v0.6.0: replace retired gemini agent with agy (Antigravity CLI) + integration tests

.github/workflows/ci.yml

@@ -15,13 +15,28 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v5
 
+      # One file per invocation — `bash -n a b c` only checks the first file
+      # (the rest become positional parameters).
       - name: Bash syntax
         run: |
           bash -n fixbuddy.sh
           bash -n fixbuddy-wizard.sh
+          bash -n tests/integration.sh
+          bash -n tests/stubs/agent
+          bash -n tests/stubs/gh
 
       - name: Install ShellCheck
         run: sudo apt-get update && sudo apt-get install -y shellcheck
 
       - name: ShellCheck
-        run: shellcheck fixbuddy.sh fixbuddy-wizard.sh
+        run: shellcheck fixbuddy.sh fixbuddy-wizard.sh tests/integration.sh tests/stubs/agent tests/stubs/gh
+
+  integration:
+    name: Integration tests (mocked CLIs)
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v5
+
+      - name: Run integration tests
+        run: tests/integration.sh

CHANGELOG.md

@@ -4,6 +4,38 @@ All notable changes to fixbuddy are documented here. The format is based on
 [Keep a Changelog](https://keepachangelog.com/), and this project aims to follow
 [Semantic Versioning](https://semver.org/).
 
+## [0.6.0] - 2026-06-12
+
+Google retires the Gemini CLI on 2026-06-18; its successor is the Antigravity
+CLI (`agy`). fixbuddy v0.6.0 swaps the agent and gains an offline integration
+test suite.
+
+### Added
+- **`agy` agent** (Antigravity CLI) as fix or review agent. Invocation details
+  that matter: `--add-dir <project>` (agents launch from the operator's CWD),
+  `--print-timeout` pinned 60s above `--agent-timeout` so fixbuddy's watchdog
+  classifies timeouts (agy itself exits 0 on its internal timeout — fixbuddy
+  also detects that output and treats it as `fix:blocked`/auto-requeue), and
+  `--sandbox` on verify/review as defense in depth (agy has no read-only mode).
+- **Integration tests** (`tests/integration.sh`) — deterministic, offline,
+  zero new dependencies: stub `gh`/agent CLIs plus a local bare repo as
+  `origin`, covering happy path, false positive, review rejection, check gate,
+  dry-run read-only, and crash classification. Run in CI.
+
+### Removed (breaking)
+- **`gemini` agent.** Passing `gemini` (flag or config) now exits with a
+  migration message. Replace `fix_agent`/`review_agent` values with `agy`.
+  Note: agy in verify/review runs sandboxed but NOT read-only — the old
+  `--approval-mode plan` has no equivalent in the Antigravity CLI.
+
+### Security & robustness
+- The two read-only-by-contract stages are now guarded deterministically (no
+  agent CLI offers an enforced read-only mode): after **every** verify outcome
+  (proceed, false positive, blocked, crash) worktree files the verify agent
+  wrote are stashed and commits it created on the base branch are discarded,
+  and the **review** branch is pinned to the reviewed commit — commits a
+  reviewer creates are discarded, so only the reviewed commit is ever pushed.
+
 ## [0.5.0] - 2026-06-12
 
 Security hardening from a full audit, plus five new features. No breaking changes
@@ -49,4 +81,5 @@ to existing flags.
 
 Predate this changelog. See the git history and the `v0.4.0` / `v0.3.2` tags.
 
+[0.6.0]: https://github.com/Codevena/fixbuddy/compare/v0.5.0...v0.6.0
 [0.5.0]: https://github.com/Codevena/fixbuddy/compare/v0.4.0...v0.5.0

CONTRIBUTING.md

@@ -21,9 +21,13 @@ Run these before opening a pull request:
 ```bash
 bash -n fixbuddy.sh
 bash -n fixbuddy-wizard.sh
-shellcheck fixbuddy.sh fixbuddy-wizard.sh
+shellcheck fixbuddy.sh fixbuddy-wizard.sh tests/integration.sh tests/stubs/agent tests/stubs/gh
+tests/integration.sh
 ```
 
+`tests/integration.sh` runs the full pipeline offline against stubbed `gh`/agent
+CLIs and a local bare repository — no network, no API keys, a few seconds.
+
 If `shellcheck` is not available locally, the GitHub Actions workflow will run it for pull requests.
 
 ## Pull Request Guidelines
@@ -36,7 +40,8 @@ If `shellcheck` is not available locally, the GitHub Actions workflow will run i
 
 ## Manual Test Checklist
 
-For behavior changes, test against a disposable repository when possible:
+Most of these paths are covered by `tests/integration.sh`; for behavior changes,
+additionally test against a disposable repository when possible:
 
 1. `--dry-run` lists expected issues.
 2. A false-positive verification closes or labels the issue correctly.

NEXT_SESSION.md

@@ -1,25 +1,39 @@
 # Next session
 
-All P0/P1 review items and the full P2 backlog (#6 repo topics, #7 `install.sh`, #8 GitHub Action wrapper) are done.
+v0.6.0 is implemented on branch `feat/agy-agent-v0.6.0`: the retired `gemini`
+agent is replaced by `agy` (Antigravity CLI — Gemini CLI shuts down 2026-06-18)
+and a deterministic offline integration test suite landed.
 
 ## Status snapshot
 
-- Branch `main`, working tree clean.
-- **`v0.4.0` is tagged and pushed**, with a floating **`v1`** ref pointing at it. Consumers use `uses: Codevena/fixbuddy@v1`.
-- `action.yml` (composite action, pure bash) + `.github/workflows/action-smoke.yml` (dry-run smoke test) shipped in #8. README has a "Use in GitHub Actions" section.
-- `fixbuddy.sh` `--dry-run` skips the agent-CLI presence check (agent-name validation still runs) so the smoke test works on bare runners.
-- `install.sh` pins `DEFAULT_REF="v0.4.0"`; README Quick Start one-liner points at `v0.4.0`.
-- `SHA256SUMS` holds hashes of `fixbuddy.sh` + `fixbuddy-wizard.sh`. **Regenerate it (`shasum -a 256 fixbuddy.sh fixbuddy-wizard.sh > SHA256SUMS`) whenever either script changes, before cutting a new tag** — `install.sh` verifies checksums and fails closed.
-- When cutting a new release: bump `VERSION` in `fixbuddy.sh`, the header comment, `install.sh` `DEFAULT_REF`, the README one-liner; regenerate `SHA256SUMS`; tag `vX.Y.Z`; then `git tag -f v1 vX.Y.Z && git push origin v1 --force` to move the floating major ref.
-- Definition-of-Done gate from `~/.claude/CLAUDE.md` still applies. Note from #8: `codex exec` hung at 0% CPU again — the OpenCode fallback (`opencode run --dangerously-skip-permissions "$(<file)"`) worked. Write reviewer prompts with the Write tool, not Bash heredocs.
+- Branch `feat/agy-agent-v0.6.0`, all work committed; awaiting merge to `main`
+  and the release tag.
+- Spec: `docs/superpowers/specs/2026-06-12-agy-agent-and-integration-tests-design.md`
+  (includes the locally verified agy CLI behavior: no read-only mode, `--sandbox`
+  semantics, `--print-timeout` default 5m with **exit code 0** on timeout).
+- Plan: `docs/superpowers/plans/2026-06-12-agy-agent-and-integration-tests.md`.
+- Tests: `tests/integration.sh` — 9 scenarios against stubbed `gh`/agent CLIs and
+  a local bare-repo origin. Runs in CI (`integration` job) alongside shellcheck.
+- Audit history moved to `docs/audit/2026-06-10-findings.md` (all 18 findings
+  fixed in v0.5.0).
 
-## What's next
+## Release checklist (v0.6.0)
 
-P2 is complete. Remaining work lives in the README **Roadmap**:
+Version bumps, CHANGELOG, and `SHA256SUMS` are already done on the branch. After
+merge to `main`:
+
+1. `git tag v0.6.0 && git push origin v0.6.0`
+2. `git tag -f v1 v0.6.0 && git push origin v1 --force` (floating major ref for
+   `uses: Codevena/fixbuddy@v1`)
+3. Verify the CI + action-smoke workflows are green on `main`.
+
+Remember for future releases: regenerate `SHA256SUMS`
+(`shasum -a 256 fixbuddy.sh fixbuddy-wizard.sh > SHA256SUMS`) whenever either
+script changes, before tagging — `install.sh` verifies checksums fail-closed.
+
+## What's next (README Roadmap)
 
-- Config file support
-- More deterministic integration tests with mocked CLIs
 - Optional notifications for run summaries
 - Explicit resume mode for interrupted runs
 
-Pick one and run it through the brainstorm → plan → DoD-gate flow. Verify the action-smoke workflow went green on GitHub after the v0.4.0 push.
+Pick one and run it through the brainstorm → plan → DoD-gate flow.

README.md

@@ -6,7 +6,7 @@
 [![version](https://img.shields.io/github/v/tag/Codevena/fixbuddy?label=version)](https://github.com/Codevena/fixbuddy/tags)
 [![license](https://img.shields.io/badge/license-MIT-green.svg)](LICENSE)
 [![shell](https://img.shields.io/badge/shell-bash-black.svg)](fixbuddy.sh)
-[![agents](https://img.shields.io/badge/agents-claude%20%7C%20codex%20%7C%20opencode%20%7C%20gemini-purple.svg)](#supported-agents)
+[![agents](https://img.shields.io/badge/agents-claude%20%7C%20codex%20%7C%20opencode%20%7C%20agy-purple.svg)](#supported-agents)
 
 fixbuddy reads open GitHub issues, asks one agent to verify and fix each issue, asks a second agent to review the committed diff, then opens a pull request. If enabled, it requests auto-merge after review approval.
 
@@ -22,12 +22,12 @@ The goal is controlled automation: one issue per branch, one issue per PR, expli
 
 Most AI issue-fixers let a single agent write a fix and, at best, review its own work. fixbuddy splits the job across **two different agents from two different vendors**: by default `claude` writes the fix and `codex` reviews the committed diff with a fresh context. The fixer never approves its own work.
 
-It needs no cloud service, no Docker, and no separate API-key broker — it drives the AI coding CLIs you already have installed (`claude`, `codex`, `opencode`, `gemini`), so it runs on the subscriptions you already pay for. The orchestrator is ~1,200 lines of readable Bash.
+It needs no cloud service, no Docker, and no separate API-key broker — it drives the AI coding CLIs you already have installed (`claude`, `codex`, `opencode`, `agy`), so it runs on the subscriptions you already pay for. The orchestrator is ~1,200 lines of readable Bash.
 
 |  | fixbuddy | Copilot coding agent | claude-code-action | OpenHands resolver |
 |---|---|---|---|---|
 | Fix **and** review | two agents, cross-vendor (fixer ≠ reviewer) | one vendor | one vendor | one agent |
-| Choice of agent | claude · codex · opencode · gemini | Copilot's models | Claude only | bring your own LLM |
+| Choice of agent | claude · codex · opencode · agy | Copilot's models | Claude only | bring your own LLM |
 | Where it runs | your machine **or** a GitHub Action | GitHub cloud | GitHub Action | local / Docker |
 | Infra required | bash · git · gh · jq | none (hosted) | GitHub Actions | Docker + API keys |
 | Cost | your existing CLI subscriptions | paid Copilot (premium requests) | API / subscription | your API + compute |
@@ -57,15 +57,15 @@ VERIFY -> FIX -> REVIEW -> PUSH/PR -> optional auto-merge
 Install with the one-liner (macOS and Linux, including WSL2):
 
 ```bash
-curl -fsSL https://raw.githubusercontent.com/Codevena/fixbuddy/v0.5.0/install.sh | bash
+curl -fsSL https://raw.githubusercontent.com/Codevena/fixbuddy/v0.6.0/install.sh | bash
 ```
 
-This downloads the pinned `v0.5.0` scripts into `~/.local/bin` (or `/usr/local/bin`), makes them executable, and prints a PATH hint if needed. Override the location with `| bash -s -- --prefix /custom/bin` or track the latest commit with `--ref main`.
+This downloads the pinned `v0.6.0` scripts into `~/.local/bin` (or `/usr/local/bin`), makes them executable, and prints a PATH hint if needed. Override the location with `| bash -s -- --prefix /custom/bin` or track the latest commit with `--ref main`.
 
 **Prefer to read before you run?** The installer is short — inspect it first, then run it:
 
 ```bash
-curl -fsSL https://raw.githubusercontent.com/Codevena/fixbuddy/v0.5.0/install.sh -o install.sh
+curl -fsSL https://raw.githubusercontent.com/Codevena/fixbuddy/v0.6.0/install.sh -o install.sh
 less install.sh        # read it
 bash install.sh        # then run it
 ```
@@ -121,7 +121,7 @@ Both `--fix-agent` and `--review-agent` must be installed. They may point to the
 | `claude` | `claude --dangerously-skip-permissions -p -` | Full tool access. |
 | `codex` | `codex exec --dangerously-bypass-approvals-and-sandbox` | Full tool access. |
 | `opencode` | `opencode run --dangerously-skip-permissions` | Full tool access. |
-| `gemini` | `gemini -p ... --approval-mode {plan\|yolo}` | Read-only style `plan` mode for verify/review; `yolo` for fix. |
+| `agy` | `agy --dangerously-skip-permissions --add-dir <project> -p ...` | Antigravity CLI (Gemini's successor). Verify/review add `--sandbox` (terminal restrictions — not read-only). |
 
 These agent invocations are intentionally powerful. Run fixbuddy only against repositories and issue content you trust.
 
@@ -134,8 +134,8 @@ These agent invocations are intentionally powerful. Run fixbuddy only against re
 | `--label <name>` | Include only issues with this label. Repeatable | none |
 | `--severity <level>` | Include issues labeled `severity:<level>` | none |
 | `--max <n>` | Maximum issues to process in this run | unlimited |
-| `--fix-agent <agent>` | `claude`, `codex`, `opencode`, or `gemini` | `claude` |
-| `--review-agent <agent>` | `claude`, `codex`, `opencode`, or `gemini` | `codex` |
+| `--fix-agent <agent>` | `claude`, `codex`, `opencode`, or `agy` | `claude` |
+| `--review-agent <agent>` | `claude`, `codex`, `opencode`, or `agy` | `codex` |
 | `--max-retries <n>` | Retry count after review rejection | `1` |
 | `--agent-timeout <secs>` | Wall-clock timeout per agent call | `1200` |
 | `--crash-abort <n>` | Abort after consecutive agent crashes | `3` |
@@ -215,8 +215,10 @@ fixbuddy creates and manages these labels:
 
 - fixbuddy refuses to start if the target checkout has a dirty working tree.
 - Each issue gets a fresh `fix/issue-N` branch.
+- The verify stage is read-only by contract, but no agent CLI enforces that: files it writes are stashed and commits it creates on the base branch are discarded before the fix branch is created.
 - The fix agent is instructed to stage only relevant files and to avoid generated artifacts.
 - The review agent receives the committed diff and must reject unrelated changes.
+- If the reviewer creates commits, the branch is reset to the reviewed commit — only the reviewed commit is ever pushed.
 - Push happens only after review approval.
 - `fix:applied` is added only after GitHub reports that the PR is merged.
 - Cleanup stashes uncommitted agent output before deleting temporary branches.
@@ -275,11 +277,11 @@ Use one agent for both roles:
   --fix-agent claude --review-agent claude --max 3
 ```
 
-Use Gemini as a read-only reviewer:
+Use agy (Antigravity CLI) as a cross-vendor reviewer:
 
 ```bash
 ./fixbuddy.sh --repo owner/repo --project ~/code/repo \
-  --fix-agent claude --review-agent gemini
+  --fix-agent claude --review-agent agy
 ```
 
 ## Use in GitHub Actions
@@ -343,7 +345,7 @@ If `github-token` is empty the action fails fast with a clear error. Pass a diff
 
 ### Prerequisites in CI
 
-GitHub-hosted runners do **not** ship the agent CLIs (`claude`, `codex`, `opencode`, `gemini`). Install whichever ones you pass to `fix-agent` / `review-agent` in a step *before* the `Codevena/fixbuddy` step — pinning them to a known version is recommended. Consult each agent's own documentation for the current install command.
+GitHub-hosted runners do **not** ship the agent CLIs (`claude`, `codex`, `opencode`, `agy`). Install whichever ones you pass to `fix-agent` / `review-agent` in a step *before* the `Codevena/fixbuddy` step — pinning them to a known version is recommended. Consult each agent's own documentation for the current install command. Note that `agy` has no npm package — install it with the vendor script: `curl -fsSL https://antigravity.google/cli/install.sh | bash`.
 
 The action does not read API keys itself; each agent CLI reads its own environment variable (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, and so on). Provide them from `secrets` at the job or workflow level, as shown above.
 
@@ -392,7 +394,6 @@ Native Windows is not tested. WSL2 is the recommended Windows environment.
 
 ## Roadmap
 
-- More deterministic integration tests with mocked CLIs
 - Optional notifications for run summaries
 - Explicit resume mode for interrupted runs (interrupting a run cleans up the in-progress branch; the issue is retried automatically on the next run)
 

SHA256SUMS

@@ -1,2 +1,2 @@
-288eda7831f6b3bd7078417e46ac8270a7f8b0175a288748cf9f3bac55c23cb8  fixbuddy.sh
-d6868daf05471c46ab969fe50b441af004d6622f664c529912da82d91665ac73  fixbuddy-wizard.sh
+b63b3e549ddb30f154482abf585313d56edd281e2493d878b6f154693c1d1674  fixbuddy.sh
+ba3ba121a60ecb37fc5ca796f3cf5279a24a54566aeb2bb0dca3c3a1f4b7bd2f  fixbuddy-wizard.sh

action.yml

@@ -15,11 +15,11 @@ inputs:
     required: false
     default: "."
   fix-agent:
-    description: "Fix agent: claude | codex | opencode | gemini"
+    description: "Fix agent: claude | codex | opencode | agy"
     required: false
     default: "claude"
   review-agent:
-    description: "Review agent: claude | codex | opencode | gemini"
+    description: "Review agent: claude | codex | opencode | agy"
     required: false
     default: "codex"
   severity: