@action-llama/action-llama 0.1.1

package/README.md

@@ -0,0 +1,448 @@
+<p align="center">
+  <img src="action_llama.jpg" alt="Action Llama" />
+</p>
+
+# Action Llama
+
+It's like a Lambda that runs an agent. Triggered either by cron or webhooks.  BYOM- bring your own model.
+
+It's a very simple wrapper around whatever your favourite flavour of LLM is:
+
+1. Either a webhook or cron wakes up the agent
+2. The agent runs according the instructions in AGENTS.md you define
+3. The agent shuts down
+
+Allows you to create:
+
+- A developer agent that watches for new Github issues and reacts
+- A reviewer agent that watches for new Github Pull Requests and reviews them then merges if all ok
+
+Have as many agents as you like. Customize the behaviour as you wish. The system is MIT licensed and fully extensible.
+
+Built on [pi.dev](https://github.com/badlogic/pi-mono) as the agent harness.
+
+## Built-in Agents
+
+The project includes a few default agents to get you started
+
+| Agent | Trigger | Action |
+|-------|---------|--------|
+| **Developer** | Webhook: issue labeled; or poll for labeled issues | Checks out a worktree, implements the fix/feature, runs tests, opens a PR |
+| **PR Reviewer** | Webhook: PR opened/updated; or poll for open PRs | Reviews code for correctness, style, security; approves+merges or requests changes |
+| **DevOps** | Poll for CI failures/Sentry errors | Creates Github issues describing problem and potential fix |
+
+## Prerequisites
+
+- Node.js >= 20
+- Git
+- Docker
+- A GitHub Personal Access Token with `repo` and `workflow` scopes
+- Anthropic auth — one of:
+  - Existing pi auth (`pi /login`) or Claude Code auth (`claude setup-token`)
+  - An Anthropic API key (`sk-ant-api...`)
+  - An OAuth token (`sk-ant-oat...`)
+- (Optional) A Sentry auth token
+- (Optional) A GitHub webhook secret (for webhook-triggered agents)
+
+## Install
+
+```bash
+npx @action-llama/action-llama init my-project
+cd my-project
+npm install
+```
+
+This creates a new project directory with a `package.json` (including `@action-llama/action-llama` as a dependency), agent configs, and credentials.
+
+## Quick start
+
+```bash
+# 1. Initialize a project (interactive setup)
+npx @action-llama/action-llama init my-project
+cd my-project
+npm install
+
+# 2. Start the agents
+npx al start
+```
+
+## Architecture
+
+Action Llama separates three concerns:
+
+| Directory | Purpose | Created by |
+|-----------|---------|------------|
+| `~/.al-credentials/` | Secrets (shared across projects) | `al init` |
+| `./<project-name>/` | Per-project config, agent instructions, scratch space | `al init` |
+
+Agents run with credentials injected directly — `GITHUB_TOKEN`, `SENTRY_AUTH_TOKEN`, etc. are set as environment variables, and credential files are available at `/credentials/` in Docker mode. Agents use standard tools (`gh` CLI, `git`, `curl`) to interact with external services.
+
+## Setup walkthrough
+
+The setup CLI (`al init <name>`) walks through three steps:
+
+### Step 1: Credentials
+- Paste your GitHub PAT
+- (Optional) Paste your Sentry auth token
+- Choose your Anthropic authentication method:
+  - **Use existing pi auth** — if you already ran `pi /login` or `claude setup-token`
+  - **Enter an API key** — a standard `sk-ant-api...` key, validated against the API
+  - **Enter an OAuth token** — a `sk-ant-oat...` token from `claude setup-token`, format-checked
+
+### Step 2: LLM defaults
+- Select a model (default: `claude-sonnet-4-20250514`)
+- Select thinking level (default: `medium`)
+
+### Step 3: Agents
+- Select which built-in agents to enable (dev, reviewer, devops) and/or add custom agents
+- For each agent, configure:
+  - **Name** (allows multiple instances, e.g., `dev-frontend`, `dev-backend`)
+  - **Repos** to monitor
+  - Type-specific options: trigger label / assignee (dev), Sentry org / projects (devops)
+  - **Webhooks** — listen for GitHub events (default webhook filter per agent type)
+  - **Schedule** — poll on a cron interval (optional if webhooks are enabled)
+- If any agent uses webhooks, you'll be asked for a **GitHub webhook secret** (used to verify `x-hub-signature-256` on incoming payloads)
+
+## Running
+
+```bash
+# Start all agents (host mode, default)
+npx al start
+
+# Check agent status
+npx al status
+```
+
+The scheduler runs as a single Node.js process. Agents wake on incoming webhooks or their cron schedule (or both), do their work (or log `[SILENT]` if there's nothing to do), then wait for the next trigger. If any agent uses webhooks, a broker server starts automatically to receive `POST /webhooks/github` requests. Press `Ctrl+C` for graceful shutdown.
+
+### Docker mode (opt-in)
+
+When `docker.enabled` is set in `config.json`, agents run in isolated Docker containers with credentials mounted read-only at `/credentials/`. Containers have internet access and use standard tools (`gh`, `git`, `curl`) directly. A broker server provides a shutdown endpoint for the anti-exfiltration kill switch.
+
+```bash
+# Requires Docker installed and running
+# Enable in config.json: "docker": { "enabled": true }
+npx al start
+```
+
+See the [Docker mode](#docker-mode-1) section below for details.
+
+### Logs
+
+Structured JSON logs are written to daily files at `<project>/.al/logs/<agent>-<YYYY-MM-DD>.log` and also printed to stdout. In Docker mode, container logs are forwarded through the same pino logger — tool events and errors all appear in the standard log files.
+
+## CLI commands
+
+| Command | Description |
+|---------|-------------|
+| `npx @action-llama/action-llama init <name>` | Interactive setup, creates project dir + credentials |
+| `al start` | Start scheduler (cron + webhooks) |
+| `al status` | Show agent status |
+| `al logs` | View agent logs |
+| `al agent add` | Add a new agent to an existing project |
+
+## Using the developer agent
+
+1. Create an issue in one of your monitored repos
+2. Add the trigger label (default: `agent`) and assign it to the configured user
+3. The developer agent wakes (immediately via webhook, or on the next poll) and will:
+   - Find the issue via `gh issue list`
+   - Clone the repo and create a branch
+   - Read `AGENTS.md`/`CLAUDE.md` for project conventions
+   - Implement the changes, run tests
+   - Push and open a PR via `gh pr create`
+
+## Using the PR reviewer agent
+
+The reviewer automatically picks up open PRs on each poll. It:
+- Gets the diff via `gh pr diff`, checks CI status
+- Reviews for correctness, style, tests, and security
+- Approves and squash-merges clean PRs with green CI
+- Requests changes with specific feedback on problematic PRs
+- Skips PRs it has already reviewed at the same commit SHA
+
+## Using the DevOps agent
+
+The DevOps agent monitors for failures:
+- **GitHub Actions**: finds failed workflow runs via `gh run list`
+- **Sentry**: finds new unresolved error groups via `curl` to the Sentry API (if configured)
+
+For each new error, it creates a GitHub issue with the error details and a link to the source. Errors are deduplicated by fingerprint so the same failure is never filed twice.
+
+## Project structure
+
+```
+~/.al-credentials/                  # Secrets (shared across projects)
+  github-token
+  sentry-token                      # (optional)
+  anthropic-key                     # (optional, if not using pi_auth)
+  github-webhook-secret             # (optional, for webhook-triggered agents)
+
+./<project-name>/                   # Per-project (created by npx @action-llama/action-llama init)
+  package.json                      # Includes @action-llama/action-llama as a dependency
+  config.json                       # Global config: docker, broker, webhooks (no secrets)
+  dev/                              # Agent directory (CWD for agent sessions)
+    config.json                     # Agent config: repos, webhooks, schedule, prompt, etc.
+    AGENTS.md                       # Instructions (written during init, edit to customize)
+  dev-backend/                      # Multiple instances are supported
+    config.json
+    AGENTS.md
+  reviewer/
+    config.json
+    AGENTS.md
+  devops/
+    config.json
+    AGENTS.md
+  node_modules/                     # Dependencies (after npm install)
+  .workspace/                       # Git clones and worktrees (gitignored)
+  .al/
+    state/{dev,dev-backend,...}/     # Dedup/tracking state per agent
+    logs/                           # Structured logs
+```
+
+## Security
+
+### Host mode
+
+Secrets are isolated from agent context:
+
+1. Credential files live in `~/.al-credentials/` (mode 600, directory mode 700)
+2. The runner injects credentials as environment variables (`GITHUB_TOKEN`, etc.) — agents never see raw credential files
+3. Agents use standard tools (`gh` CLI, `git`, `curl`) which read credentials from env vars
+4. Agents have no extensions loaded (`noExtensions: true`) — only bash, read, edit, write tools for working in worktrees
+5. Anti-exfiltration policy is injected into agent prompts — agents are instructed to never output credentials in logs, comments, or PRs
+
+### Docker mode
+
+Docker mode adds stronger isolation:
+
+1. **Credentials mounted read-only** — credential files are symlinked into a temp staging dir and mounted at `/credentials/` (read-only)
+2. **Minimal privileges** — `--read-only` root FS, `--cap-drop ALL`, `--security-opt no-new-privileges`, non-root user, PID/memory/CPU limits
+3. **Kill switch** — each container gets a unique shutdown secret; if exfiltration is detected, the agent calls `POST /shutdown` to immediately kill the container
+4. **Tmpfs workspace** — all writable space is tmpfs (`/workspace`, `/tmp`), nothing persists after container exit
+5. **Standard tooling** — containers include `gh` CLI, `git`, and `curl`; no custom proxy or command routing
+
+## Configuration
+
+Config is split between a global `config.json` and per-agent `config.json` files.
+
+**`<project>/config.json`** — global settings (Docker, broker, webhooks):
+
+```json
+{
+  "docker": { "enabled": false },
+  "broker": { "port": 8080 },
+  "webhooks": { "githubSecretCredential": "github-webhook-secret" }
+}
+```
+
+**`<project>/<agent>/config.json`** — per-agent (includes model):
+
+```json
+{
+  "credentials": ["github-token", "anthropic-key"],
+  "model": { "provider": "anthropic", "model": "claude-sonnet-4-20250514", "thinkingLevel": "medium", "authType": "pi_auth" },
+  "prompt": "An issue was just assigned to you. Implement the changes described in the issue.",
+  "repos": ["acme/frontend"],
+  "triggerLabel": "agent",
+  "assignee": "bot-user",
+  "webhooks": {
+    "filters": [{
+      "source": "github",
+      "repos": ["acme/frontend"],
+      "events": ["issues"],
+      "actions": ["labeled"],
+      "labels": ["agent"],
+      "assignee": "bot-user"
+    }]
+  }
+}
+```
+
+Agents can use webhooks, a cron schedule, or both. Add a `"schedule"` field (e.g., `"*/5 * * * *"`) for polling. Each agent carries its own model config, so you can run different agents on different models (e.g., Opus for dev, Haiku for devops).
+
+Edit these files directly to change triggers, add repos, or switch models. Re-run `al init` for a guided reconfiguration.
+
+### Docker config options
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `docker.enabled` | `false` | Enable Docker container mode |
+| `docker.image` | `"al-agent:latest"` | Docker image for agent containers |
+| `docker.memory` | `"4g"` | Memory limit per container |
+| `docker.cpus` | `2` | CPU limit per container |
+| `docker.timeout` | `3600` | Max container runtime in seconds |
+| `broker.port` | `8080` | Broker server listen port |
+| `webhooks.githubSecretCredential` | `"github-webhook-secret"` | Credential name for the GitHub webhook HMAC secret |
+
+## Webhooks
+
+Agents can be triggered by GitHub webhooks for real-time responses instead of (or in addition to) polling. The setup wizard configures default webhook filters per agent type:
+
+| Agent | Default filter |
+|-------|---------------|
+| **Developer** | `issues` event, `labeled` action, matching trigger label + assignee |
+| **PR Reviewer** | `pull_request` event, `opened` / `synchronize` actions |
+| **DevOps** | `workflow_run` event, `completed` action |
+
+### Setting up the GitHub webhook
+
+1. Run `al init` and enable webhooks for your agents — the wizard will ask for a webhook secret
+2. In your GitHub repo (or org) settings, add a webhook:
+   - **Payload URL**: `http://<your-host>:8080/webhooks/github`
+   - **Content type**: `application/json`
+   - **Secret**: the same secret you entered during `al init`
+   - **Events**: select the events your agents listen for (or "Send me everything")
+3. Start AL — the broker server listens for incoming webhooks automatically
+
+Webhook payloads are validated using HMAC-SHA256 (`x-hub-signature-256`). If the signature doesn't match, the request is rejected with 401.
+
+### Webhook filter options
+
+Filters are configured in `<agent>/config.json` under `webhooks.filters`. Each filter can match on:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `source` | `"github"` | Required — the webhook source |
+| `repos` | `string[]` | Only match events from these repos |
+| `events` | `string[]` | GitHub event types (`issues`, `pull_request`, `push`, etc.) |
+| `actions` | `string[]` | Event actions (`opened`, `labeled`, `synchronize`, etc.) |
+| `labels` | `string[]` | Match if the issue/PR has any of these labels |
+| `assignee` | `string` | Match if assigned to this user |
+| `author` | `string` | Match if authored by this user |
+| `branches` | `string[]` | Match if targeting one of these branches |
+
+All specified fields must match (AND logic). Omitted fields are not checked.
+
+## Customizing agent behavior
+
+Each agent has an `AGENTS.md` file written during `al init`. Edit `<project>/<agent>/AGENTS.md` to customize agent behavior — changes take effect on the next run.
+
+Agent config values (repos, trigger label, etc.) are automatically injected into the prompt as an `<agent-config>` block, so AGENTS.md can reference them without hardcoding. When triggered by a webhook, a `<webhook-trigger>` block is also injected with event details (issue title, PR number, labels, etc.).
+
+## Testing
+
+Tests use [Vitest](https://vitest.dev/) with globals enabled and V8 coverage.
+
+```bash
+# Run all tests once
+npm test
+
+# Watch mode (re-runs on file changes)
+npm run test:watch
+
+# Run with coverage report
+npm run test:coverage
+```
+
+Test files live in `test/` mirroring the `src/` layout:
+
+```
+test/
+  helpers.ts                        # Shared test utilities
+  shared/                           # Unit tests for shared modules
+  agents/                           # Agent runner, prompt builder, default agent tests
+  setup/                            # Setup validators + scaffolding tests
+  scheduler/                        # Scheduler, webhook integration tests
+  webhooks/                         # Webhook registry + GitHub provider tests
+```
+
+Coverage is collected for all `src/**/*.ts` files, excluding entry points (`cli/main.ts`), interactive prompts (`setup/prompts.ts`), and pure type definitions (`scheduler/types.ts`).
+
+## Docker mode
+
+Docker mode runs each agent session in an isolated container. Credentials are mounted read-only at `/credentials/`, and the container has internet access to use standard tools directly.
+
+```
+HOST                                         DOCKER (al-net)
+┌──────────────────────────────────┐         ┌──────────────────────────────┐
+│  Scheduler                      │         │  Agent Container (per run)   │
+│    generates shutdown secret    │         │    pi-coding-agent session   │
+│    stages credentials           │         │    coding tools (bash,r/w)   │
+│    launches container           │         │    gh, git, curl             │
+│    waits for exit               │         │                              │
+│                                 │         │  /credentials/ (read-only)   │
+│  Broker (in-process)            │◄────────│    anthropic-key             │
+│    ● Health endpoint            │  HTTP   │    github-token              │
+│    ● Shutdown kill switch       │         │    sentry-token (optional)   │
+│    ● Webhook receiver           │         │                              │
+│                                 │         │  /workspace/ (tmpfs)         │
+│  Credentials (~/.al-creds/)     │         │  Internet access: yes        │
+│  Workspace (project/.workspace/)│         └──────────────────────────────┘
+└──────────────────────────────────┘
+```
+
+### Prerequisites
+
+- Docker installed and running
+- Set `"docker": { "enabled": true }` in `config.json`
+
+The agent Docker image is built automatically on first run from `docker/Dockerfile`.
+
+## Publishing to npm
+
+The package is configured for standard npm publishing under the `@action-llama` scope.
+
+### Prerequisites
+
+1. An [npm account](https://www.npmjs.com/signup) with access to the `@action-llama` org
+2. Login to npm:
+   ```bash
+   npm login
+   ```
+
+### Version and publish
+
+Use `npm version` to bump the version (updates `package.json`, creates a git tag, and pushes):
+
+```bash
+# Patch release (0.1.0 → 0.1.1)
+npm version patch
+
+# Minor release (0.1.0 → 0.2.0)
+npm version minor
+
+# Major release (0.1.0 → 1.0.0)
+npm version major
+```
+
+Then publish:
+
+```bash
+npm publish --access public
+```
+
+The `prepublishOnly` script automatically runs the build and tests before publishing. If either fails, the publish is aborted.
+
+### First-time publish
+
+For the very first publish of a scoped package:
+
+```bash
+npm publish --access public
+```
+
+The `--access public` flag is required for scoped packages on the first publish (subsequent publishes remember the setting).
+
+### What gets published
+
+Only these files are included in the npm tarball (controlled by the `files` field in `package.json`):
+
+- `dist/` — compiled JavaScript, source maps, and type declarations
+- `docker/` — Dockerfile for container mode
+- `README.md`
+- `LICENSE`
+- `package.json` (always included by npm)
+
+### Build scripts
+
+| Script | Description |
+|--------|-------------|
+| `npm run build` | Compile TypeScript to `dist/` and copy agent definition assets |
+| `npm run clean` | Remove the `dist/` directory |
+| `npm test` | Run all tests |
+| `npm version <patch\|minor\|major>` | Bump version, tag, and push |
+
+## License
+
+MIT

package/dist/agents/container-entry.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=container-entry.d.ts.map

package/dist/agents/container-entry.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"container-entry.d.ts","sourceRoot":"","sources":["../../src/agents/container-entry.ts"],"names":[],"mappings":""}

package/dist/agents/container-entry.js

@@ -0,0 +1,173 @@
+import { readFileSync, existsSync } from "fs";
+import { getModel } from "@mariozechner/pi-ai";
+import { AuthStorage, createAgentSession, DefaultResourceLoader, SessionManager, SettingsManager, createCodingTools, } from "@mariozechner/pi-coding-agent";
+// Structured log line — written to stdout, parsed by ContainerAgentRunner on the host
+function emitLog(level, msg, data) {
+    console.log(JSON.stringify({ _log: true, level, msg, ...data, ts: Date.now() }));
+}
+function readCredential(name) {
+    const path = `/credentials/${name}`;
+    if (!existsSync(path))
+        return undefined;
+    return readFileSync(path, "utf-8").trim();
+}
+async function main() {
+    // Switch CWD to /workspace so child processes (git, bash, etc.) default to it.
+    // Node must resolve from /app (WORKDIR at build time), so we chdir after startup.
+    process.chdir("/workspace");
+    const brokerUrl = process.env.BROKER_URL;
+    const shutdownSecret = process.env.SHUTDOWN_SECRET;
+    // Parse agent config from env var
+    const agentConfigStr = process.env.AGENT_CONFIG;
+    if (!agentConfigStr) {
+        emitLog("error", "missing AGENT_CONFIG env var");
+        process.exit(1);
+    }
+    const parsed = JSON.parse(agentConfigStr);
+    const agentsMd = parsed._agentsMd;
+    delete parsed._agentsMd;
+    const agentConfig = parsed;
+    const modelId = agentConfig.model.model;
+    const modelThinking = agentConfig.model.thinkingLevel;
+    emitLog("info", "container starting", { agentName: agentConfig.name, modelId, brokerUrl });
+    // Read Anthropic API key from credentials volume
+    const anthropicKey = readCredential("anthropic-key");
+    if (!anthropicKey) {
+        emitLog("error", "missing anthropic-key credential");
+        process.exit(1);
+    }
+    // Set GitHub and Sentry tokens in environment if available
+    const githubToken = readCredential("github-token");
+    if (githubToken) {
+        process.env.GITHUB_TOKEN = githubToken;
+        process.env.GH_TOKEN = githubToken;
+    }
+    const sentryToken = readCredential("sentry-token");
+    if (sentryToken) {
+        process.env.SENTRY_AUTH_TOKEN = sentryToken;
+    }
+    const cwd = "/workspace";
+    const model = getModel("anthropic", modelId);
+    const authStorage = AuthStorage.create();
+    authStorage.setRuntimeApiKey("anthropic", anthropicKey);
+    // AGENTS.md content is passed via the serialized config from the host
+    const agentsContent = agentsMd || `# ${agentConfig.name} Agent\n\nCustom agent.\n`;
+    const agentsFile = "/tmp/AGENTS.md";
+    const resourceLoader = new DefaultResourceLoader({
+        noExtensions: true,
+        agentsFilesOverride: () => ({
+            agentsFiles: [
+                { path: agentsFile, content: agentsContent },
+            ],
+        }),
+    });
+    await resourceLoader.reload();
+    const settingsManager = SettingsManager.inMemory({
+        compaction: { enabled: true },
+        retry: { enabled: true, maxRetries: 2 },
+    });
+    emitLog("info", "creating agent session", { model: modelId, thinking: modelThinking });
+    const { session } = await createAgentSession({
+        cwd,
+        model,
+        thinkingLevel: modelThinking,
+        authStorage,
+        resourceLoader,
+        tools: createCodingTools(cwd),
+        sessionManager: SessionManager.inMemory(),
+        settingsManager,
+    });
+    emitLog("info", "session created, sending prompt");
+    // Mirror the host-mode AgentRunner's session event logging
+    const pendingCmds = new Map();
+    let outputText = "";
+    let eventCount = 0;
+    session.subscribe((event) => {
+        eventCount++;
+        // Log all event types for debugging
+        if (event.type !== "message_update") {
+            const extra = { type: event.type, eventCount };
+            // Dump message events to see what the SDK is doing
+            if (event.type === "message_start" || event.type === "message_end") {
+                extra.role = event.role || event.message?.role;
+                extra.content = JSON.stringify(event.content || event.message?.content || []).slice(0, 500);
+                extra.stopReason = event.stopReason || event.stop_reason;
+            }
+            if (event.type === "turn_end") {
+                extra.turnResult = JSON.stringify(event).slice(0, 500);
+            }
+            emitLog("debug", "event", extra);
+        }
+        if (event.type === "error") {
+            emitLog("error", "session error", { error: String(event.error || event.message || JSON.stringify(event)) });
+        }
+        if (event.type === "message_update" && event.assistantMessageEvent?.type === "text_delta") {
+            outputText += event.assistantMessageEvent.delta;
+        }
+        if (event.type === "tool_execution_start") {
+            const cmd = String(event.args?.command || "");
+            if (event.toolName === "bash") {
+                pendingCmds.set(event.toolCallId, cmd);
+                emitLog("info", "bash", { cmd: cmd.slice(0, 200) });
+            }
+            else {
+                emitLog("debug", "tool start", { tool: event.toolName });
+            }
+        }
+        if (event.type === "tool_execution_end") {
+            const resultStr = typeof event.result === "string"
+                ? event.result
+                : JSON.stringify(event.result);
+            pendingCmds.delete(event.toolCallId);
+            if (event.isError) {
+                emitLog("error", "tool error", { tool: event.toolName, result: resultStr.slice(0, 1000) });
+            }
+            else {
+                emitLog("debug", "tool done", { tool: event.toolName, resultLength: resultStr.length });
+            }
+        }
+    });
+    // Prompt is pre-built by the scheduler and passed via PROMPT env var
+    const fullPrompt = process.env.PROMPT;
+    if (!fullPrompt) {
+        emitLog("error", "missing PROMPT env var");
+        process.exit(1);
+    }
+    // Retry on rate limit errors with exponential backoff
+    const MAX_PROMPT_RETRIES = 5;
+    const DEFAULT_BACKOFF_MS = 30_000;
+    const MAX_BACKOFF_MS = 300_000;
+    let result;
+    for (let attempt = 0; attempt <= MAX_PROMPT_RETRIES; attempt++) {
+        try {
+            result = await session.prompt(fullPrompt);
+            break;
+        }
+        catch (promptErr) {
+            const msg = String(promptErr?.message || promptErr || "");
+            const isRateLimit = msg.includes("rate_limit") || msg.includes("429") || msg.includes("529") || msg.includes("overloaded");
+            if (!isRateLimit || attempt === MAX_PROMPT_RETRIES) {
+                throw promptErr;
+            }
+            const delayMs = Math.min(DEFAULT_BACKOFF_MS * Math.pow(2, attempt), MAX_BACKOFF_MS);
+            emitLog("warn", "rate limited, retrying prompt", { attempt: attempt + 1, delayMs });
+            await new Promise((r) => setTimeout(r, delayMs));
+        }
+    }
+    emitLog("info", "prompt returned", { eventCount, resultType: typeof result, resultKeys: result ? Object.keys(result) : [] });
+    if (outputText.includes("[SILENT]")) {
+        emitLog("info", "no work to do");
+        console.log("[SILENT]");
+    }
+    else {
+        emitLog("info", "run completed", { outputLength: outputText.length });
+        console.log(outputText.slice(0, 2000));
+    }
+    session.dispose();
+    process.exit(0);
+}
+main().catch((err) => {
+    emitLog("error", "container entry error", { error: err.message, stack: err.stack?.split("\n").slice(0, 5).join("\n") });
+    process.exit(1);
+});
+//# sourceMappingURL=container-entry.js.map

package/dist/agents/container-entry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"container-entry.js","sourceRoot":"","sources":["../../src/agents/container-entry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,QAAQ,EAAE,MAAM,qBAAqB,CAAC;AAC/C,OAAO,EACL,WAAW,EACX,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,eAAe,EACf,iBAAiB,GAClB,MAAM,+BAA+B,CAAC;AAGvC,sFAAsF;AACtF,SAAS,OAAO,CAAC,KAAa,EAAE,GAAW,EAAE,IAA0B;IACrE,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,EAAE,GAAG,IAAI,EAAE,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,CAAC;AACnF,CAAC;AAED,SAAS,cAAc,CAAC,IAAY;IAClC,MAAM,IAAI,GAAG,gBAAgB,IAAI,EAAE,CAAC;IACpC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACxC,OAAO,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;AAC5C,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,+EAA+E;IAC/E,kFAAkF;IAClF,OAAO,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;IAE5B,MAAM,SAAS,GAAG,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC;IACzC,MAAM,cAAc,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC;IAEnD,kCAAkC;IAClC,MAAM,cAAc,GAAG,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC;IAChD,IAAI,CAAC,cAAc,EAAE,CAAC;QACpB,OAAO,CAAC,OAAO,EAAE,8BAA8B,CAAC,CAAC;QACjD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IACD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC;IAC1C,MAAM,QAAQ,GAAW,MAAM,CAAC,SAAS,CAAC;IAC1C,OAAO,MAAM,CAAC,SAAS,CAAC;IACxB,MAAM,WAAW,GAAgB,MAAM,CAAC;IACxC,MAAM,OAAO,GAAG,WAAW,CAAC,KAAK,CAAC,KAAK,CAAC;IACxC,MAAM,aAAa,GAAG,WAAW,CAAC,KAAK,CAAC,aAAa,CAAC;IAEtD,OAAO,CAAC,MAAM,EAAE,oBAAoB,EAAE,EAAE,SAAS,EAAE,WAAW,CAAC,IAAI,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IAE3F,iDAAiD;IACjD,MAAM,YAAY,GAAG,cAAc,CAAC,eAAe,CAAC,CAAC;IACrD,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,OAAO,CAAC,OAAO,EAAE,kCAAkC,CAAC,CAAC;QACrD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,2DAA2D;IAC3D,MAAM,WAAW,GAAG,cAAc,CAAC,cAAc,CAAC,CAAC;IACnD,IAAI,WAAW,EAAE,CAAC;QAChB,OAAO,CAAC,GAAG,CAAC,YAAY,GAAG,WAAW,CAAC;QACvC,OAAO,CAAC,GAAG,CAAC,QAAQ,GAAG,WAAW,CAAC;IACrC,CAAC;IAED,MAAM,WAAW,GAAG,cAAc,CAAC,cAAc,CAAC,CAAC;IACnD,IAAI,WAAW,EAAE,CAAC;QAChB,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,WAAW,CAAC;IAC9C,CAAC;IAED,MAAM,GAAG,GAAG,YAAY,CAAC;IAEzB,MAAM,KAAK,GAAG,QAAQ,CAAC,WAAW,EAAE,OAAc,CAAC,CAAC;IAEpD,MAAM,WAAW,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC;IACzC,WAAW,CAAC,gBAAgB,CAAC,WAAW,EAAE,YAAY,CAAC,CAAC;IAExD,sEAAsE;IACtE,MAAM,aAAa,GAAG,QAAQ,IAAI,KAAK,WAAW,CAAC,IAAI,2BAA2B,CAAC;IAEnF,MAAM,UAAU,GAAG,gBAAgB,CAAC;IAEpC,MAAM,cAAc,GAAG,IAAI,qBAAqB,CAAC;QAC/C,YAAY,EAAE,IAAI;QAClB,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC;YAC1B,WAAW,EAAE;gBACX,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,aAAa,EAAE;aAC7C;SACF,CAAC;KACH,CAAC,CAAC;IACH,MAAM,cAAc,CAAC,MAAM,EAAE,CAAC;IAE9B,MAAM,eAAe,GAAG,eAAe,CAAC,QAAQ,CAAC;QAC/C,UAAU,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE;QAC7B,KAAK,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE;KACxC,CAAC,CAAC;IAEH,OAAO,CAAC,MAAM,EAAE,wBAAwB,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,aAAa,EAAE,CAAC,CAAC;IAEvF,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,kBAAkB,CAAC;QAC3C,GAAG;QACH,KAAK;QACL,aAAa,EAAE,aAAa;QAC5B,WAAW;QACX,cAAc;QACd,KAAK,EAAE,iBAAiB,CAAC,GAAG,CAAC;QAC7B,cAAc,EAAE,cAAc,CAAC,QAAQ,EAAE;QACzC,eAAe;KAChB,CAAC,CAAC;IAEH,OAAO,CAAC,MAAM,EAAE,iCAAiC,CAAC,CAAC;IAEnD,2DAA2D;IAC3D,MAAM,WAAW,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC9C,IAAI,UAAU,GAAG,EAAE,CAAC;IACpB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,OAAO,CAAC,SAAS,CAAC,CAAC,KAAK,EAAE,EAAE;QAC1B,UAAU,EAAE,CAAC;QACb,oCAAoC;QACpC,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;YACpC,MAAM,KAAK,GAAwB,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,CAAC;YACpE,mDAAmD;YACnD,IAAI,KAAK,CAAC,IAAI,KAAK,eAAe,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;gBACnE,KAAK,CAAC,IAAI,GAAI,KAAa,CAAC,IAAI,IAAK,KAAa,CAAC,OAAO,EAAE,IAAI,CAAC;gBACjE,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,SAAS,CAAE,KAAa,CAAC,OAAO,IAAK,KAAa,CAAC,OAAO,EAAE,OAAO,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;gBAC9G,KAAK,CAAC,UAAU,GAAI,KAAa,CAAC,UAAU,IAAK,KAAa,CAAC,WAAW,CAAC;YAC7E,CAAC;YACD,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;gBAC9B,KAAK,CAAC,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;YACzD,CAAC;YACD,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;QACnC,CAAC;QACD,IAAK,KAAa,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YACpC,OAAO,CAAC,OAAO,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,MAAM,CAAE,KAAa,CAAC,KAAK,IAAK,KAAa,CAAC,OAAO,IAAI,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;QAChI,CAAC;QACD,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,IAAI,KAAK,CAAC,qBAAqB,EAAE,IAAI,KAAK,YAAY,EAAE,CAAC;YAC1F,UAAU,IAAI,KAAK,CAAC,qBAAqB,CAAC,KAAK,CAAC;QAClD,CAAC;QACD,IAAI,KAAK,CAAC,IAAI,KAAK,sBAAsB,EAAE,CAAC;YAC1C,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,EAAE,OAAO,IAAI,EAAE,CAAC,CAAC;YAC9C,IAAI,KAAK,CAAC,QAAQ,KAAK,MAAM,EAAE,CAAC;gBAC9B,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;gBACvC,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;YACtD,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;YAC3D,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,IAAI,KAAK,oBAAoB,EAAE,CAAC;YACxC,MAAM,SAAS,GAAG,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ;gBAChD,CAAC,CAAC,KAAK,CAAC,MAAM;gBACd,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;YACjC,WAAW,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;YAErC,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;gBAClB,OAAO,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,QAAQ,EAAE,MAAM,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC7F,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,OAAO,EAAE,WAAW,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,QAAQ,EAAE,YAAY,EAAE,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC;YAC1F,CAAC;QACH,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,qEAAqE;IACrE,MAAM,UAAU,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC;IACtC,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,wBAAwB,CAAC,CAAC;QAC3C,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,sDAAsD;IACtD,MAAM,kBAAkB,GAAG,CAAC,CAAC;IAC7B,MAAM,kBAAkB,GAAG,MAAM,CAAC;IAClC,MAAM,cAAc,GAAG,OAAO,CAAC;IAE/B,IAAI,MAAW,CAAC;IAChB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,kBAAkB,EAAE,OAAO,EAAE,EAAE,CAAC;QAC/D,IAAI,CAAC;YACH,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;YAC1C,MAAM;QACR,CAAC;QAAC,OAAO,SAAc,EAAE,CAAC;YACxB,MAAM,GAAG,GAAG,MAAM,CAAC,SAAS,EAAE,OAAO,IAAI,SAAS,IAAI,EAAE,CAAC,CAAC;YAC1D,MAAM,WAAW,GAAG,GAAG,CAAC,QAAQ,CAAC,YAAY,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC;YAC3H,IAAI,CAAC,WAAW,IAAI,OAAO,KAAK,kBAAkB,EAAE,CAAC;gBACnD,MAAM,SAAS,CAAC;YAClB,CAAC;YACD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,kBAAkB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,cAAc,CAAC,CAAC;YACpF,OAAO,CAAC,MAAM,EAAE,+BAA+B,EAAE,EAAE,OAAO,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC;YACpF,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;QACnD,CAAC;IACH,CAAC;IAED,OAAO,CAAC,MAAM,EAAE,iBAAiB,EAAE,EAAE,UAAU,EAAE,UAAU,EAAE,OAAO,MAAM,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAE7H,IAAI,UAAU,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC;QACpC,OAAO,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;QACjC,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;IAC1B,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,MAAM,EAAE,eAAe,EAAE,EAAE,YAAY,EAAE,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;QACtE,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;IACzC,CAAC;IAED,OAAO,CAAC,OAAO,EAAE,CAAC;IAClB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACnB,OAAO,CAAC,OAAO,EAAE,uBAAuB,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,OAAO,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACxH,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}

package/dist/agents/container-runner.d.ts

@@ -0,0 +1,20 @@
+import type { GlobalConfig, AgentConfig } from "../shared/config.js";
+import type { Logger } from "../shared/logger.js";
+import type { StatusTracker } from "../tui/status-tracker.js";
+export declare class ContainerAgentRunner {
+    private _running;
+    private globalConfig;
+    private agentConfig;
+    private logger;
+    private registerContainer;
+    private brokerUrl;
+    private projectPath;
+    private statusTracker?;
+    constructor(globalConfig: GlobalConfig, agentConfig: AgentConfig, logger: Logger, registerContainer: (secret: string, containerName: string) => void, brokerUrl: string, projectPath: string, statusTracker?: StatusTracker);
+    get isRunning(): boolean;
+    private forwardLogLine;
+    private streamContainerLogs;
+    private waitForContainer;
+    run(prompt: string): Promise<void>;
+}
+//# sourceMappingURL=container-runner.d.ts.map

package/dist/agents/container-runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"container-runner.d.ts","sourceRoot":"","sources":["../../src/agents/container-runner.ts"],"names":[],"mappings":"AAMA,OAAO,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AACrE,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,qBAAqB,CAAC;AAMlD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAE9D,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,QAAQ,CAAS;IACzB,OAAO,CAAC,YAAY,CAAe;IACnC,OAAO,CAAC,WAAW,CAAc;IACjC,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,iBAAiB,CAAkD;IAC3E,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,aAAa,CAAC,CAAgB;gBAGpC,YAAY,EAAE,YAAY,EAC1B,WAAW,EAAE,WAAW,EACxB,MAAM,EAAE,MAAM,EACd,iBAAiB,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,KAAK,IAAI,EAClE,SAAS,EAAE,MAAM,EACjB,WAAW,EAAE,MAAM,EACnB,aAAa,CAAC,EAAE,aAAa;IAW/B,IAAI,SAAS,IAAI,OAAO,CAEvB;IAED,OAAO,CAAC,cAAc;IAyCtB,OAAO,CAAC,mBAAmB;IAkC3B,OAAO,CAAC,gBAAgB;IA6BlB,GAAG,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;CAyFzC"}