Clawboo vs Mirror CLI

Clawboo 83

Harness reading of the framework line, as for Paperclip. The REST API's 199 routes are documented per resource with parameters, bodies and responses, and the MCP tools reference gives typed arguments for every tool on the four servers, but no OpenAPI file was found (17). llms.txt, llms-full.txt and a Markdown copy of every page at .md on docs.claw.boo (10). The docs say what each part does and does not cover, with known issues, production defaults and a limits page that names the code line behind each number (17). Tool arguments are typed with required and optional fields and documented caps, such as 24 live subtasks per parent and limits lowered to 1,000, with free-text task bodies (12). curl examples on each API page, an events and errors reference, and the exact error strings each MCP tool returns (14). A dated changelog for every release with an Unreleased section, though API paths carry no version and every MCP server reports 0.1.0 (13).

Mirror CLI 65

Framework reading. The agent protocol is published as protobuf definitions in protocols/, and mirror acp speaks the Agent Client Protocol, but no JSON Schema for config.json or the MCP definition files was found (15 of 25). llms.txt with a Markdown copy of every page (10). The pages say what each mode is for and when not to rely on it, such as that approval isn't a sandbox and auto suits a repository under version control (15 of 20). Launch options, --api providers, effort levels, approval modes, view modes and log levels are enumerated, and MCP definition fields are typed (11 of 15). Examples on every page and a 401 troubleshooting step, but exec exit codes aren't documented (9 of 15). A version number and an update check, but no changelog (5 of 15).

Clawboo 75

Harness reading, scored on what an agent or pipeline driving it has to supply. Four MCP servers keep tool lists small (12 Tasks tools, 4 Memory tools, 2 or 3 TeamChat tools, and Tools with four demonstration builtins plus granted connectors), Clawboo Native gets a read-only Tasks subset, and Tools results over 16 KiB are trimmed with a handle to read the rest (22). Circuit breakers stop task-runner runs after more than 30 tool calls or 200,000 tokens a minute, and team chat caps hand-offs at 8 a turn and three levels. Budgets warn only unless set to a hard cap, and a hard cap stops most runs only between runs (14). REST errors use one {error} shape, refused MCP calls return denied: with a reason prefix, and 429s carry Retry-After (16). claim_task and assign_task are atomic, illegal status changes are refused by name, the board survives restarts and routines resume after failure. No MCP tool annotations are documented (14). One npx command and a provider key or a local Ollama, but Claude Code teammates need a global install with the Agent SDK, and there is no client SDK (9).

Mirror CLI 58

Framework reading, adapted to a harness driven by a pipeline. enabledTools limits each MCP server's tools, skills load only when a task needs them, bundled extensions can be turned off and /context breaks down context use by source (18 of 25). --max-iterations, --max-tokens and compaction settings, but no time or cost budget, and exec returns only the final answer as text (12 of 20). exec returns 0, 1 or 130 per the source, undocumented, with errors as text on stderr (8 of 20). Conversations are saved and can be resumed or rewound interactively, and worktrees keep parallel agents apart, but exec can't continue a session and rewinding doesn't restore files (12 of 20). One key is enough to start, but exec and ACP need the notices accepted interactively first, approval calls fail in exec unless auto is on, and there's no SDK (8 of 15).

Clawboo 52

Read as local software. The official npm package clawboo states Node.js 22.12 or newer, and the README says macOS, Linux and Windows are supported (20). ci.yml runs lint, typecheck, Vitest and a real-concurrency suite on Linux and platform suites on Windows and macOS, and publish.yml reruns build, lint, typecheck, tests and a clean-install boot before it publishes. 0.5.0 carries provenance from that workflow on main, so its checks passed at that commit, but run results on main couldn't be read (15). github.com was unreachable, so open crash and regression issues are unchecked. A known-issues page lists 25 entries with a status each, three of them fixed on main (5). Changesets write a changelog entry per release, and removals and the 0.2.0 database reset are called out, though breaking changes arrive in minor versions before 1.0 (12). Version 0.5.0, and SECURITY.md says pre-1.0 (0).

Mirror CLI 37

Local-package reading. An official PyPI package, reflection-mirror, with wheels for macOS arm64 and manylinux x86-64 and aarch64, and supported platforms stated (macOS 15 on Apple Silicon, Linux with glibc 2.28, Python 3.12 or later) (20). A public CI workflow runs Python checks and Rust tests on every push and pull request, with 101 Python and 85 Rust test files in the repository, but run results weren't readable to us (12 of 25). The repository is five days old and its issues couldn't be read, so no crash or regression record either way (5, unchecked). No changelog, no release tags and one version (0). 0.1.0, labelled early beta in the docs (0).

Clawboo 70

Harness reading of the framework checklist, as for the other harnesses. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. The docs say Clawboo sends no telemetry and list the GitHub catalogue fetch, the npm version check, OpenRouter's public model list, Google Fonts and the GitHub star count, of which only the catalogue address can be changed (27). Clawboo Native shell commands are off by default and asked one at a time, and connector grants are per agent with approval on writes. Claude Code, Codex and Hermes run with their prompts and sandboxes off and no setting to change it, they can read Clawboo's database and key vault, and with no access token by default any local program can answer approval cards (8). Images in agent text don't load on their own, peer posts are marked as evidence, and a run that has read private data and untrusted content asks before any outbound connector call (11). An audit log of budget, breaker, approval, verification, tool-call and install events, with secrets masked (13). SECURITY.md with private reporting, CodeQL, Dependabot and npm provenance. No security.txt or bug bounty, OSV lists no advisory for clawboo, and the proxy advisory is on the docs with a workaround while its fix is unreleased (11).

Mirror CLI 58

Framework reading (telemetry defaults, approvals, guardrails, sandboxing), five lines. Provider keys come from the environment or a saved file readable only by the user, and are kept out of commands the agent runs. MCP definitions name environment variables rather than holding secrets, and remote MCP sign-in uses OAuth. Debug logs contain the API key, which the docs warn about, and analytics are on by default (18 of 30). Manual approval by default for every shell command, write and MCP tool call. Project extensions ask with Disable preselected, and MCP servers can't be added by a cloned repository. Auto mode, once chosen, applies in every directory (15 of 20). SECURITY.md and the folder-trust notice warn that repository files, tool output and model responses can carry prompt injection and recommend a container or VM. The microVM sandbox is opt-in and covers file and shell tools only (9 of 15). Conversations with their tool calls are kept in ~/.mirror/state.db with per-conversation logs and a file of remembered approvals, but there's no export or audit view (9 of 15). SECURITY.md gives a private email and a report scope, but there's no security.txt, bug bounty or published advisory (7 of 20). Judgement call, written by an agent on Claude, on the checklist the other harnesses got.