SwarmMojoBLOCK
Full-fledged local-first AI agent harness combining ROMS (RAG, OKF, MCP, Skills), Decision AI, correction memory, and reviewable workflow preparation.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[!NOTE] In Plain English: When running multiple autonomous AI agents locally, developers constantly face three painful bottlenecks: sluggish runtimes that burn memory, chaotic multi-agent loops that hallucinate and overwrite each other's files, and context bloat from massive logs and unindexed tools. SwarmMojo is a unified, local-first meta-harness engineered for maximum speed, strict determinism, and zero cloud dependency. Powered by 21 specialized engines accelerated by native Mojo SIMD and Python, SwarmMojo coordinates autonomous coding, enterprise UI design, multimodal studio production, human-grade prose authoring, DAG workflow automation, and 24/7 personal executive assistance. With mathematical drift guardrails (<65°), microsecond symbol indexing (<20 μs), content-addressed snapshot rollbacks, and offline workflow rehearsals, SwarmMojo delivers sovereign, enterprise-grade AI execution directly on your local machine.
A unified, local-first AI agent harness combining decision intelligence, specialist reviews, workflow rehearsal, and universal agent configuration.
SwarmMojo brings together three core pillars:
- ROMS Engine: Decision AI, RAG retrieval, Open Knowledge Files (OKF), SQLite-Vec, MCP tools, and skills to prepare bounded, reviewable actions.
- Swarm Mojo (Aeon): Source-grounded specialist reviews, TriggerTangle workflow rehearsal, bounded execution, and optional mSGL inference adapters.
- PolyHarness: Universal transpiler and package manager for AI agent configurations (compiling to
CLAUDE.md,AGENTS.md,dsh.config.json,.cursorrules, scoped.cursor/rules/*.mdc,.claude/rules/*.md, MCP registries, and deterministic pre-commit guardrails).
The portable Python path is used on Windows and Linux. Omarchy is an optional integration; it is not required to use SwarmMojo.
Core Pillars & Capabilities
0ae753cbd911OBSERVED · 2026-10-10Host compatibility
What the documentation claims. We have not run a compatibility test.
| Host | Status | Notes |
|---|---|---|
| claude-code | mentioned | |
| cursor | mentioned | |
| windsurf | mentioned |
Trust audit
BLOCKgrade F · trust 49/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | WARN |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (11 observation(s))
- Network
- declared (15 observation(s))
- Shell
- declared (7 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- found
Findings (25)
"eval": "Use of eval() is strictly forbidden in sandbox tasks.",
"exec": "Use of exec() is strictly forbidden in sandbox tasks.",
'id_rsa', 'id_ed25519', '.npmrc', '.pypirc', 'runtime.env'}
mod = importlib.import_module(module_name)
r"/etc/shadow", # System shadow
r"/etc/passwd", # System passwd
result = subprocess.run(['/usr/local/bin/goose', *args], capture_output=True,
subprocess.run(['/usr/local/bin/goose', '--remove-source', source_id], check=True, timeout=30)
subprocess.run(['/usr/local/bin/goose', {'/project': '--project-workshop', '/system': '--system', '/voice': '--voice'}[prompt]], check=False)"beacon of",
"tapestry", "delve", "beacon", "groundbreaking", "multifaceted",
endpoint = "http://127.0.0.1:30000"
endpoint = "http://127.0.0.1:30001"
endpoint = "http://127.0.0.1:30002"
endpoint = "http://127.0.0.1:30003"
endpoint = "http://127.0.0.1:30004"
'café e\u0301 — 日本語 中文 العربية', '👩🏽💻 🚀', '<s> <think> literal </think>', 'before\x00after']
secret = 'sensitiveuniquevalue12345'
secret = 'uniquecredential123456789'
(repo / 'private.txt').write_text('-----BEGIN OPENSSH PRIVATE KEY-----')implementations.append(getattr(importlib.import_module(module), function))
SOURCE = {'a.py': b'exec(code)\n', 'notes.md': b'Not a supported language\n'}" eval(user_input)\n"
h = int(hashlib.md5(word.encode("utf-8")).hexdigest(), 16)Gates applied: no_behavioural_pass.
0ae753cbd911full audit observations/trust-audit/skill/buzburg__swarmmojo.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-10 | 0ae753cbd911 | BLOCK | F | 49 | first audit |
Questions
What does the SwarmMojo skill do?
Full-fledged local-first AI agent harness combining ROMS (RAG, OKF, MCP, Skills), Decision AI, correction memory, and reviewable workflow preparation.
Is SwarmMojo safe to install?
No — not without reading the findings first. The audit graded it F (49/100) and found 3 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What can SwarmMojo access on my machine?
The audit observed that it reaches the network, runs shell commands and reads or writes files. Each of those is consistent with what it says it does. Secrets in the source: found — see the findings.
Which assistants does SwarmMojo work with?
Its documentation mentions claude-code, cursor and windsurf. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (0ae753cbd911), read on 2026-10-10. The repository is watched, and a new audit runs when it changes — this is the first audit.