localclawd 2.0.9 → 2.2.1

package/README.md

@@ -22,7 +22,7 @@ The project focuses on four areas:
 - Compact-context controls for local models that degrade before their advertised context limit.
 - Multimodal passthrough for local models that support image and screenshot input.
 - **Lattice memory scoring** — memory files tagged with `tags:` frontmatter are ranked using Jaccard similarity and co-occurrence lattice math. Works offline as a fallback when a hosted side-query model is unavailable.
-- **`/keepgoing`** — autonomous task continuation loop. The model works through all outstanding steps without waiting for user input and re-queues itself after each response. Stops when the model emits `TASK COMPLETE:` or `NEEDS INPUT:`. Aliases: `/kg`, `/continue`.
+- **`/keepgoing`** — autonomous task continuation loop. After each round, a lightweight synthesis agent analyzes the full conversation and writes a precise directive for the next round — the model doesn't need to self-direct. Stops when the user presses Ctrl+C or sends `/stop`. Aliases: `/kg`, `/continue`.
 - **`/buddy`** — spawns a named ASCII animal companion for the session with a personality. Use `/buddy pet` to hear their thoughts on the current codebase.
 - **`/images`** — quick-start slash command that forwards into the project-local image pipeline setup flow, with ComfyUI-first defaults and helper scaffolding.
 - **`/image-pipeline`** — scaffolds and uses a project-local image generation workflow for game textures, sprites, and related art assets under `.localclawd/image-pipeline/`, then visually reviews outputs when the current model/runtime supports image reads.
@@ -152,9 +152,10 @@ localclawd does not require an account or managed login flow. Connect it directl
 
 The CLI is built for longer coding loops, not only one-shot prompts.
 
-- `/keepgoing` continues through pending work until the model emits a completion or input-needed signal.
-- `/thinkharder` increases verification rigor and model self-checking for complex changes.
-- `/buddy` creates a persistent ASCII companion persona for the current session.
+- `/keepgoing` continues through pending work until the model emits a completion or input-needed signal.
+- `/thinkharder` increases verification rigor and model self-checking for complex changes.
+- `/buddy` creates a persistent ASCII companion persona for the current session.
+- `/includememory` removes `CLAUDE.local.md` from gitignore so local memory can be committed when you want to share it.
 
 ### Project-local image workflow
 
@@ -216,16 +217,14 @@ localclawd
 
 localclawd does not require any account, login, or subscription. Connect it to a local model (vLLM, Ollama) or any OpenAI-compatible endpoint and start coding immediately. Use `/setup` at any time to configure or change your backend.
 
-If you want to use the Anthropic API directly, set `ANTHROPIC_API_KEY` in your environment — no login flow needed.
-
-## Release status
+## Release status
 
 `v1.7.2` is live on npm. Install globally with `npm install -g localclawd` or run without installing with `npx localclawd`.
 
 Current release highlights:
 
 - `1.7.2` adds `/images`, the bundled `/image-pipeline` workflow, project-local image pipeline scaffolding, ComfyUI helper/templates, and visual review for generated assets when the runtime supports image reads.
-- `1.7.1` introduced the autonomous `/keepgoing` loop, the `/buddy` companion flow, lattice-ranked local memory fallback, and the paired `/thinkharder` / `/thinknormal` operating modes.
+- `1.7.1` introduced the autonomous `/keepgoing` loop, the `/buddy` companion flow, and the paired `/thinkharder` / `/thinknormal` operating modes.
 - `1.0.0` was the initial public release.
 
 External native update metadata is now expected under `release-manifests/`, the main verification workflow lives in `.github/workflows/ci.yml`, and the native asset publication workflow lives in `.github/workflows/publish-release-assets.yml`. See `docs/release.md` for the expected asset set and publish sequence.
@@ -312,7 +311,7 @@ During first-run setup, localclawd asks for a compact context window cap. Use th
 
 ## Why use it instead of the upstream hosted CLI
 
-- You can point the CLI at your own inference stack instead of a Claude-hosted backend.
+- You can point the CLI at your own inference stack instead of a hosted backend.
 - Backend configuration lives inside the app, not only in shell variables.
 - `doctor` validates the configured local backend instead of only checking install state.
 - vLLM, Ollama, and generic OpenAI-compatible gateways are all supported under the same terminal UX.