Release 1.1.3 — long skills keep their later steps

Fix release for 1.1.2: the 14 longest skills keep their later phases
after a compaction, a question nobody answered is never taken as an
answer, settings read the same from a subfolder, the commit check works
for a project in a subfolder of a larger repository, every skill opens
with a title, and the shipped allow rules no longer approve deleting a
branch or writing files without asking. Details in CHANGELOG.md; upgrade
steps in UPGRADING.md (v1.1.2 → v1.1.3).

Refs #83
Refs #133

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0185en4t76fPn7315q4c3nko

.claude/rules/agent-memory.md

@@ -5,16 +5,25 @@ paths:
 
 # Agent Memory Rules
 
-**Agent memory is the one place an agent may write without asking — and the only
-one.** `.claude/agent-memory/` is gitignored, per-agent, and never ships. That
-exemption exists because the memory is the agent's own working notes about how to
-do its job, not a project artifact.
+**Agent memory is one of three writes that need no "May I write?"** The other two
+are named in `CLAUDE.md`:
 
-It does **not** extend anywhere else. The Collaboration Protocol still governs
-every file under the code root (`src/`, `Assets/` or `Source/`), `design/`,
-`docs/`, `production/` and `assets/`: ask
-first, naming the path. Writing a memory is never a substitute for asking, and a
-memory must never be used to record something the user declined to have written.
+- **Appends to `production/session-state/active.md`** — the session checkpoint
+  the hooks and the next session rely on. An append, never a rewrite of what is
+  already there.
+- **A subagent's new file at the path its orchestrating skill named.** Only a
+  new file under `production/`, `docs/` or `tests/`, only at a path the skill's
+  prompt names (never one you chose), and only inside a phase the user approved.
+
+`.claude/agent-memory/` is gitignored, per-agent, and never ships. Its exemption
+exists because the memory is the agent's own working notes about how to do its
+job, not a project artifact.
+
+None of the three extends anywhere else. The Collaboration Protocol still governs
+every other write under the code root (`src/`, `Assets/` or `Source/`),
+`design/`, `docs/`, `production/` and `assets/`: ask first, naming the path.
+Writing a memory is never a substitute for asking, and a memory must never be
+used to record something the user declined to have written.
 
 ## Never record a temporary absence as a durable fact
 

.claude/rules/skill-authoring.md

@@ -78,32 +78,16 @@ list.** A gate that fires on forty files does not get fixed, it gets muted — a
 a muted gate is worse than a narrow one, because it reads as coverage. When you
 keep a list, record *why* derivation failed and state the obligation to extend it.
 
-Two further gate rules earned the hard way:
+Two further gate rules:
 
 - **Assert something that cannot be true by accident.** A check for the word
-  "unity" was satisfied by a file path, and case-insensitively by "community".
+  "unity" is satisfied by a file path, and case-insensitively by "community".
   Assert a distinctive token instead.
 - **A gate you have not watched fail is not a gate.** Break the thing it guards,
   confirm it fails, restore. One mutation per tool invocation — a batch that
   times out leaves the repo holding a broken file, and `trap … EXIT` does not
   survive `SIGTERM`.
 
-## Evidence
-
-Every row below is a real instance this rule was written to prevent.
-
-| Where | What was indistinguishable from success |
-|---|---|
-| `team-*` engine step | Specialist skipped when no engine configured — output identical to a run that consulted one |
-| `team-ui` accessibility gate | Passed vacuously; its criterion file did not exist |
-| `yaml-helper` | Locked key in `project.local.yaml` dropped in silence; user saw the default they were overriding |
-| all nine `team-*` | Pipeline printed six agents and ran one, with no statement of the difference |
-| `/security-audit` | Godot-only greps returned zero hits on Unity/Unreal, and zero hits read as clean |
-| `unity/VERSION.md` | "Post-cutoff" table listed pre-cutoff versions, inverting the signal it exists to send |
-| `/security-audit` | `platform.multiplayer` had a reader and no writer; the network category never ran, on any project |
-| `test-evidence-review` | No verdict for "could not check", so an unverifiable story got a verdict saying it was verified |
-| the gate meant to catch all of this | Was a hand-written list, and read green over the gap |
-
 ## The test to apply
 
 Before shipping a skill or gate, ask: **if this step silently did not run, what