@lmctl-ai/lmctl 0.1.0

package/dist/config/ai_test_templates/example-test.md.template

@@ -0,0 +1,50 @@
+---
+name: example-test
+description: Replace with a one-line description of what this test verifies
+type: smoke
+tags: [example]
+last_run_at: never
+last_run_status: unknown
+last_run_id: 0
+---
+
+# Test: example
+
+Replace this content with your real test case. The structure below is what the QA Tester agent expects.
+
+## Setup
+
+Describe what the system state must be before running this test. Examples:
+- A project exists with a configured team
+- An image fixture is present at a known path
+- The serve daemon is running on $LMCTL_API_URL
+
+Use shell-style commands (the Tester will execute them) or describe in prose if manual steps are needed.
+
+## Action
+
+What the test actually does. Be specific about commands, inputs, and expected intermediate observations.
+
+```bash
+# Example action
+curl -X GET $LMCTL_API_URL/api/status \
+  -H "Authorization: Bearer $LMCTL_API_TOKEN"
+```
+
+## Expected
+
+What "pass" looks like:
+- HTTP 200 response
+- JSON body contains key `jobs` with `total >= 0`
+- Response time under 1 second
+
+The Interpreter agent reads this section + the Action's actual observations to emit STANCE: pass / fail / inconclusive.
+
+## Cleanup
+
+Idempotent cleanup steps. Should be safe to run even if Setup or Action failed.
+
+```bash
+# Example cleanup
+# (nothing to clean up for this read-only example)
+```

package/dist/config/ai_test_templates/index.md.template

@@ -0,0 +1,27 @@
+# AI Test Suite - index
+
+This directory holds **AI-driven manual tests** for this project. Each chapter (`*.md`) describes one test case that an LLM agent (the QA Tester) executes; an Interpreter agent judges pass/fail.
+
+**This is NOT for unit/integration tests** - those live in `tests/` and run via your test framework (e.g., `npm test`). `ai-test/` is for things that need LLM-quality judgment: vision quality, prompt-driven behavior, end-to-end scenarios that exercise the full stack.
+
+## Conventions
+
+- One test case per `.md` file
+- YAML frontmatter (agent-writable: `last_run_*` fields)
+- Markdown body sections: `## Setup`, `## Action`, `## Expected`, `## Cleanup`
+- Issues found by qa-suite reference the chapter via `ai_test_path` (e.g. `ai-test/test-image-upload-basic.md`)
+
+## Test cases
+
+See individual chapters in this directory. Use `grep -l 'type: smoke' ai-test/*.md` (etc.) to filter.
+
+## How to add a new test case
+
+1. Copy a chapter as a template: `cp ai-test/example-test.md ai-test/test-my-new-case.md`
+2. Edit frontmatter (name, description, type, tags)
+3. Edit body (Setup, Action, Expected, Cleanup)
+4. Commit alongside the code change that motivated the test
+
+## Pointer to lmctl's own ai-test/ (canonical example)
+
+See `<lmctl-repo>/ai-test/` for the test corpus used to verify lmctl itself.

package/dist/config/durable_memory_templates/index.md.template

@@ -0,0 +1,29 @@
+---
+name: index
+description: Index of all durable-memory chapters for this project
+type: project
+---
+
+# Durable-memory index
+
+This directory contains durable knowledge accumulated by agents working on this project.
+
+## How to use
+
+ON EVERY NEW TASK:
+1. Read this file first. Scan the chapter list below.
+2. Pull only the chapters relevant to your task. Don't load everything at once.
+3. If you learn something worth preserving (a convention, a decision, a gotcha), write it back:
+   create or update `durable-memory/<descriptive-name>.md` with YAML frontmatter
+   (name / description / type) and a body. The engine syncs the write automatically.
+
+## Canonical example
+
+If you're new to durable-memory, lmctl's own `durable-memory/` directory is the canonical reference.
+Read its `index.md` + `architecture-foundations.md` + `agent-onboarding-pattern.md` to see the pattern in production use.
+Path: `<lmctl-repo>/durable-memory/`
+
+## Chapters
+
+<!-- Agents: add a one-line entry here when you create a new chapter. -->
+<!-- Format: - **<name>** (`<filename>.md`) — <one-sentence description> -->

package/dist/config/durable_memory_templates/skills_general.md.template

@@ -0,0 +1,42 @@
+---
+name: Skills — general agent discipline
+description: Core agent habits for durable-memory, STANCE format, and bypass-test discipline
+type: feedback
+tags: [skills, discipline]
+---
+
+# General agent discipline
+
+## STANCE format
+
+Always end a response or turn with a STANCE line so the engine can parse your outcome:
+
+```
+STANCE: <outcome_key>
+```
+
+Common outcomes: `completed`, `needs_more_clarity_from_lead`, `blocked_with_question`, `no_writeback`.
+
+Use `STANCE: no_writeback` after a write-back turn only when there is genuinely nothing new to record.
+
+## Durable-memory read-before-write
+
+- Read `durable-memory/index.md` at the start of every new task.
+- Pull only the chapters relevant to your task; don't load the full directory.
+- Before creating a new chapter, check whether an existing one covers the topic.
+
+## Write-back discipline
+
+Write back when you learn something durable:
+- A project convention that future agents must follow
+- An architectural decision and its rationale
+- A non-obvious constraint, integration gotcha, or environment quirk
+
+Do NOT write back:
+- Information derivable from reading the current code or git history
+- Ephemeral task state or in-progress notes
+- Anything already documented in CLAUDE.md or equivalent config files
+
+## Bypass-test discipline
+
+Tests that verify a behaviour must be "load-bearing": removing the code under test must make the test fail. If a test would still pass after the fix is removed, it is not a bypass test — rewrite it so it is.

package/dist/config/durable_memory_templates/skills_lmdebug.md.template

@@ -0,0 +1,31 @@
+---
+name: Skills — lmdebug
+description: When to use lmdebug, common invocations, and project-specific gotchas
+type: feedback
+tags: [skills, lmdebug]
+---
+
+# lmdebug skills
+
+<!-- Agents: fill in concrete examples from your first walk-through of this project. -->
+<!-- This stub gives structure; you supply the project-specific content.             -->
+
+## When to use lmdebug
+
+[TODO: describe the scenarios in this project where lmdebug is the right tool —
+e.g., inspecting a specific failure mode, tracing an unexpected code path, etc.]
+
+## Common invocations
+
+[TODO: list the lmdebug commands / flags most useful for this project, with a one-line
+description of what each does and when to reach for it.]
+
+```
+# Example placeholder — replace with real commands:
+# lmdebug attach --target <...> --breakpoint <...>
+```
+
+## Gotchas
+
+[TODO: note any project-specific pitfalls — environment requirements, symbol resolution
+issues, known interpreter quirks — discovered during real usage.]

package/dist/config/durable_memory_templates/skills_lmprobe.md.template

@@ -0,0 +1,31 @@
+---
+name: Skills — lmprobe
+description: When to use lmprobe, common invocations, and project-specific gotchas
+type: feedback
+tags: [skills, lmprobe]
+---
+
+# lmprobe skills
+
+<!-- Agents: fill in concrete examples from your first walk-through of this project. -->
+<!-- This stub gives structure; you supply the project-specific content.             -->
+
+## When to use lmprobe
+
+[TODO: describe the scenarios in this project where lmprobe is the right tool —
+e.g., reproducing a specific class of failure, validating a fix under load, etc.]
+
+## Common invocations
+
+[TODO: list the lmprobe commands / flags most useful for this project, with a one-line
+description of what each does and when to reach for it.]
+
+```
+# Example placeholder — replace with real commands:
+# lmprobe run --target <...> --scenario <...>
+```
+
+## Gotchas
+
+[TODO: note any project-specific pitfalls — environment requirements, known flaky
+scenarios, interpretation quirks — discovered during real usage.]

package/package.json

@@ -0,0 +1,42 @@
+{
+  "name": "@lmctl-ai/lmctl",
+  "version": "0.1.0",
+  "description": "lmctl — a durable multi-step agent workflow engine and CLI (public preview).",
+  "type": "module",
+  "bin": {
+    "lmctl": "bin/lmctl"
+  },
+  "files": [
+    "dist/cli/index.js",
+    "dist/cli/schema.sql",
+    "dist/cli/side_effect_classifier.json",
+    "dist/config/",
+    "workflows/",
+    "bin/",
+    "package.json",
+    "LICENSE",
+    "README.md"
+  ],
+  "engines": {
+    "node": ">=20"
+  },
+  "license": "SEE LICENSE IN LICENSE",
+  "publishConfig": {
+    "access": "public",
+    "tag": "staging"
+  },
+  "dependencies": {
+    "@aws-sdk/client-cognito-identity": "^3.1053.0",
+    "@aws-sdk/client-s3": "^3.1053.0",
+    "@aws-sdk/credential-provider-cognito-identity": "^3.972.36",
+    "@hono/node-server": "1.13.8",
+    "better-sqlite3": "11.5.0",
+    "docx": "^9.6.1",
+    "exceljs": "^4.4.0",
+    "hono": "4.6.20",
+    "jexl": "^2.3.0",
+    "marked": "^18.0.3",
+    "node-pty": "^1.1.0",
+    "zod": "4.4.3"
+  }
+}