@ssheleg/agent-sync 1.10.0 → 1.11.0

package/CHANGELOG.md

@@ -1,3 +1,60 @@
+## v1.11.0 — releasing stops rewinding a cell, and a register with no pattern says so
+
+Two more defects in the same afternoon that produced v1.10.1, both from using this tool
+on a board that had just been worked through.
+
+### Fixed
+
+- **Releasing no longer rewinds a cell that moved.** The restore was verbatim, and in this
+  family the claim cell IS the status cell — so `close then release` silently reopened a
+  row closed with evidence minutes earlier. Caught by `finish` reporting the board
+  uncommitted, not by anybody reading it. The protocol's intent is to remove *this run's*
+  marker, not to rewind the cell, so a changed cell keeps its change, loses only the
+  marker, and the note says so. An untouched cell is still restored exactly.
+- **An id register with no pattern reports instead of crashing.** The script read
+  `nextFreeIdPattern`; every config this family ships writes `pattern`. Both are accepted
+  now — but the defect was the fallback: absence became `re.search("", text)`, which
+  matches the empty string at position 0, so `check` took the found branch and died with
+  `IndexError: no such group` rather than saying the register has no pattern. A component
+  that never received its input approved and then fell over.
+
+### Added
+
+- **`test/claim_cell_test.py`** — 6 cases driving the shipped script as a process against
+  real project directories, because these defects are about what the command does to a
+  file on disk. **Four were watched failing** against the pre-fix script, including that
+  exact `IndexError`. One of the six is a regression test for v1.10.1's own fix: a board
+  where an id is cited by two other rows must still tag the row that owns it.
+
+## v1.10.1
+
+The claim tag could not be placed on nearly half a real board, and the lease was granted
+anyway.
+
+### Fixed
+
+- **A row that CITES an id no longer defeats that id's claim.** `claimTags` in `cell` mode
+  selected every row *containing* the id, so an ordinary board sentence — "closed by
+  B-12", "blocked until T-1 ships" — made the id ambiguous and the tag was refused.
+  Cross-referencing is what boards do: measured on `sshlg-skills`' own board on
+  2026-08-14, **14 of 41 rows cited others, leaving 19 ids that could never be tagged**.
+
+  The refusal itself was correct — guessing which row carries a claim is how a claim lands
+  on somebody else's work. What was wrong is that `acquire` still returned `won`: the
+  lease was granted while the registry silently carried no claim, which is precisely the
+  state a claim exists to make visible.
+
+  A markdown board declares which row is which in its **first cell**. `acquire` now
+  narrows to the row whose id cell equals the key, and refuses only when no row's id cell
+  matches. Tried *after* the existing marker narrowing, so `release` still trusts the
+  marker it actually wrote rather than re-deriving the row.
+
+  Covered by `check_claim_lands_on_the_row_whose_id_cell_matches()`, which builds a two-row
+  board where the second row cites the first, and asserts the tag lands on exactly one row
+  and that it is the right one — then that release still restores the file byte for byte.
+  Watched failing before the fix: *"acquire wrote no claim — a row citing T-1 defeated the
+  selector, and the lease was granted with the registry unmarked"*.
+
 ## v1.10.0
 
 Following `task-pipeline` v1.53.0, which renamed the artifact root's default from

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@ssheleg/agent-sync",
-  "version": "1.10.0",
-  "description": "Let concurrent coding agents share one project without colliding — leases with TTL, race-free id reservation, a run journal and a generated board, over a pluggable knowledge cloud.",
+  "version": "1.11.0",
+  "description": "Let concurrent coding agents share one project without colliding \u2014 leases with TTL, race-free id reservation, a run journal and a generated board, over a pluggable knowledge cloud.",
   "bin": {
     "agent-sync": "bin/agent-sync.js"
   },
@@ -15,7 +15,7 @@
     "LICENSE"
   ],
   "scripts": {
-    "test": "python3 test/validate.py && python3 test/validate.py --self-test",
+    "test": "python3 test/validate.py && python3 test/validate.py --self-test && python3 test/claim_cell_test.py",
     "prepublishOnly": "python3 test/validate.py"
   },
   "publishConfig": {

package/plugins/agent-sync/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "agent-sync",
   "displayName": "Agent Sync",
-  "version": "1.10.0",
+  "version": "1.11.0",
   "description": "Coordination layer for multi-agent repositories — leases with TTL, race-free ID reservation, a run journal, a cross-repo signal feed and a generated board, over a pluggable knowledge cloud.",
   "author": {
     "name": "ssheleg",

package/plugins/agent-sync/skills/agent-sync/SKILL.md

@@ -4,7 +4,7 @@ description: "Use when several coding agents work one repository at the same tim
 compatibility: "Requires the task-pipeline skill for its stages (npx sshlg-skills install). Needs python3 3.9+ (stdlib only, HTTP included - nothing to pip install) and bash for the hooks. The knowledge backend is configured per project; with none configured it degrades to git-file leases. Enforcement hooks are Claude Code only - on other agents the same checks run as a self-check."
 license: MIT
 metadata:
-  version: "1.10.0"
+  version: "1.11.0"
   author: ssheleg
 ---
 

package/plugins/agent-sync/skills/agent-sync/scripts/agent_sync.py

@@ -33,7 +33,7 @@ from datetime import datetime, timezone
 from pathlib import Path
 from typing import Any
 
-VERSION = "1.10.0"
+VERSION = "1.11.0"
 
 CONFIG_PATH = Path(".claude/agent-sync.json")
 ENV_FILE = Path(".env.agent-sync")
@@ -201,6 +201,24 @@ class Fail(Exception):
 _GLOB_CACHE: dict[str, re.Pattern[str]] = {}
 
 
+
+def id_pattern(spec: dict) -> str | None:
+    """The regex that finds a register's next free id, under either accepted key.
+
+    The script read `nextFreeIdPattern`; every config this family ships writes `pattern`.
+    Both are accepted now, because a key mismatch is a documentation problem and should
+    not be a crash — but ABSENCE returns None rather than `""`, which is the whole defect:
+    `re.search("", text)` matches the empty string at position 0, so `check` took the found
+    branch and died with `IndexError: no such group` instead of saying the register has no
+    pattern. A component that never received its input approved and then fell over.
+    """
+    for k in ("nextFreeIdPattern", "pattern"):
+        v = spec.get(k)
+        if v:
+            return v
+    return None
+
+
 def matches_glob(rel: str, pattern: str) -> bool:
     """Repo-root-anchored glob — ONE implementation, for the guard and for `check`.
 
@@ -1372,7 +1390,12 @@ class Sync:
         path = self.root / spec["file"]
         if not path.exists():
             raise Fail(f"register file {spec['file']} does not exist")
-        m = re.search(spec["nextFreeIdPattern"], path.read_text())
+        pat = id_pattern(spec)
+        if not pat:
+            raise Fail(f"register has no id pattern — set `pattern` (or the legacy "
+                       f"`nextFreeIdPattern`) for {spec['file']}; without one there is "
+                       f"nothing to read a next free id out of")
+        m = re.search(pat, path.read_text())
         if not m:
             raise Fail(f"could not read the next free id out of {spec['file']}")
         return int(m.group(1))
@@ -1668,6 +1691,22 @@ class Sync:
                     if len(narrowed) == 1:
                         hits = narrowed
 
+            # A board says which row is which in its FIRST cell. Every other match is a
+            # row CITING this id — "closed by B-12", "blocked until T-1 ships" — and boards
+            # cross-reference constantly: on the family's own board 14 rows cited others, so
+            # 19 of 41 ids had two candidate rows and their claim was refused. The refusal
+            # was right; granting the lease anyway was not, because the registry then carries
+            # no claim while `acquire` reports `won`. Tried after the marker so a release
+            # still trusts what it actually wrote.
+            if len(hits) > 1:
+                own = []
+                for i in hits:
+                    cells = self._row_cells(lines[i])
+                    if cells and cells[0].strip() == key:
+                        own.append(i)
+                if len(own) == 1:
+                    hits = own
+
             if len(hits) > 1:
                 notes.append(f"{rel}: `{key}` appears in {len(hits)} table rows — refusing "
                              "to guess which one is the claim. Narrow the pattern or edit by hand")
@@ -1704,7 +1743,23 @@ class Sync:
             else:
                 if saved is None:
                     continue                      # nothing of ours to undo
-                cells[idx] = saved
+                # Restoring VERBATIM loses any edit made while the claim was held, and in
+                # this family the claim cell IS the status cell — so `close then release`
+                # silently reopened a row closed with evidence minutes earlier (B-35,
+                # 2026-08-14, caught by `finish` reporting the board uncommitted, not by
+                # anybody reading it). The protocol's intent is to remove this run's
+                # marker, not to rewind the cell, so when the text moved on, strip the
+                # marker and keep the movement.
+                template = spec.get("held") or "{prev} (claimed: {holder})"
+                marker = template.replace("{prev}", "").replace("{holder}", self.rid).strip()
+                expected = template.replace("{prev}", saved.strip()).replace("{holder}", self.rid)
+                if current.strip() == expected.strip() or not marker:
+                    cells[idx] = saved
+                else:
+                    stripped = current.replace(marker, "").rstrip()
+                    cells[idx] = (stripped if stripped.strip() else saved)
+                    notes.append(f"{rel}: `{key}`'s cell changed while the claim was held — "
+                                 f"kept the change and removed only the claim marker")
                 state.get(key, {}).pop(str(rel), None)
                 if not state.get(key):
                     state.pop(key, None)
@@ -3166,10 +3221,17 @@ def check_setup(root: Path) -> tuple[list[str], list[str], list[str]]:
             problems.append(f"register {reg}: file '{spec.get('file')}' does not exist")
             continue
         text = f.read_text()
+        pat = id_pattern(spec)
+        if not pat:
+            problems.append(f"register {reg}: no id pattern — set `pattern` (or the legacy "
+                            f"`nextFreeIdPattern`). An absent one used to become "
+                            f"`re.search(\"\", text)`, which matches at position 0, so this "
+                            f"check took the found branch and crashed instead of reporting")
+            continue
         try:
-            m = re.search(spec.get("nextFreeIdPattern", ""), text)
+            m = re.search(pat, text)
         except re.error as exc:
-            problems.append(f"register {reg}: nextFreeIdPattern is not valid regex ({exc})")
+            problems.append(f"register {reg}: id pattern is not valid regex ({exc})")
             continue
         if not m:
             problems.append(f"register {reg}: nextFreeIdPattern matches nothing in "