lilflow 0.1.1 → 0.2.1

package/.claude-plugin/plugin.json

@@ -0,0 +1,7 @@
+{
+  "name": "lilflow",
+  "version": "0.2.1",
+  "description": "Drive lilflow workflows from inside a Claude Code session. Exposes the flow session-bridge CLI as a skill so the agent can request the next step, update workflow state, evaluate gates, and converse with the user.",
+  "homepage": "https://github.com/iVintik/lilflow",
+  "license": "MIT"
+}

package/README.md

@@ -1,112 +1,254 @@
+<!-- codeharness:readme -->
 # lilflow
 
-Repo-native workflow engine CLI. Step-level resumable YAML workflows with state stored under `.flow/` in your repo.
+Repo-native workflow engine CLI. Define multi-step workflows in YAML, execute them with step-level persistence, and resume from failure -- all stored under `.flow/` in your repo. No external services, no databases, no daemon.
 
-## Install
+## What it does
+
+lilflow turns YAML workflow definitions into executable, resumable pipelines that live inside your repository. It persists every step's start, completion, and failure as append-only JSONL events, so you can resume a failed run from exactly where it broke. It has first-class support for LLM coding agents (Claude Code and OpenCode) as step types, with session management across steps and two execution modes: classic (one agent process per step) and session (the entire workflow runs inside a single long-lived agent session for prompt-cache efficiency). It's for developers who want CI-like orchestration without leaving the repo.
+
+## Key features
+
+- **Step-level resumability** -- append-only JSONL event logs let you resume from the exact point of failure
+- **LLM agent steps** -- run Claude Code or OpenCode as workflow steps with session continuation and cost tracking
+- **Session mode** -- run the whole workflow inside one agent session; agent calls `flow session-bridge` CLI to advance steps. Keeps the prompt cache warm for the entire run
+- **Conversational agent mode** -- `agent.mode: conversational` lets the agent iterate with the user until the step is complete, then advance
+- **Structured JSON output** -- `output_file` + `output_format: json` on agent steps extracts and validates JSON from the agent's stdout, no parsing scripts required
+- **Parallel execution** -- mark steps with `parallel: true` or group them with `parallel_group` for concurrent batches
+- **For-each loops** -- expand a step across an array of items with `{{item}}` templating
+- **Quality gates** -- conditional checkpoints that halt or warn based on expressions
+- **Subflow composition** -- call child workflows with parameter passing
+- **Wait triggers** -- block on file existence or external signals with `flow signal`
+- **Hierarchical config** -- global, project, workflow-local, env vars, and CLI flags merge in order
+
+## Tech stack
+
+- **Runtime:** Node.js >= 22
+- **Language:** JavaScript (ES modules)
+- **Dependencies:** js-yaml (single runtime dependency)
+- **Test runner:** node:test (built-in)
+- **Coverage:** c8
+
+## Getting started
+
+### Install
 
 ```bash
 npm install -g lilflow
 ```
 
-Binary is exposed as both `lilflow` and `flow`.
-
-## Quick Start
+### Run the simplest possible example
 
 ```bash
-flow init        # scaffold workflow.yaml + .flow/ state dir
-flow run         # execute the workflow
-flow status      # inspect run state
-flow resume      # resume a failed run from the last completed step
+flow init && flow run
 ```
 
-## Parallel Steps
+Expected output:
+```
+Initialized .flow/ and workflow.yaml
+Running workflow 'hello-world'...
+[setup] echo "Hello from flow"
+Hello from flow
+Workflow completed.
+```
+
+### Next step
+
+See [Usage](#usage) for common workflows or [docs/index.md](./docs/index.md) for full docs.
 
-The runner supports contiguous parallel batches inside `workflow.yaml`.
+## Usage
+
+### Define and run a multi-step workflow
 
 ```yaml
-name: ci
+name: build-and-test
 steps:
+  - name: install
+    run: npm ci
   - name: lint
     run: npm run lint
-    parallel: true
+  - name: test
+    run: npm test
+```
+
+```bash
+flow run
+```
+
+### Run steps in parallel
 
+```yaml
+steps:
+  - name: lint
+    run: npm run lint
+    parallel: true
   - name: test
     run: npm test
     parallel: true
-
   - name: build
     run: npm run build
 ```
 
-Parallel step output is streamed in real time with step-name prefixes such as `[lint] ...`, and `flow status <run-id>` shows each started step with its own `running`, `completed`, or `failed` state.
+Parallel step output streams in real time with `[lint] ...` / `[test] ...` prefixes.
+
+### Resume a failed run
+
+```bash
+flow status <run-id>    # see what failed
+flow resume <run-id>    # pick up from the failed step
+```
 
-Parallel batch concurrency is controlled by the resolved `parallelism` config value. The default is `4`, and you can override it in `.flow/config.yaml`, `~/.flow/config.yaml`, or with `FLOW_PARALLELISM`.
+### Use LLM agents as steps
 
 ```yaml
-parallelism: 2
+steps:
+  - name: implement
+    agent:
+      provider: claude-code
+      prompt: "Implement the dashboard component"
+  - name: review
+    agent:
+      provider: claude-code
+      prompt: "Review the implementation"
+      session: continue
 ```
 
-## Retry Logic
+### Agent interaction modes
 
-Steps can retry transient failures with exponential backoff.
+Agent steps take an optional `mode` field:
+
+- `autonomous` (default) -- agent works alone, runs headless, reports when done
+- `conversational` -- agent works with the user iteratively in a TTY, advances only after user approval
 
 ```yaml
-name: deploy
 steps:
-  - name: flaky-check
-    run: npm run flaky-check
-    retry: 3
-    retry_delay: 5s
+  - name: build
+    agent:
+      provider: claude-code
+      prompt: "Build the feature described in spec.md"
+      mode: conversational
 ```
 
-`retry` is the number of retries after the initial attempt, so `retry: 3` allows up to 4 total attempts. `retry_delay` is the base delay and doubles on each retry. The runner prints numbered attempt labels such as `[attempt 2/4]`, shows the failure reason for each retryable attempt, and prints wait lines such as `Waiting 10s before retry 2/3...`.
+### Extract JSON output from an agent step
+
+```yaml
+steps:
+  - name: analyze
+    agent:
+      provider: claude-code
+      prompt: ./analyze.md
+      output_file: out/analysis.json
+      output_format: json
+```
 
-## For-Each Loops
+lilflow strips markdown fences, finds the first balanced JSON object/array in stdout, validates it parses, and writes the cleaned JSON to the file. The step fails with a structured error code if the output isn't valid JSON.
 
-Steps can expand an inline array into concrete iterations with `for_each`.
+### Run the entire workflow in one agent session
 
 ```yaml
-name: deploy
+name: iterative-dev
+mode: session
+session:
+  provider: claude-code       # or: opencode (via oh-my-opencode)
+  model: claude-opus-4-6
+  allow_tools: [Bash, Read, Write, Edit, Grep, Glob]
+  plugins: [lilflow]
 steps:
-  - name: deploy-target
-    run: echo "Deploying to {{item}} ({{item_number}}/{{item_count}})"
-    for_each: [dev, staging, prod]
+  - name: scaffold
+    run: mkdir -p src/components
+  - name: implement
+    agent:
+      provider: claude-code
+      prompt: "Build dashboard components"
+      mode: conversational
+  - name: test
+    run: npm test
 ```
 
-Each iteration becomes a concrete runtime step with a stable label such as `deploy-target [item-prod]`. The templating variables `{{item}}`, `{{item_index}}`, `{{item_number}}`, `{{item_count}}`, `{{item_first}}`, and `{{item_last}}` are available in string step fields. If one iteration fails, the remaining iterations still run before the workflow stops.
+`flow run` spawns one `claude` (or `opencode`) process, injects a workflow-aware system prompt, and exposes the `flow session-bridge` CLI through the bundled `lilflow` skill. The agent drives the workflow (`flow session-bridge next` -> execute -> `flow session-bridge update`) without per-step process spawns, keeping the prompt cache warm for the entire run.
 
-## Workflow Templates
+OpenCode sessions work through [oh-my-opencode](https://github.com/opensoft/oh-my-opencode)'s Claude Code compatibility layer -- the same bundled plugin loads unmodified.
 
-`flow init` can scaffold `workflow.yaml` from reusable local templates stored under `.flow/templates/<name>/workflow.yaml`.
+### Iterate with for-each
 
-```bash
-flow init --list-templates
-flow init --template ci-pipeline
+```yaml
+steps:
+  - name: deploy
+    run: ./deploy.sh {{item}}
+    for_each: [dev, staging, prod]
 ```
 
-Template files can mix prompt-backed placeholders and config-backed placeholders:
+### Wait for external input
 
 ```yaml
-name: {{template.workflow_name|ci-pipeline}}
-parallelism: {{config.parallelism}}
 steps:
-  - name: test
-    run: npm test
+  - name: await-approval
+    wait:
+      trigger: signal
+      timeout: 1h
+```
+
+Then from another terminal: `flow signal <run-id> await-approval --data '{"approved": true}'`
+
+## CLI
+
 ```
+flow init [--template <name>]
+flow config
+flow run [<workflow>]
+flow resume <run-id>
+flow set-step <run-id> <step-index>
+flow signal <run-id> <step-name> [--data '{}']
+flow status <run-id>
+flow list
+flow logs <run-id> [--step <step>]
+flow session-bridge <subcommand>     # agent-facing bridge for session-mode workflows
+```
+
+## Claude Code plugin
 
-- `{{template.name}}` prompts for a value
-- `{{template.name|default}}` prompts and uses the default when you press enter
-- `{{config.parallelism}}` reads from the resolved lilflow config
+This repo is also a Claude Code plugin. `.claude-plugin/plugin.json` is the marker file; `skills/` holds the `lilflow-workflow-driver` skill used by session mode. The plugin version is kept in lock-step with the npm package version via the `npm version` lifecycle hook, so a single `v{version}` tag releases both.
 
-Any directory matching `.flow/templates/<name>/workflow.yaml` is a usable custom template.
+Install as a plugin from a Claude Code session:
+```bash
+claude plugin install github:iVintik/lilflow
+```
+
+OpenCode users install the same plugin via [oh-my-opencode](https://github.com/opensoft/oh-my-opencode)'s Claude Code compatibility layer.
+
+## Project structure
+
+```
+lilflow/
+├── src/                    # Application source (ES modules)
+│   ├── cli.js              # Entry point, command router
+│   ├── run-workflow.js     # Core workflow engine
+│   ├── config.js           # Hierarchical config system
+│   ├── init-project.js     # Project scaffolding
+│   ├── session-bridge.js   # Agent-facing CLI for session mode
+│   ├── session-runner.js   # Single-process execution path
+│   ├── session-prompt.js   # System prompt generator
+│   └── agents/             # LLM agent provider adapters + output extraction
+├── .claude-plugin/         # Claude Code plugin manifest
+│   └── plugin.json
+├── skills/                 # Claude Code skills bundled with the plugin
+│   └── lilflow-workflow-driver/SKILL.md
+├── scripts/                # Release helpers (sync-plugin-version.js)
+├── tests/                  # Test suite (node:test)
+├── docs/                   # Generated documentation + requirements specs
+└── .github/workflows/      # CI/CD (lint + test + npm publish)
+```
 
-## Configuration
+## Documentation
 
-- Project config: `.flow/config.yaml`
-- Global config: `~/.flow/config.yaml`
-- Env var override: `FLOW_*`
-- Override order: defaults → global → project → env → flags
+- [Documentation Index](./docs/index.md)
+- [Architecture](./docs/architecture.md)
+- [Component Inventory](./docs/component-inventory.md)
+- [Source Tree Analysis](./docs/source-tree-analysis.md)
+- [Development Guide](./docs/development-guide.md)
+- [Session Mode Spec](./docs/requirements/session-mode.md)
 
 ## License
 
 MIT
+<!-- /codeharness:readme -->

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "lilflow",
-  "version": "0.1.1",
+  "version": "0.2.1",
   "description": "Repo-native workflow engine CLI",
   "type": "module",
   "bin": {
@@ -9,6 +9,8 @@
   },
   "files": [
     "src",
+    ".claude-plugin",
+    "skills",
     "README.md",
     "AGENTS.md"
   ],
@@ -33,7 +35,9 @@
   "scripts": {
     "test": "c8 --reporter=text --reporter=lcov --reporter=json-summary node --test",
     "coverage": "npm test",
-    "lint": "eslint src/"
+    "lint": "eslint src/",
+    "sync-plugin-version": "node scripts/sync-plugin-version.js",
+    "version": "npm run sync-plugin-version && git add .claude-plugin/plugin.json"
   },
   "engines": {
     "node": ">=22"

package/skills/lilflow-workflow-driver/SKILL.md

@@ -0,0 +1,110 @@
+---
+name: lilflow-workflow-driver
+description: Drive a lilflow session-mode workflow from inside this agent session (Claude Code natively, or OpenCode via the oh-my-opencode Claude Code compatibility layer). Use this skill whenever the FLOW_RUN_ID environment variable is set — it means lilflow has started a workflow and expects you to advance it via the `flow session-bridge` CLI. Call `flow session-bridge next` to get the next step, execute it, then call `flow session-bridge update <step> completed|failed` to advance. For conversational agent steps, work with the user iteratively before marking complete.
+---
+
+# lilflow Workflow Driver
+
+You are driving a **lilflow** workflow. Each step in the workflow must be fetched, executed, and reported back through the `flow session-bridge` CLI. Do not try to guess the workflow — always ask the bridge what's next.
+
+This skill works identically in Claude Code and in OpenCode (via [oh-my-opencode](https://github.com/opensoft/oh-my-opencode)'s Claude Code compatibility layer). The bridge CLI is the same in both environments.
+
+## When this skill activates
+
+You can tell this skill applies when:
+
+- The environment variable `FLOW_RUN_ID` is set (always check `echo $FLOW_RUN_ID` first).
+- The user's initial prompt references a workflow, a "run", or explicitly mentions lilflow.
+
+If `FLOW_RUN_ID` is not set, stop and tell the user — the workflow runner did not start this session.
+
+## Core loop
+
+Follow this loop until `flow session-bridge next` reports `workflow_status: "completed"`:
+
+1. **Get the next step:**
+   ```bash
+   flow session-bridge next
+   ```
+   Returns JSON: `{ step, steps, remaining, workflow_status }`. `step` is non-null for a single eligible step; `steps` is a list for a parallel batch.
+
+2. **Execute the step** based on its `type`:
+
+   - **`run`** — execute `step.command` via the `Bash` tool. Capture stdout/exit code.
+   - **`agent`** with `mode: autonomous` — execute `step.agent.prompt` yourself. You are the agent.
+   - **`agent`** with `mode: conversational` — see [Conversational steps](#conversational-steps) below.
+   - **`agent`** with a different provider (e.g. `opencode`) — mark the step `deferred`; the runner handles foreign providers.
+   - **`gate`** — call `flow session-bridge gate <name>`. It evaluates the condition and returns `{ passed, workflow_should_stop }`.
+   - **`interactive`** — call `flow session-bridge ask <prompt>` to get user input, then run the command with the input.
+   - **`wait`** — call `flow session-bridge wait <name>`. Blocks until the trigger fires.
+   - **`subflow`** — not yet supported in session mode. Mark the step `deferred`.
+
+3. **Report the result:**
+   ```bash
+   flow session-bridge update <step-name> completed --output "..." --exit-code 0
+   flow session-bridge update <step-name> failed --reason "..."
+   ```
+
+4. **On failure:** call `flow session-bridge update <name> failed` and stop the loop unless the user instructs otherwise.
+
+5. **Repeat** until `workflow_status === "completed"`.
+
+## Conversational steps
+
+When `step.agent.mode === "conversational"`:
+
+1. Present the step's prompt and your proposed approach to the user.
+2. Work iteratively — implement, show results, gather feedback, adjust.
+3. Do NOT call `flow session-bridge update` until the user has confirmed satisfaction OR you have completed the task to its written specification.
+4. Explicitly tell the user when you consider the step complete before advancing.
+5. If the user says "skip", "abort", or similar, call `flow session-bridge update <name> failed --reason "user aborted"`.
+
+Example:
+```
+User: (workflow starts)
+Agent: The workflow says: "Create React components for the dashboard.
+       Iterate with the user until they approve." Here's my plan: ...
+User: Add a sidebar too.
+Agent: [revises, implements, shows output]. Anything else?
+User: Looks good.
+Agent: Marking "implement" complete, advancing to the next step.
+  [runs: flow session-bridge update implement completed]
+```
+
+## Parallel batches
+
+If `flow session-bridge next` returns multiple `steps`, they are eligible in parallel. You may execute them sequentially, or use sub-agents (the `Agent` tool) to run them concurrently. Call `flow session-bridge update` once per step.
+
+## User-initiated questions
+
+If you need user input in the middle of an autonomous step (an ambiguity the workflow didn't cover), call `flow session-bridge ask "<question>"` — the bridge prompts the user and returns their response on stdout.
+
+## Context management
+
+After long workflows, your context may accumulate noise from completed steps. Call:
+```bash
+flow session-bridge compact
+```
+This returns a condensed summary of completed steps. Use the summary to anchor your memory and treat earlier tool results as droppable.
+
+## Reporting notes and decisions
+
+To record observations that don't change workflow state:
+```bash
+flow session-bridge log "observation or decision"
+```
+
+## Inspecting state
+
+To see the full workflow state (what's done, what's pending):
+```bash
+flow session-bridge status
+```
+
+## Rules
+
+1. **Always** call `flow session-bridge next` before executing anything. Do not infer the workflow.
+2. **Always** call `flow session-bridge update` after finishing a step, even on failure. The bridge is the source of truth.
+3. **Never** invent step names or pretend a step succeeded. The event log is authoritative.
+4. When a gate fails with `workflow_should_stop: true`, stop the loop and report the failure to the user.
+5. Respect `mode: conversational` — do not advance these steps without user interaction.

package/src/agents/index.js

@@ -5,6 +5,7 @@ import { runOpencode } from "./opencode.js";
 import { runClaudeCode } from "./claude-code.js";
 import { resolvePromptInput } from "./prompt.js";
 import { getSessionId, loadSessionStore, saveSessionId } from "./session-store.js";
+import { AgentOutputError, writeAgentOutput } from "./output-file.js";
 
 const PROVIDERS = {
   opencode: {
@@ -101,6 +102,18 @@ export async function executeAgentStep(options) {
   const templatedAppendSystemPrompt = agentSpec.append_system_prompt === undefined
     ? undefined
     : templateFn(agentSpec.append_system_prompt);
+
+  if (agentSpec.interactive !== undefined && typeof warn === "function") {
+    warn(
+      `Agent step '${step.name}' uses deprecated agent.interactive; prefer agent.mode: ${agentSpec.interactive === true ? "conversational" : "autonomous"}.`
+    );
+  }
+
+  // mode: conversational maps to TTY passthrough. Legacy agent.interactive
+  // is respected for back-compat until removal.
+  const resolvedInteractive = agentSpec.mode === "conversational"
+    || agentSpec.interactive === true;
+
   const result = await provider.run({
     bin: binary,
     prompt: templatedPrompt,
@@ -109,7 +122,7 @@ export async function executeAgentStep(options) {
     timeoutMs,
     cwd,
     env,
-    interactive: agentSpec.interactive === true,
+    interactive: resolvedInteractive,
     plugins: agentSpec.plugins ?? [],
     allowTools: agentSpec.allow_tools ?? [],
     appendSystemPrompt: templatedAppendSystemPrompt,
@@ -128,6 +141,33 @@ export async function executeAgentStep(options) {
     }
   }
 
+  if (agentSpec.output_file !== undefined && result.exitCode === 0) {
+    try {
+      const writeResult = await writeAgentOutput({
+        outputFile: agentSpec.output_file,
+        format: agentSpec.output_format ?? "text",
+        stdout: result.stdout,
+        cwd,
+        stepName: step.name
+      });
+
+      if (typeof warn === "function") {
+        warn(`Agent step '${step.name}' wrote ${writeResult.bytesWritten} bytes to ${agentSpec.output_file}.`);
+      }
+    } catch (error) {
+      if (error instanceof AgentOutputError) {
+        // Surface as step failure without corrupting the agent result payload.
+        return {
+          ...result,
+          exitCode: result.exitCode === 0 ? 1 : result.exitCode,
+          stderr: `${result.stderr ?? ""}\n${error.message}`.trim(),
+          outputError: { code: error.code, message: error.message }
+        };
+      }
+      throw error;
+    }
+  }
+
   return result;
 }