with a clip of that size. Height is capped at
+            # 16384 px so a very long page cannot ask for a giant image. Any
+            # failure falls back to the window, and the log says so.
+            try:
+                import base64
+                driver.execute_script("window.scrollTo({top: 0, left: 0, behavior: 'instant'});")
+                metrics = driver.execute_cdp_cmd("Page.getLayoutMetrics", {})
+                size = metrics.get("cssContentSize") or metrics.get("contentSize") or {}
+                width = max(1, int(size.get("width") or 1920))
+                height = max(1, min(int(size.get("height") or 1080), 16384))
+                shot = driver.execute_cdp_cmd("Page.captureScreenshot", {
+                    "format": "png",
+                    "captureBeyondViewport": True,
+                    "clip": {"x": 0, "y": 0, "width": width, "height": height, "scale": 1},
+                })
+                screenshot_path.write_bytes(base64.b64decode(shot["data"]))
+            except Exception as exc:
+                logger.warning(f"⚠️ Full-page screenshot failed; saving the window instead: {exc}")
+                driver.save_screenshot(str(screenshot_path))
             artifacts['screenshot'] = str(screenshot_path)
 
+        # --- Response bodies: first-party scripts and data, bounded ---
+        # WHERE DID THIS TEXT COME FROM (2026-10-02, the PocketRender ride): a
+        # single-page app sends a small shell and builds every view from
+        # scripts and fetches, so the answer is in those bodies, not in
+        # source.html, and the browser forgets them when it closes. Kept:
+        # Script, XHR and Fetch responses whose host ends in the final URL's
+        # last two labels (app.botify.com and irf.production.botify.com both
+        # end in botify.com; a co.uk-style host takes in more, never less),
+        # read off the drained ledger with Network.getResponseBody, so nothing
+        # is requested twice. Bounds: 60 files, 2 MB each (cut, and marked),
+        # 20 MB in all, 120 tries. Only bodies/index.jsonl (address without
+        # its query, status, type, bytes) is an artifact; the bodies stay
+        # beside the capture, under the same gitignored browser_cache/ as
+        # everything else the page held.
+        try:
+            import base64
+            final_host = (urlparse(final_url).hostname or "").lower()
+            site = ".".join(final_host.split(".")[-2:])
+            bodies_dir = output_dir / "bodies"
+            rows, seen, kept, total = [], set(), 0, 0
+            for ev in cdp_events:
+                params_ev = ev.get("params", {})
+                response = params_ev.get("response", {})
+                body_url = response.get("url", "")
+                host = (urlparse(body_url).hostname or "").lower()
+                request_id = params_ev.get("requestId")
+                kind = params_ev.get("type")
+                if (ev.get("method") != "Network.responseReceived"
+                        or kind not in ("Script", "XHR", "Fetch")
+                        or not request_id or request_id in seen
+                        or not site or not (host == site or host.endswith("." + site))):
+                    continue
+                if kept >= 60 or total >= 20_000_000 or len(rows) >= 120:
+                    break
+                seen.add(request_id)
+                row = {"url": body_url.split("#", 1)[0].split("?", 1)[0],
+                       "status": response.get("status"), "type": kind,
+                       "mime": response.get("mimeType", "")}
+                try:
+                    got = driver.execute_cdp_cmd("Network.getResponseBody", {"requestId": request_id})
+                except Exception as exc:
+                    row["error"] = (str(exc).strip().splitlines() or [type(exc).__name__])[0][:200]
+                    rows.append(row)
+                    continue
+                if got.get("base64Encoded"):
+                    data, ext = base64.b64decode(got.get("body", "")), ".bin"
+                else:
+                    data = got.get("body", "").encode("utf-8")
+                    ext = ".js" if kind == "Script" else ".json" if "json" in row["mime"] else ".txt"
+                cut = len(data) > 2_000_000
+                data = data[:2_000_000]
+                name = f"{kept + 1:03d}-{kind.lower()}-{hashlib.sha256(body_url.encode('utf-8')).hexdigest()[:10]}{ext}"
+                bodies_dir.mkdir(exist_ok=True)
+                (bodies_dir / name).write_bytes(data)
+                kept += 1
+                total += len(data)
+                row.update({"file": name, "bytes": len(data), "cut": cut})
+                rows.append(row)
+            if rows:
+                bodies_dir.mkdir(exist_ok=True)
+                index_path = bodies_dir / "index.jsonl"
+                index_path.write_text("".join(json.dumps(r) + "\n" for r in rows), encoding="utf-8")
+                artifacts['bodies_index'] = str(index_path)
+            if verbose:
+                logger.info(f"📦 Kept {kept} first-party response bodies ({total} bytes) in {bodies_dir.name}/")
+        except Exception as exc:
+            logger.warning(f"⚠️ Could not keep response bodies: {exc}")
+
 
 
         # 2. Create LLM-Optimized Simplified DOMs (The Symmetrical Lens)