← All reportsPublic report

ultraworkers/claw-code

An agent-managed museum exhibit, built in Rust with Gajae-Code / LazyCodex — developed and maintained with no human intervention.

195k stars Rust View on GitHubprofiled 11d ago
Try GitZoid
6PRs this week
2Contributors
0Deps scanned
0Issues found
01 · Repo overview

How claw-code is put together

Claw Code is a dual-workspace repository implementing `claw`, an agentic coding CLI harness. The canonical implementation is a Rust cargo workspace (`rust/`) whose crates provide multi-provider LLM API access (Anthropic native plus OpenAI-compatible providers: OpenAI, xAI/Grok, DashScope/Qwen, local Ollama), a tool-loop agent harness with explicit permissions and a workspace jail, and a workspace RAG service backed by SQLite. A companion Python package (`src/`) is a non-runtime 'porting/audit' workspace that mirrors command/tool inventories from an archived TypeScript snapshot and runs parity audits. Data flow: user prompt -> CLI -> ProviderClient (selected by model-name detection and env credentials) -> streaming chat completion API -> tool loop; sessions/transcripts are persisted locally as JSON files under `.port_sessions/` and `.claude/sessions/`. The repo self-describes as an agent-managed exhibit maintained by external harnesses (LazyCodex/Gajae-Code).

Languages

RustPython

Frameworks

tokioaxumclapreqwestrusqlitecriterionargparse (stdlib)

Datastores

SQLite (bundled, via rusqlite in claw-rag-service)Qdrant (optional feature flag 'qdrant-index')

Infrastructure

Docker (Containerfile)docker-compose.ymlrust/crates/claw-rag-service/Dockerfileinstall.shLocal OpenAI-compatible servers (Ollama via OLLAMA_HOST)

Major components

rust/crates/api

Provider abstraction layer: AnthropicClient and OpenAiCompatClient, model alias/provider detection, message translation between internal MessageRequest and OpenAI wire format, prompt cache, SSE streams, retries/backoff, and a rich ApiError taxonomy.

rust/crates/runtime

Shared runtime crate consumed by `api` and `claw-analog` (internal support library for the harnesses).

rust/crates/rusty-claude-cli

The `claw` CLI binary crate (includes build.rs), exposing subcommands such as `prompt`, `doctor`, and the `acp` status entrypoint.

rust/crates/claw-analog

Minimal agent harness: tool loop with explicit permissions (deny-list via globset) and a workspace jail, walking the filesystem with walkdir/ignore.

rust/crates/claw-rag-service

Standalone workspace RAG service: indexes the workspace into SQLite (blake3 hashing), queries OpenAI-compatible embeddings, and serves a query API over axum; optional Qdrant backend behind the `qdrant-index` feature.

rust/crates/mock-anthropic-service

Dev-only mock Anthropic API service used as a test double by claw-analog integration tests and the parity harness.

rust/crates/commands, plugins, compat-harness

Supporting workspace crates for command definitions, plugin handling, and compatibility-harness logic (manifests present, sources not shown in detail).

src/ (Python porting workspace)

Non-primary audit/porting toolkit: builds manifests and graphs of the archived TypeScript snapshot, mirrors command/tool inventories, simulates runtime branching modes, persists/loads sessions, and runs parity audits against the archive.

Over the past few weeks, work focused on making the sandbox environment run more reliably on systems where certain user-mapping tools are restricted or missing. The team added an automatic fallback mode when user mapping isn't allowed, documented the extra software it depends on, and wrapped up with a knowledge-base document for contributors.

Week by week

2026-08-10A documentation week: a structured knowledge base for agents was added to the repo.latest1 change

Docs

Hierarchical AGENTS.md knowledge base

Added a layered AGENTS.md file so automated agents and new contributors can quickly find project guidelines and context.

2026-08-03The sandbox fallback fix was finalized and merged after checking the full launcher configuration, not just its mapping settings.2 changes

Fix

Sandbox launcher probe now checks the full shape

The sandbox now verifies the entire launcher setup before deciding how to start, preventing incorrect assumptions that could break launches.

Chore

Merged sandbox fallback fix (PR #3280)

The pull request containing the automatic fallback behavior was reviewed and merged into the main codebase.

2026-07-27The groundwork was laid for running the sandbox on restricted systems, with an automatic fallback and supporting docs and cleanup.3 changes

Fix

Automatic --map-auto fallback in sandbox

When the system restricts root-user mapping, the sandbox now automatically switches to an alternate mode (--map-auto) instead of failing.

Refactor

Cached mapping probe helper

The check that detects whether user mapping is allowed was pulled into its own reusable function with caching, so repeated checks are faster and cleaner.

Docs

Documented newuidmap/newgidmap dependency

Added notes explaining that the fallback mode needs the newuidmap/newgidmap tools installed on the machine.

03 · Security check

Dependencies and code review

0 dependencies scanned

Dependency advisories

Security Watch

No known advisories across 0 scanned dependencies.

No known advisories in the scanned dependencies.

Code review

No concrete code-level issues confirmed.

Get this report every week for your repos.

GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.

$19 a month, flat · First 10 outputs free · No card required