openclaw-code-agent 4.2.0 → 4.2.2

package/README.md

@@ -17,33 +17,41 @@ Use it when you want to start coding work from Telegram, Discord, or another Ope
 - **Full session lifecycle**. Suspend, resume, fork, interrupt, and recover sessions across restarts with persisted metadata and output.
 - **Explicit goal-task loops**. Opt into verifier-driven repair loops or Ralph-style completion loops when you need iterative autonomous execution toward a specific goal.
 - **Real operator visibility**. `agent_sessions`, `agent_output`, and `agent_stats` show status, buffered output, duration, and USD cost.
-- **Two harnesses, one control plane**. Claude Code and Codex share the same tools, routing, notification pipeline, and worktree strategy model while each backend uses its own native execution substrate.
+- **Two harnesses, one control plane**. Claude Code and Codex share the same tools, routing, notification pipeline, and worktree strategy model while each backend uses its own adapter and resume substrate.
 - **One continuation path**. Follow-ups, approvals, revisions, interrupts, and redirects all continue the existing session instead of launching a duplicate.
 
 This plugin is separate from OpenClaw's bundled `acpx` runtime plugin and bundled core `codex` plugin. Those own adjacent OpenClaw runtime/provider surfaces; `openclaw-code-agent` owns chat orchestration and repository follow-through for its own Claude Code and Codex harnesses. See [docs/ACP-COMPARISON.md](docs/ACP-COMPARISON.md) for the boundary details.
 
-## From Prompt To Shipped Branch
+## From Chat To Resolved Work
 
 1. Ask OpenClaw to launch a coding session from chat.
-2. Review the plan in the same thread before implementation starts.
-3. Let the agent finish in an isolated worktree.
-4. Merge into the base branch, open a PR, defer the decision, or discard the sandbox from the same control plane.
+2. Choose the review style you want: direct execution, user plan approval, delegated review, or explicit worktree decisions.
+3. Let the agent finish in an isolated worktree when branch follow-through is enabled.
+4. Merge into the base branch, open a PR, defer the decision, or discard the sandbox from the same thread.
 
-### Plan First
+### Direct Completion
 
-The core loop is plan review. Claude Code and Codex feed the same approval UX: the plugin receives a structured plan artifact, keeps execution blocked until approval, and continues the same session after the plan is approved. If the user asks for revisions, the revised plan becomes the latest actionable version for that same session.
+For small trusted changes, an orchestrator can launch a session, let Codex or Claude Code finish, and report the verified outcome back to chat. The session stays observable through launch, completion, cost, duration, and commit summary.
 
-### Finish Cleanly
+![Direct completion](https://raw.githubusercontent.com/goldmar/openclaw-code-agent/main/assets/no-plan.png)
 
-When a task completes, the plugin can leave the branch for review, merge it automatically, open or update a PR, or wake the orchestrator with diff context. In `ask`, the user gets explicit decision buttons in the originating thread. In `delegate`, the orchestrator reviews the worktree result and escalates user-facing decisions such as PR creation.
+### Plan Review
+
+The default review loop is plan-first. Claude Code and Codex feed the same approval UX: the plugin receives a structured plan artifact, blocks implementation until approval, and continues the same session after the plan is approved. The user can approve, request a revision, or reject the plan from the originating thread.
+
+![Plan review](https://raw.githubusercontent.com/goldmar/openclaw-code-agent/main/assets/plan-review.png)
+
+### Worktree Decisions
+
+In `ask`, the user controls branch follow-through after the agent finishes. Current buttons adapt to state: new branches can show **Merge**, **Open PR**, **Later**, and **Discard**; branches with an existing PR can show **View PR** and **Sync PR** instead of **Open PR**.
 
-![Delegated worktree flow](https://raw.githubusercontent.com/goldmar/openclaw-code-agent/main/assets/delegate-readme.gif)
+![Ask-mode worktree decisions](https://raw.githubusercontent.com/goldmar/openclaw-code-agent/main/assets/worktree-ask.png)
 
-*`delegate` keeps the main checkout clean while the branch lifecycle happens in the worktree. The chat thread stays current on what was attempted, what changed, and what follow-through is needed.*
+### Delegated Worktrees
 
-![Ask-mode worktree decisions](https://raw.githubusercontent.com/goldmar/openclaw-code-agent/main/assets/ask-readme.gif)
+In `delegate`, the orchestrator reviews the completed worktree and attempts the merge follow-through when the change is clean. The agent edits files in the managed worktree so the main checkout is not touched during implementation; after review, delegated follow-through merges the finished branch back to the base branch unless a conflict, error, or explicit policy requires escalation.
 
-*`ask` keeps the human in the loop for branch follow-through. Current buttons adapt to state: new branches can show **Merge**, **Open PR**, **Later**, and **Discard**; branches with an existing PR can show **View PR** and **Sync PR** instead of **Open PR**. Older recordings may show prior wording.*
+![Delegated worktree flow](https://raw.githubusercontent.com/goldmar/openclaw-code-agent/main/assets/worktree-delegate.png)
 
 ### Worktree Lifecycle
 
@@ -70,6 +78,19 @@ openclaw gateway restart
 openclaw plugins inspect openclaw-code-agent --runtime --json
 ```
 
+If OpenClaw blocks installation with a dangerous-code scanner finding for
+`child_process`, that is expected for this trusted plugin because it launches
+local coding harnesses and git tooling. Review the rationale in
+[docs/SECURITY.md](docs/SECURITY.md), then rerun the trusted package/source with
+the unsafe-install override:
+
+```bash
+openclaw plugins install openclaw-code-agent --force --pin --dangerously-force-unsafe-install
+```
+
+Use that override only for a package/source you already trust. When validating a
+specific reviewed release, add its version after the package name.
+
 Add the smallest useful config under `plugins.entries["openclaw-code-agent"]` in `~/.openclaw/openclaw.json`:
 
 ```json
@@ -103,7 +124,7 @@ Because worktree isolation defaults to `delegate`, `defaultWorkdir` should norma
 
 Chat-launched sessions route updates back to their originating chat thread. For agent-launched tool sessions without an origin route, configure `fallbackChannel` or `agentChannels` in the reference guide.
 
-This release targets the OpenClaw SDK package `openclaw@2026.5.5`, while keeping the plugin peer floor at `>=2026.4.21`.
+This release targets the OpenClaw SDK package `openclaw@2026.5.6`, while keeping the plugin peer floor at `>=2026.4.21`.
 
 If you use Codex, make sure the local `codex` command or `OPENCLAW_CODEX_APP_SERVER_COMMAND` override is available and authenticated. When Codex auth is inconsistent, this is the recommended `~/.codex/config.toml` setting: