Sync crill 0.5.2 public release inputs

Author: spilist

.release-notes/v0.5.2.md

@@ -0,0 +1,20 @@
+# v0.5.2
+
+Public binary mirror for the private crill release.
+
+- Source repo: `corca-ai/crill`
+- Source release: https://github.com/corca-ai/crill/releases/tag/v0.5.2
+- Public repo: `corca-ai/crill-cli`
+
+## Assets
+
+- `checksums.txt` - `sha256:5de1ad7b9a3d39b41b5ffe47f44bc80d3d0a6c668908fc645d337e320aefcd1f`
+- `crill_0.5.2_darwin_arm64.tar.gz` - `sha256:dc3076ca683cfe56cb216a33b7cdd902b6bfd2d18030573e9fa0bab90ec4dab2`
+
+## Access Control
+
+Binary download is public. Gated crill commands still require `crill auth login <email>`.
+
+## Release Boundary
+
+This sync uploads the public release assets, then pushes the matching tag so `corca-ai/crill-cli` CI can update `corca-ai/homebrew-tap`.

README.md

@@ -246,6 +246,6 @@ crill runs audit runs/current/ --baseline runs/baseline/ --max-llm-avg-regressio
 
 ## Latest Public Release
 
-- Version: `0.5.1`
-- Release: https://github.com/corca-ai/crill-cli/releases/tag/v0.5.1
-- Mirrored from upstream release: https://github.com/corca-ai/crill/releases/tag/v0.5.1
+- Version: `0.5.2`
+- Release: https://github.com/corca-ai/crill-cli/releases/tag/v0.5.2
+- Mirrored from upstream release: https://github.com/corca-ai/crill/releases/tag/v0.5.2

docs/install.md

@@ -29,7 +29,8 @@ The current macOS binary is unsigned. If macOS blocks the first launch, open
 Finder and approve the binary once:
 
 1. Open `/opt/homebrew/bin/crill` on Apple Silicon, or `/usr/local/bin/crill`
-   on Intel Macs.
+   on Intel Macs. If the symlink does not expose the approval prompt cleanly,
+   open the real bundled binary instead: `.../Cellar/crill/<version>/libexec/crill`.
 2. Right-click, choose **Open**, and confirm once.
 3. Return to the terminal and run `crill` normally.
 
@@ -53,6 +54,19 @@ crill skills install
 The installed public `crill` skill now includes a shared onboarding reference
 at `references/onboarding.md`.
 
+If you want one quick readiness snapshot before switching sessions, run:
+
+```bash
+crill doctor --json
+```
+
+The important rows for the handoff are `gate.session`, `install.skill.global`,
+and `install.skill.claude-link`.
+
+That is only the install-to-skill handoff. In the next session, the skill may
+still continue onboarding until `providers.default` and `home.saved-ios-device`
+also become `ok`.
+
 Now close this agent session. Open a new one, invoke the `crill` skill, and
 have it read its own `references/onboarding.md` before it continues. That
 reference owns provider selection, iOS setup, app resolution, the first-run

docs/internal-ios-trial.md

@@ -56,7 +56,9 @@ export 받습니다.
 - `@corca.ai` 주소라면 activation code / invitation key 없이 로그인돼야 한다.
 - 사람 개입이 필요한 단계는 에이전트가 스스로 우회하지 말고 정확히 요청한다.
 - `crill skills install` 까지 끝낸다.
-- 마지막에 `crill doctor` 또는 `crill doctor --json` 으로 상태를 확인한다.
+- 마지막에 `crill doctor --json` 으로 상태를 확인한다.
+- 이때 특히 `gate.session`, `install.skill.global`, `install.skill.claude-link`
+  가 기대대로 보이는지 확인한다.
 - 끝나면 "이제 새 세션에서 `crill` skill 을 실행해 보세요" 라고 안내한다.
 
 호스트마다 skill 호출 표기가 다를 수 있으므로, 기본 원칙은 **현재 에이전트가
@@ -71,7 +73,15 @@ export 받습니다.
 - skill 이 먼저 `crill doctor --json` 으로 현재 상태를 읽는다.
 - skill 이 자기 `references/onboarding.md` 계약을 읽고 첫 미완료 단계부터 이어간다.
 - 첫 skill 실행 상태라면 온보딩 경로로 들어간다.
-- 아직 iOS 기기 선택이 안 되어 있으면 `crill setup --ios` 를 진행한다.
+- `gate.session`, `providers.default`, `home.saved-ios-device` 가 각각
+  온보딩 phase gate 로 쓰인다.
+- 다만 `gate.reachable` 가 깨졌으면 재로그인이 아니라 gate/network 복구로
+  먼저 돌려야 한다.
+- 아직 iOS 기기 선택이 안 되어 있거나, 저장된 기기가 지금 연결되어 있지 않으면
+  `crill setup --ios` 를 진행한다.
+- `home.saved-ios-device == ok` 는 저장된 기기가 아직 연결돼 있고 signing/provider
+  readiness 가 크게 깨져 보이지 않는다는 뜻이지, 모든 WDA/trust/Appium seam 을
+  완전히 증명한 것은 아니다.
 - 앱을 무엇을 테스트할지 물어본다.
 - `crill ios apps --json` 으로 설치된 앱을 보고 앱 이름 또는 bundle id 를
   해석한다.
@@ -96,8 +106,12 @@ export 받습니다.
 - `@corca.ai` 로그인에서 activation code 없이 자연스럽게 들어가는가
 - 에이전트가 사람 개입이 필요한 순간을 정확히 구분하는가
 - provider 를 어떻게 감지했고 무엇을 쓰겠다고 말하는지가 충분히 자연스러운가
+- provider 선택/저장이 `crill provider status --json` 과 `crill config provider set`
+  흐름으로 자연스럽게 이어지는가
 - 새 세션 handoff 가 자연스러운가
 - `crill` skill 첫 실행이 `doctor` 기반으로 온보딩에 들어가는가
+- 첫 LLM 결정이 실패하면 애매하게 back/skip 하지 않고 provider 복구로 되돌아가는가
+- 그때 기존 `providers.default == ok` 만 믿고 같은 provider 로 바로 재시도하지 않는가
 - 앱 선택 뒤 첫 `scan -> report` 경로가 실제로 닫히는가
 
 ## 참여자에게 보낼 Slack 문안

scripts/release/crill.rb

@@ -15,7 +15,15 @@ class Crill < Formula
     sha256 "__ARM64_SHA__"
 
     def install
-      bin.install "crill"
+      libexec.install Dir["*"]
+      bin.install_symlink libexec/"crill"
+    end
+
+    def post_install
+      # Best-effort warm-up so the first operator launch after install or
+      # upgrade does not also pay the cold-start bundle load cost. Keep the
+      # install successful even if Gatekeeper or host policy blocks execution.
+      quiet_system libexec/"crill", "--version"
     end
   end
 

skills/crill/SKILL.md

@@ -25,10 +25,24 @@ crill doctor --json
 ```
 
 2. Use the onboarding contract phases to decide what is already complete and
-   what still needs work. Skip completed phases and resume from the first
-   incomplete phase.
+   what still needs work. Treat these rows as the phase gates:
 
-3. If there is no operator session yet, stop and tell the user exactly:
+- `gate.session` must be `ok`
+- `providers.default` must be `ok`
+- `home.saved-ios-device` must be `ok`
+
+Skip completed phases and resume from the first incomplete phase.
+
+3. If `gate.reachable` is `fail`, or `gate.session.extra.reason` is
+   `gate_unreachable` / `server_unreachable`, stop on the gate/network problem
+   and ask the operator to restore reachability before continuing from:
+
+```bash
+crill doctor --json
+```
+
+4. If `gate.session` is not `ok` for missing or stale-session reasons, tell
+   the user exactly:
 
 ```bash
 Run `crill auth login <email>` in your terminal first, then continue from this onboarding contract.
@@ -42,9 +56,14 @@ When the onboarding contract reaches provider selection, inspect the available
 auth sources:
 
 ```bash
-crill provider status
+crill provider status --json
 ```
 
+If `providers.default` is already `ok`, keep it and continue unless the
+immediately previous first scan failed on the first LLM decision. In that
+failure case, re-run provider recovery instead of trusting the saved default
+again.
+
 If exactly one provider is available, say which one was detected and persist it:
 
 ```bash
@@ -127,6 +146,9 @@ Human-only blockers to treat this way:
 3. Always include the exact resume command.
 4. Use [`references/ios-real-device.md`](references/ios-real-device.md) only
    after reading the current `crill` output and error text.
+5. If the first LLM decision fails during the first scan, stop and return to
+   provider recovery instead of letting the scan drift. Do not trust
+   `providers.default == ok` alone after that failure.
 
 ## What is not gated
 

skills/crill/references/onboarding.md

@@ -19,7 +19,19 @@ Read the current state first:
 crill doctor --json
 ```
 
-When the machine has no operator session yet, stop and tell the user exactly:
+Completion rule: `gate.session` from `crill doctor --json` must be `ok`.
+
+If `gate.reachable` is `fail`, or `gate.session.extra.reason` is
+`gate_unreachable` / `server_unreachable`, stop on the gate/network problem.
+Do not tell the operator to log in again yet. Ask them to restore gate reachability,
+then resume with:
+
+```bash
+crill doctor --json
+```
+
+When `gate.session` is missing, invalid, or bound to a different gate, stop
+and tell the user exactly:
 
 ```bash
 Run `crill auth login <email>` in your terminal first, then continue from this onboarding contract.
@@ -34,11 +46,16 @@ Goal: one preferred provider is chosen and persisted for future scans.
 Inspect available auth sources:
 
 ```bash
-crill provider status
+crill provider status --json
 ```
 
 Rules:
 
+0. If `providers.default` from `crill doctor --json` is already `ok`, keep it
+   and continue, unless the immediately previous first scan failed on the first
+   LLM decision. In that failure case, re-enter Phase 2 and require a fresh
+   provider recovery step instead of trusting the saved default again.
+
 1. If exactly one provider is clearly available, tell the operator which one
    was detected and persist it:
 
@@ -77,9 +94,16 @@ explicit `--provider` flag overrides it.
 
 ## Phase 3 — iOS Device Setup
 
-Goal: one real iOS device is selected and usable for `crill`.
+Goal: one real iOS device is selected, connected, and not obviously blocked on
+host-side signing/provider readiness.
 
-If no saved iOS device is present yet, run:
+Completion rule: `home.saved-ios-device` from `crill doctor --json` must be `ok`.
+This proves the saved device is still connected and signing/provider readiness
+looks plausible; it does not replace `crill setup --ios` if later app listing
+or scan output shows a real Appium/WDA/trust blocker.
+
+If no saved iOS device is present yet, or the saved one is no longer connected,
+run:
 
 ```bash
 crill setup --ios
@@ -152,6 +176,11 @@ crill report runs/<timestamp>/
 The golden path is `scan -> report`. `crill runs audit` is maintainer
 follow-up, not part of the participant default path.
 
+If the first LLM decision fails, stop and return to provider recovery instead
+of letting the scan continue with blind back-navigation. Do not trust the
+existing `providers.default == ok` row alone after that failure; require a
+fresh provider recovery step first.
+
 ## Recovery Rules
 
 1. Prefer current `crill` output over stale memory.